Search

Large annotated datasets are required for training deep learning models, but in medical imaging data sharing is often complicated due to ethics, anonymization and data protection legislation. Generative AI models, such as generative adversarial networks (GANs) and diffusion models, can today produce very realistic synthetic images, and can potentially facilitate data sharing. However, in order to share synthetic medical images it must first be demonstrated that they can be used for training different networks with acceptable performance. Here, we therefore comprehensively evaluate four GANs (progressive GAN, StyleGAN 1–3) and a diffusion model for the task of brain tumor segmentation (using two segmentation networks, U-Net and a Swin transformer). Our results show that segmentation networks trained on synthetic images reach Dice scores that are 80%–90% of Dice scores when training with real images, but that memorization of the training images can be a problem for diffusion models if the original dataset is too small. Our conclusion is that sharing synthetic medical images is a viable option to sharing real images, but that further work is required. The trained generative models and the generated synthetic images are shared on AIDA data hub., Funding Agencies|ITEA/VINNOVA project ASSIST [2021-01420]; LiU Cancer; VINNOVA AIDA [M22-0088]; Ake Wiberg foundation; Wallenberg Center for Molecular Medicine as an associated clinical fellow; [2021-01954]

Atypical femur fractures (AFF) represent a very rare type of fracture that can be difficult to discriminate radiologically from normal femur fractures (NFF). AFFs are associated with drugs that are administered to prevent osteoporosis-related fragility fractures, which are highly prevalent in the elderly population. Given that these fractures are rare and the radiologic changes are subtle currently only 7% of AFFs are correctly identified, which hinders adequate treatment for most patients with AFF. Deep learning models could be trained to classify automatically a fracture as AFF or NFF, thereby assisting radiologists in detecting these rare fractures. Historically, for this classification task, only imaging data have been used, using convolutional neural networks (CNN) or vision transformers applied to radiographs. However, to mimic situations in which all available data are used to arrive at a diagnosis, we adopted an approach of deep learning that is based on the integration of image data and tabular data (from electronic health records) for 159 patients with AFF and 914 patients with NFF. We hypothesized that the combinatorial data, compiled from all the radiology departments of 72 hospitals in Sweden and the Swedish National Patient Register, would improve classification accuracy, as compared to using only one modality. At the patient level, the area under the ROC curve (AUC) increased from 0.966 to 0.987 when using the integrated set of imaging data and seven pre-selected variables, as compared to only using imaging data. More importantly, the sensitivity increased from 0.796 to 0.903. We found a greater impact of data fusion when only a randomly selected subset of available images was used to make the image and tabular data more balanced for each patient. The AUC then increased from 0.949 to 0.984, and the sensitivity increased from 0.727 to 0.849. These AUC improvements are not large, mainly because of the already excellent performance of the CNN (AUC of 0.9, Funding: ITEA/VINNOVA [2021-01954]; Region Ostergotland; Knut and Alice Wallenberg Foundation; Swedish research council [2023-01942]

Large annotated datasets are required for training deep learning models, but in medical imaging data sharing is often complicated due to ethics, anonymization and data protection legislation. Generative AI models, such as generative adversarial networks (GANs) and diffusion models, can today produce very realistic synthetic images, and can potentially facilitate data sharing. However, in order to share synthetic medical images it must first be demonstrated that they can be used for training different networks with acceptable performance. Here, we therefore comprehensively evaluate four GANs (progressive GAN, StyleGAN 1–3) and a diffusion model for the task of brain tumor segmentation (using two segmentation networks, U-Net and a Swin transformer). Our results show that segmentation networks trained on synthetic images reach Dice scores that are 80%–90% of Dice scores when training with real images, but that memorization of the training images can be a problem for diffusion models if the original dataset is too small. Our conclusion is that sharing synthetic medical images is a viable option to sharing real images, but that further work is required. The trained generative models and the generated synthetic images are shared on AIDA data hub., Funding Agencies|ITEA/VINNOVA project ASSIST [2021-01420]; LiU Cancer; VINNOVA AIDA [M22-0088]; Ake Wiberg foundation; Wallenberg Center for Molecular Medicine as an associated clinical fellow; [2021-01954]

Atypical femur fractures (AFF) represent a very rare type of fracture that can be difficult to discriminate radiologically from normal femur fractures (NFF). AFFs are associated with drugs that are administered to prevent osteoporosis-related fragility fractures, which are highly prevalent in the elderly population. Given that these fractures are rare and the radiologic changes are subtle currently only 7% of AFFs are correctly identified, which hinders adequate treatment for most patients with AFF. Deep learning models could be trained to classify automatically a fracture as AFF or NFF, thereby assisting radiologists in detecting these rare fractures. Historically, for this classification task, only imaging data have been used, using convolutional neural networks (CNN) or vision transformers applied to radiographs. However, to mimic situations in which all available data are used to arrive at a diagnosis, we adopted an approach of deep learning that is based on the integration of image data and tabular data (from electronic health records) for 159 patients with AFF and 914 patients with NFF. We hypothesized that the combinatorial data, compiled from all the radiology departments of 72 hospitals in Sweden and the Swedish National Patient Register, would improve classification accuracy, as compared to using only one modality. At the patient level, the area under the ROC curve (AUC) increased from 0.966 to 0.987 when using the integrated set of imaging data and seven pre-selected variables, as compared to only using imaging data. More importantly, the sensitivity increased from 0.796 to 0.903. We found a greater impact of data fusion when only a randomly selected subset of available images was used to make the image and tabular data more balanced for each patient. The AUC then increased from 0.949 to 0.984, and the sensitivity increased from 0.727 to 0.849. These AUC improvements are not large, mainly because of the already excellent performance of the CNN (AUC of 0.9, Funding: ITEA/VINNOVA [2021-01954]; Region Ostergotland; Knut and Alice Wallenberg Foundation; Swedish research council [2023-01942]

Large annotated datasets are required for training deep learning models, but in medical imaging data sharing is often complicated due to ethics, anonymization and data protection legislation. Generative AI models, such as generative adversarial networks (GANs) and diffusion models, can today produce very realistic synthetic images, and can potentially facilitate data sharing. However, in order to share synthetic medical images it must first be demonstrated that they can be used for training different networks with acceptable performance. Here, we therefore comprehensively evaluate four GANs (progressive GAN, StyleGAN 1–3) and a diffusion model for the task of brain tumor segmentation (using two segmentation networks, U-Net and a Swin transformer). Our results show that segmentation networks trained on synthetic images reach Dice scores that are 80%–90% of Dice scores when training with real images, but that memorization of the training images can be a problem for diffusion models if the original dataset is too small. Our conclusion is that sharing synthetic medical images is a viable option to sharing real images, but that further work is required. The trained generative models and the generated synthetic images are shared on AIDA data hub., Funding Agencies|ITEA/VINNOVA project ASSIST [2021-01420]; LiU Cancer; VINNOVA AIDA [M22-0088]; Ake Wiberg foundation; Wallenberg Center for Molecular Medicine as an associated clinical fellow; [2021-01954]

Atypical femur fractures (AFF) represent a very rare type of fracture that can be difficult to discriminate radiologically from normal femur fractures (NFF). AFFs are associated with drugs that are administered to prevent osteoporosis-related fragility fractures, which are highly prevalent in the elderly population. Given that these fractures are rare and the radiologic changes are subtle currently only 7% of AFFs are correctly identified, which hinders adequate treatment for most patients with AFF. Deep learning models could be trained to classify automatically a fracture as AFF or NFF, thereby assisting radiologists in detecting these rare fractures. Historically, for this classification task, only imaging data have been used, using convolutional neural networks (CNN) or vision transformers applied to radiographs. However, to mimic situations in which all available data are used to arrive at a diagnosis, we adopted an approach of deep learning that is based on the integration of image data and tabular data (from electronic health records) for 159 patients with AFF and 914 patients with NFF. We hypothesized that the combinatorial data, compiled from all the radiology departments of 72 hospitals in Sweden and the Swedish National Patient Register, would improve classification accuracy, as compared to using only one modality. At the patient level, the area under the ROC curve (AUC) increased from 0.966 to 0.987 when using the integrated set of imaging data and seven pre-selected variables, as compared to only using imaging data. More importantly, the sensitivity increased from 0.796 to 0.903. We found a greater impact of data fusion when only a randomly selected subset of available images was used to make the image and tabular data more balanced for each patient. The AUC then increased from 0.949 to 0.984, and the sensitivity increased from 0.727 to 0.849. These AUC improvements are not large, mainly because of the already excellent performance of the CNN (AUC of 0.9, Funding: ITEA/VINNOVA [2021-01954]; Region Ostergotland; Knut and Alice Wallenberg Foundation; Swedish research council [2023-01942]

Atypical femur fractures (AFF) represent a very rare type of fracture that can be difficult to discriminate radiologically from normal femur fractures (NFF). AFFs are associated with drugs that are administered to prevent osteoporosis-related fragility fractures, which are highly prevalent in the elderly population. Given that these fractures are rare and the radiologic changes are subtle currently only 7% of AFFs are correctly identified, which hinders adequate treatment for most patients with AFF. Deep learning models could be trained to classify automatically a fracture as AFF or NFF, thereby assisting radiologists in detecting these rare fractures. Historically, for this classification task, only imaging data have been used, using convolutional neural networks (CNN) or vision transformers applied to radiographs. However, to mimic situations in which all available data are used to arrive at a diagnosis, we adopted an approach of deep learning that is based on the integration of image data and tabular data (from electronic health records) for 159 patients with AFF and 914 patients with NFF. We hypothesized that the combinatorial data, compiled from all the radiology departments of 72 hospitals in Sweden and the Swedish National Patient Register, would improve classification accuracy, as compared to using only one modality. At the patient level, the area under the ROC curve (AUC) increased from 0.966 to 0.987 when using the integrated set of imaging data and seven pre-selected variables, as compared to only using imaging data. More importantly, the sensitivity increased from 0.796 to 0.903. We found a greater impact of data fusion when only a randomly selected subset of available images was used to make the image and tabular data more balanced for each patient. The AUC then increased from 0.949 to 0.984, and the sensitivity increased from 0.727 to 0.849. These AUC improvements are not large, mainly because of the already excellent performance of the CNN (AUC of 0.9, Funding: ITEA/VINNOVA [2021-01954]; Region Ostergotland; Knut and Alice Wallenberg Foundation; Swedish research council [2023-01942]

Large annotated datasets are required for training deep learning models, but in medical imaging data sharing is often complicated due to ethics, anonymization and data protection legislation. Generative AI models, such as generative adversarial networks (GANs) and diffusion models, can today produce very realistic synthetic images, and can potentially facilitate data sharing. However, in order to share synthetic medical images it must first be demonstrated that they can be used for training different networks with acceptable performance. Here, we therefore comprehensively evaluate four GANs (progressive GAN, StyleGAN 1–3) and a diffusion model for the task of brain tumor segmentation (using two segmentation networks, U-Net and a Swin transformer). Our results show that segmentation networks trained on synthetic images reach Dice scores that are 80%–90% of Dice scores when training with real images, but that memorization of the training images can be a problem for diffusion models if the original dataset is too small. Our conclusion is that sharing synthetic medical images is a viable option to sharing real images, but that further work is required. The trained generative models and the generated synthetic images are shared on AIDA data hub., Funding Agencies|ITEA/VINNOVA project ASSIST [2021-01420]; LiU Cancer; VINNOVA AIDA [M22-0088]; Ake Wiberg foundation; Wallenberg Center for Molecular Medicine as an associated clinical fellow; [2021-01954]

Atypical femur fractures (AFF) represent a very rare type of fracture that can be difficult to discriminate radiologically from normal femur fractures (NFF). AFFs are associated with drugs that are administered to prevent osteoporosis-related fragility fractures, which are highly prevalent in the elderly population. Given that these fractures are rare and the radiologic changes are subtle currently only 7% of AFFs are correctly identified, which hinders adequate treatment for most patients with AFF. Deep learning models could be trained to classify automatically a fracture as AFF or NFF, thereby assisting radiologists in detecting these rare fractures. Historically, for this classification task, only imaging data have been used, using convolutional neural networks (CNN) or vision transformers applied to radiographs. However, to mimic situations in which all available data are used to arrive at a diagnosis, we adopted an approach of deep learning that is based on the integration of image data and tabular data (from electronic health records) for 159 patients with AFF and 914 patients with NFF. We hypothesized that the combinatorial data, compiled from all the radiology departments of 72 hospitals in Sweden and the Swedish National Patient Register, would improve classification accuracy, as compared to using only one modality. At the patient level, the area under the ROC curve (AUC) increased from 0.966 to 0.987 when using the integrated set of imaging data and seven pre-selected variables, as compared to only using imaging data. More importantly, the sensitivity increased from 0.796 to 0.903. We found a greater impact of data fusion when only a randomly selected subset of available images was used to make the image and tabular data more balanced for each patient. The AUC then increased from 0.949 to 0.984, and the sensitivity increased from 0.727 to 0.849. These AUC improvements are not large, mainly because of the already excellent performance of the CNN (AUC of 0.9, Funding: ITEA/VINNOVA [2021-01954]; Region Ostergotland; Knut and Alice Wallenberg Foundation; Swedish research council [2023-01942]

Large annotated datasets are required for training deep learning models, but in medical imaging data sharing is often complicated due to ethics, anonymization and data protection legislation. Generative AI models, such as generative adversarial networks (GANs) and diffusion models, can today produce very realistic synthetic images, and can potentially facilitate data sharing. However, in order to share synthetic medical images it must first be demonstrated that they can be used for training different networks with acceptable performance. Here, we therefore comprehensively evaluate four GANs (progressive GAN, StyleGAN 1–3) and a diffusion model for the task of brain tumor segmentation (using two segmentation networks, U-Net and a Swin transformer). Our results show that segmentation networks trained on synthetic images reach Dice scores that are 80%–90% of Dice scores when training with real images, but that memorization of the training images can be a problem for diffusion models if the original dataset is too small. Our conclusion is that sharing synthetic medical images is a viable option to sharing real images, but that further work is required. The trained generative models and the generated synthetic images are shared on AIDA data hub., Funding Agencies|ITEA/VINNOVA project ASSIST [2021-01420]; LiU Cancer; VINNOVA AIDA [M22-0088]; Ake Wiberg foundation; Wallenberg Center for Molecular Medicine as an associated clinical fellow; [2021-01954]

Analysis of brain connectivity is important for understanding how information is processed by the brain. We propose a novel Bayesian vector autoregression hierarchical model for analysing brain connectivity within resting-state functional magnetic resonance imaging, and apply it to simulated data and a real data set with subjects in different groups. Our approach models functional and effective connectivity simultaneously and allows for both group- and single-subject inference. We combine analytical marginalization with Hamiltonian Monte Carlo to obtain highly efficient posterior sampling. We show that our model gives similar inference for effective connectivity compared to models with a common covariance matrix to all subjects, but more accurate inference for functional connectivity between regions compared to models with more restrictive covariance structures. A Stan implementation of our model is available on GitHub., Funding Agencies|Riksbankens Jubileumsfond [P16-028:1]; Center for Industrial Information Technology (CENIIT) at Linkoping University; Swedish Research Council [2020-02846]

There is mounting evidence of the long-term effects of COVID-19 on the central nervous system, with patients experiencing diverse symptoms, often suggesting brain involvement. Conventional brain MRI of these patients shows unspecific patterns, with no clear connection of the symptomatology to brain tissue abnormalities, whereas diffusion tensor studies and volumetric analyses detect measurable changes in the brain after COVID-19. Diffusion MRI exploits the random motion of water molecules to achieve unique sensitivity to structures at the microscopic level, and new sequences employing generalized diffusion encoding provide structural information which are sensitive to intravoxel features. In this observational study, a total of 32 persons were investigated: 16 patients previously hospitalized for COVID-19 with persisting symptoms of post-COVID condition (mean age 60 years: range 41–79, all male) at 7-month follow-up and 16 matched controls, not previously hospitalized for COVID-19, with no post-COVID symptoms (mean age 58 years, range 46–69, 11 males). Standard MRI and generalized diffusion encoding MRI were employed to examine the brain white matter of the subjects. To detect possible group differences, several tissue microstructure descriptors obtainable with the employed diffusion sequence, the fractional anisotropy, mean diffusivity, axial diffusivity, radial diffusivity, microscopic anisotropy, orientational coherence (Cc) and variance in compartment’s size (CMD) were analysed using the tract-based spatial statistics framework. The tract-based spatial statistics analysis showed widespread statistically significant differences (P < 0.05, corrected for multiple comparisons using the familywise error rate) in all the considered metrics in the white matter of the patients compared to the controls. Fractional anisotropy, microscopic anisotropy and Cc were lower in the patient group, while axial diffusivity, radial diffusivity, mean diffusivity and CMD were higher, Funding: Analytic Imaging Diagnostic Arena (AIDA), a Medtech4Health initiative; ITEA/ VINNOVA (The Swedish Innovation Agency) project ASSIST (Automation, Surgery Support and Intuitive 3D visualization to optimize workflow in IGT SysTems) [2021-01954]; Wallenberg Center for Molecular Medicine

Classifying subjects as healthy or diseased using neuroimaging data has gained a lot of attention during the last 10 years, and recently, different deep learning approaches have been used. Despite this fact, there has not been any investigation regarding how 3D augmentation can help to create larger datasets, required to train deep networks with millions of parameters. In this study, deep learning was applied to derivatives from resting state functional MRI data, to investigate how different 3D augmentation techniques affect the test accuracy. Specifically, resting state derivatives from 1112 subjects in ABIDE (Autism Brain Imaging Data Exchange) preprocessed were used to train a 3D convolutional neural network (CNN) to classify each subject according to presence or absence of autism spectrum disorder. The results show that augmentation only provide minor improvements to the test accuracy., Funding: Swedish research council [2017-04889]; ITEA/VINNOVA [2021-01954]; Ake Wiberg foundation [M22-0088]

Using 3D CNNs on high-resolution medical volumes is very computationally demanding, especially for large datasets like UK Biobank, which aims to scan 100,000 subjects. Here, we demonstrate that using 2D CNNs on a few 2D projections (representing mean and standard deviation across axial, sagittal and coronal slices) of 3D volumes leads to reasonable test accuracy (mean absolute error of about 3.5 years) when predicting age from brain volumes. Using our approach, one training epoch with 20,324 subjects takes 20–50 s using a single GPU, which is two orders of magnitude faster than a small 3D CNN. This speedup is explained by the fact that 3D brain volumes contain a lot of redundant information, which can be efficiently compressed using 2D projections. These results are important for researchers who do not have access to expensive GPU hardware for 3D CNNs., Funding: ITEA/VINNOVA [2021-01954]; Ake Wiberg foundation

Background: Stereotactic radiosurgery (SRS) can be an effective primary or adjuvant treatment option for intracranial tumors. However, it carries risks of various radiation toxicities, which can lead to functional deficits for the patients. Current inverse planning algorithms for SRS provide an efficient way for sparing organs at risk (OARs) by setting maximum radiation dose constraints in the treatment planning process.Purpose: We propose using activation maps from functional MRI (fMRI) to map the eloquent regions of the brain and define functional OARs (fOARs) for Gamma Knife SRS treatment planning.Methods: We implemented a pipeline for analyzing patient fMRI data, generating fOARs from the resulting activation maps, and loading them onto the GammaPlan treatment planning software. We used the Lightning inverse planner to generate multiple treatment plans from open MRI data of five subjects, and evaluated the effects of incorporating the proposed fOARs.Results: The Lightning optimizer designs treatment plans with high conformity to the specified parameters. Setting maximum dose constraints on fOARs successfully limits the radiation dose incident on them, but can have a negative impact on treatment plan quality metrics. By masking out fOAR voxels surrounding the tumor target it is possible to achieve high quality treatment plans while controlling the radiation dose on fOARs.Conclusions: The proposed method can effectively reduce the radiation dose incident on the eloquent brain areas during Gamma Knife SRS of brain tumors., Funding: Centrum foer Industriell Informationsteknologi, Linkoepings Universitet; Vinnova [2018-02230, 2021-01954]

The infiltrative nature of malignant gliomas results in active tumor spreading into the peritumoral edema, which is not visible in conventional magnetic resonance imaging (cMRI) even after contrast injection. MR relaxometry (qMRI) measures relaxation rates dependent on tissue properties, and can offer additional contrast mechanisms to highlight the non-enhancing infiltrative tumor. To investigate if qMRI data provides additional information compared to cMRI sequences when considering deep learning-based brain tumor detection and segmentation, preoperative conventional (T1-w per- and post-contrast, T2-w and FLAIR) and quantitative (pre- and post-contrast R1, R2 and proton density) MR data was obtained from 23 patients with typical radiological findings suggestive of a high-grade malignant glioma. 2D deep learning models were trained on transversal slices (n=528) for tumor detection and segmentation using either conventional or quantitative data. Moreover, trends in quantitative R1 and R2 rates of regions identified as relevant for tumor detection by model explainability methods were qualitatively analyzed. Tumor detection and segmentation performance for models trained with a combination of qMRI pre- and post-contrast was the highest (detection MCC=0.72, segmentation DSC=0.90), however, the difference compared to cMRI was not statistically significant. Overall analysis of the relevant regions identified using model explainability showed no differences between models trained on cMRI or qMRI. When looking at the individual cases, relaxation rates of brain regions outside the annotation and identified as relevant for tumor detection exhibited changes after contrast injection similar to region inside the annotation in the majority of cases. In conclusion, models trained on qMRI data obtained similar detection and segmentation performance to those trained on cMRI data, with the advantage of quantitatively measuring brain tissue properties within similar scan time. When co, Funding: CENIIT at Linkoeping University, ITEA3 / VINNOVA funded project Intelligence based iMprovement of Personalized treatment And Clinical workflow supporT (IMPACT); ITEA4 / VINNOVA funded project Automation, Surgery Support and Intuitive 3D visualization to optimize workflow in IGT SysTems (ASSIST) [2021-01954]; Cancer Strength Area at Linkoeping University, VINOVA project via the Analytic Imaging Diagnostics Arena (AIDA) [2017-02447]; Medical Research Council of Southeast Sweden [FORSS-234551]; Swedish Research Council [2018-05250]

Using 3D CNNs on high-resolution medical volumes is very computationally demanding, especially for large datasets like UK Biobank, which aims to scan 100,000 subjects. Here, we demonstrate that using 2D CNNs on a few 2D projections (representing mean and standard deviation across axial, sagittal and coronal slices) of 3D volumes leads to reasonable test accuracy (mean absolute error of about 3.5 years) when predicting age from brain volumes. Using our approach, one training epoch with 20,324 subjects takes 20–50 s using a single GPU, which is two orders of magnitude faster than a small 3D CNN. This speedup is explained by the fact that 3D brain volumes contain a lot of redundant information, which can be efficiently compressed using 2D projections. These results are important for researchers who do not have access to expensive GPU hardware for 3D CNNs., Funding: ITEA/VINNOVA [2021-01954]; Ake Wiberg foundation

There is mounting evidence of the long-term effects of COVID-19 on the central nervous system, with patients experiencing diverse symptoms, often suggesting brain involvement. Conventional brain MRI of these patients shows unspecific patterns, with no clear connection of the symptomatology to brain tissue abnormalities, whereas diffusion tensor studies and volumetric analyses detect measurable changes in the brain after COVID-19. Diffusion MRI exploits the random motion of water molecules to achieve unique sensitivity to structures at the microscopic level, and new sequences employing generalized diffusion encoding provide structural information which are sensitive to intravoxel features. In this observational study, a total of 32 persons were investigated: 16 patients previously hospitalized for COVID-19 with persisting symptoms of post-COVID condition (mean age 60 years: range 41–79, all male) at 7-month follow-up and 16 matched controls, not previously hospitalized for COVID-19, with no post-COVID symptoms (mean age 58 years, range 46–69, 11 males). Standard MRI and generalized diffusion encoding MRI were employed to examine the brain white matter of the subjects. To detect possible group differences, several tissue microstructure descriptors obtainable with the employed diffusion sequence, the fractional anisotropy, mean diffusivity, axial diffusivity, radial diffusivity, microscopic anisotropy, orientational coherence (Cc) and variance in compartment’s size (CMD) were analysed using the tract-based spatial statistics framework. The tract-based spatial statistics analysis showed widespread statistically significant differences (P < 0.05, corrected for multiple comparisons using the familywise error rate) in all the considered metrics in the white matter of the patients compared to the controls. Fractional anisotropy, microscopic anisotropy and Cc were lower in the patient group, while axial diffusivity, radial diffusivity, mean diffusivity and CMD were higher, Funding: Analytic Imaging Diagnostic Arena (AIDA), a Medtech4Health initiative; ITEA/ VINNOVA (The Swedish Innovation Agency) project ASSIST (Automation, Surgery Support and Intuitive 3D visualization to optimize workflow in IGT SysTems) [2021-01954]; Wallenberg Center for Molecular Medicine

Classifying subjects as healthy or diseased using neuroimaging data has gained a lot of attention during the last 10 years, and recently, different deep learning approaches have been used. Despite this fact, there has not been any investigation regarding how 3D augmentation can help to create larger datasets, required to train deep networks with millions of parameters. In this study, deep learning was applied to derivatives from resting state functional MRI data, to investigate how different 3D augmentation techniques affect the test accuracy. Specifically, resting state derivatives from 1112 subjects in ABIDE (Autism Brain Imaging Data Exchange) preprocessed were used to train a 3D convolutional neural network (CNN) to classify each subject according to presence or absence of autism spectrum disorder. The results show that augmentation only provide minor improvements to the test accuracy., Funding: Swedish research council [2017-04889]; ITEA/VINNOVA [2021-01954]; Ake Wiberg foundation [M22-0088]

Background: Stereotactic radiosurgery (SRS) can be an effective primary or adjuvant treatment option for intracranial tumors. However, it carries risks of various radiation toxicities, which can lead to functional deficits for the patients. Current inverse planning algorithms for SRS provide an efficient way for sparing organs at risk (OARs) by setting maximum radiation dose constraints in the treatment planning process.Purpose: We propose using activation maps from functional MRI (fMRI) to map the eloquent regions of the brain and define functional OARs (fOARs) for Gamma Knife SRS treatment planning.Methods: We implemented a pipeline for analyzing patient fMRI data, generating fOARs from the resulting activation maps, and loading them onto the GammaPlan treatment planning software. We used the Lightning inverse planner to generate multiple treatment plans from open MRI data of five subjects, and evaluated the effects of incorporating the proposed fOARs.Results: The Lightning optimizer designs treatment plans with high conformity to the specified parameters. Setting maximum dose constraints on fOARs successfully limits the radiation dose incident on them, but can have a negative impact on treatment plan quality metrics. By masking out fOAR voxels surrounding the tumor target it is possible to achieve high quality treatment plans while controlling the radiation dose on fOARs.Conclusions: The proposed method can effectively reduce the radiation dose incident on the eloquent brain areas during Gamma Knife SRS of brain tumors., Funding: Centrum foer Industriell Informationsteknologi, Linkoepings Universitet; Vinnova [2018-02230, 2021-01954]

The infiltrative nature of malignant gliomas results in active tumor spreading into the peritumoral edema, which is not visible in conventional magnetic resonance imaging (cMRI) even after contrast injection. MR relaxometry (qMRI) measures relaxation rates dependent on tissue properties, and can offer additional contrast mechanisms to highlight the non-enhancing infiltrative tumor. To investigate if qMRI data provides additional information compared to cMRI sequences when considering deep learning-based brain tumor detection and segmentation, preoperative conventional (T1-w per- and post-contrast, T2-w and FLAIR) and quantitative (pre- and post-contrast R1, R2 and proton density) MR data was obtained from 23 patients with typical radiological findings suggestive of a high-grade malignant glioma. 2D deep learning models were trained on transversal slices (n=528) for tumor detection and segmentation using either conventional or quantitative data. Moreover, trends in quantitative R1 and R2 rates of regions identified as relevant for tumor detection by model explainability methods were qualitatively analyzed. Tumor detection and segmentation performance for models trained with a combination of qMRI pre- and post-contrast was the highest (detection MCC=0.72, segmentation DSC=0.90), however, the difference compared to cMRI was not statistically significant. Overall analysis of the relevant regions identified using model explainability showed no differences between models trained on cMRI or qMRI. When looking at the individual cases, relaxation rates of brain regions outside the annotation and identified as relevant for tumor detection exhibited changes after contrast injection similar to region inside the annotation in the majority of cases. In conclusion, models trained on qMRI data obtained similar detection and segmentation performance to those trained on cMRI data, with the advantage of quantitatively measuring brain tissue properties within similar scan time. When co, Funding: CENIIT at Linkoeping University, ITEA3 / VINNOVA funded project Intelligence based iMprovement of Personalized treatment And Clinical workflow supporT (IMPACT); ITEA4 / VINNOVA funded project Automation, Surgery Support and Intuitive 3D visualization to optimize workflow in IGT SysTems (ASSIST) [2021-01954]; Cancer Strength Area at Linkoeping University, VINOVA project via the Analytic Imaging Diagnostics Arena (AIDA) [2017-02447]; Medical Research Council of Southeast Sweden [FORSS-234551]; Swedish Research Council [2018-05250]

There is mounting evidence of the long-term effects of COVID-19 on the central nervous system, with patients experiencing diverse symptoms, often suggesting brain involvement. Conventional brain MRI of these patients shows unspecific patterns, with no clear connection of the symptomatology to brain tissue abnormalities, whereas diffusion tensor studies and volumetric analyses detect measurable changes in the brain after COVID-19. Diffusion MRI exploits the random motion of water molecules to achieve unique sensitivity to structures at the microscopic level, and new sequences employing generalized diffusion encoding provide structural information which are sensitive to intravoxel features. In this observational study, a total of 32 persons were investigated: 16 patients previously hospitalized for COVID-19 with persisting symptoms of post-COVID condition (mean age 60 years: range 41–79, all male) at 7-month follow-up and 16 matched controls, not previously hospitalized for COVID-19, with no post-COVID symptoms (mean age 58 years, range 46–69, 11 males). Standard MRI and generalized diffusion encoding MRI were employed to examine the brain white matter of the subjects. To detect possible group differences, several tissue microstructure descriptors obtainable with the employed diffusion sequence, the fractional anisotropy, mean diffusivity, axial diffusivity, radial diffusivity, microscopic anisotropy, orientational coherence (Cc) and variance in compartment’s size (CMD) were analysed using the tract-based spatial statistics framework. The tract-based spatial statistics analysis showed widespread statistically significant differences (P < 0.05, corrected for multiple comparisons using the familywise error rate) in all the considered metrics in the white matter of the patients compared to the controls. Fractional anisotropy, microscopic anisotropy and Cc were lower in the patient group, while axial diffusivity, radial diffusivity, mean diffusivity and CMD were higher, Funding: Analytic Imaging Diagnostic Arena (AIDA), a Medtech4Health initiative; ITEA/ VINNOVA (The Swedish Innovation Agency) project ASSIST (Automation, Surgery Support and Intuitive 3D visualization to optimize workflow in IGT SysTems) [2021-01954]; Wallenberg Center for Molecular Medicine

Using 3D CNNs on high-resolution medical volumes is very computationally demanding, especially for large datasets like UK Biobank, which aims to scan 100,000 subjects. Here, we demonstrate that using 2D CNNs on a few 2D projections (representing mean and standard deviation across axial, sagittal and coronal slices) of 3D volumes leads to reasonable test accuracy (mean absolute error of about 3.5 years) when predicting age from brain volumes. Using our approach, one training epoch with 20,324 subjects takes 20–50 s using a single GPU, which is two orders of magnitude faster than a small 3D CNN. This speedup is explained by the fact that 3D brain volumes contain a lot of redundant information, which can be efficiently compressed using 2D projections. These results are important for researchers who do not have access to expensive GPU hardware for 3D CNNs., Funding: ITEA/VINNOVA [2021-01954]; Ake Wiberg foundation

Classifying subjects as healthy or diseased using neuroimaging data has gained a lot of attention during the last 10 years, and recently, different deep learning approaches have been used. Despite this fact, there has not been any investigation regarding how 3D augmentation can help to create larger datasets, required to train deep networks with millions of parameters. In this study, deep learning was applied to derivatives from resting state functional MRI data, to investigate how different 3D augmentation techniques affect the test accuracy. Specifically, resting state derivatives from 1112 subjects in ABIDE (Autism Brain Imaging Data Exchange) preprocessed were used to train a 3D convolutional neural network (CNN) to classify each subject according to presence or absence of autism spectrum disorder. The results show that augmentation only provide minor improvements to the test accuracy., Funding: Swedish research council [2017-04889]; ITEA/VINNOVA [2021-01954]; Ake Wiberg foundation [M22-0088]

Brain age prediction from 3D MRI volumes using deep learning has recently become a popular research topic, as brain age has been shown to be an important biomarker. Training deep networks can be very computationally demanding for large datasets like the U.K. Biobank (currently 29,035 subjects). In our previous work, it was demonstrated that using a few 2D projections (mean and standard deviation along three axes) instead of each full 3D volume leads to much faster training at the cost of a reduction in prediction accuracy. Here, we investigated if another set of 2D projections, based on higher-order statistical central moments and eigenslices, leads to a higher accuracy. Our results show that higher-order moments do not lead to a higher accuracy, but that eigenslices provide a small improvement. We also show that an ensemble of such models provides further improvement., Funding: ITEA/VINNOVA

Brain age prediction from 3D MRI volumes using deep learning has recently become a popular research topic, as brain age has been shown to be an important biomarker. Training deep networks can be very computationally demanding for large datasets like the U.K. Biobank (currently 29,035 subjects). In our previous work, it was demonstrated that using a few 2D projections (mean and standard deviation along three axes) instead of each full 3D volume leads to much faster training at the cost of a reduction in prediction accuracy. Here, we investigated if another set of 2D projections, based on higher-order statistical central moments and eigenslices, leads to a higher accuracy. Our results show that higher-order moments do not lead to a higher accuracy, but that eigenslices provide a small improvement. We also show that an ensemble of such models provides further improvement., Funding: ITEA/VINNOVA

Brain age prediction from 3D MRI volumes using deep learning has recently become a popular research topic, as brain age has been shown to be an important biomarker. Training deep networks can be very computationally demanding for large datasets like the U.K. Biobank (currently 29,035 subjects). In our previous work, it was demonstrated that using a few 2D projections (mean and standard deviation along three axes) instead of each full 3D volume leads to much faster training at the cost of a reduction in prediction accuracy. Here, we investigated if another set of 2D projections, based on higher-order statistical central moments and eigenslices, leads to a higher accuracy. Our results show that higher-order moments do not lead to a higher accuracy, but that eigenslices provide a small improvement. We also show that an ensemble of such models provides further improvement., Funding: ITEA/VINNOVA

Brain age prediction from 3D MRI volumes using deep learning has recently become a popular research topic, as brain age has been shown to be an important biomarker. Training deep networks can be very computationally demanding for large datasets like the U.K. Biobank (currently 29,035 subjects). In our previous work, it was demonstrated that using a few 2D projections (mean and standard deviation along three axes) instead of each full 3D volume leads to much faster training at the cost of a reduction in prediction accuracy. Here, we investigated if another set of 2D projections, based on higher-order statistical central moments and eigenslices, leads to a higher accuracy. Our results show that higher-order moments do not lead to a higher accuracy, but that eigenslices provide a small improvement. We also show that an ensemble of such models provides further improvement., Funding: ITEA/VINNOVA

t-distributed stochastic neighbor embedding (t-SNE) is a method for reducing high-dimensional data to a low-dimensional representation, and is mostly used for visualizing data. In parametric t-SNE, a neural network learns to reproduce this mapping. When used for EEG analysis, the data are usually first transformed into a set of features, but it is not known which features are optimal. The principle of t-SNE was used to train convolutional neural network (CNN) encoders to learn to produce both a high- and a low-dimensional representation, eliminating the need for feature engineering. To evaluate the method, the Temple University EEG Corpus was used to create three datasets with distinct EEG characters: (1) wakefulness and sleep; (2) interictal epileptiform discharges; and (3) seizure activity. The CNN encoders produced low-dimensional representations of the datasets with a structure that conformed well to the EEG characters and generalized to new data. Compared to parametric t-SNE for either a short-time Fourier transform or wavelet representation of the datasets, the developed CNN encoders performed equally well in separating categories, as assessed by support vector machines. The CNN encoders generally produced a higher degree of clustering, both visually and in the number of clusters detected by k-means clustering. The developed principle is promising and could be further developed to create general tools for exploring relations in EEG data., Funding: Linkoping University; University Hospital of Linkoeping; ALF of Region OEstergoetland [LIO-936176, ROE-941359]; ITEA3/VINNOVA

Intraoperative guidance tools for thyroid surgery based on optical coherence tomography (OCT) could aid distinguish between normal and diseased tissue. However, OCT images are difficult to interpret, thus, real-time automatic analysis could support the clinical decision-making. In this study, several deep learning models were investigated for thyroid disease classification on 2D and 3D OCT data obtained from ex vivo specimens of 22 patients undergoing surgery and diagnosed with several thyroid pathologies. Additionally, two open-access datasets were used to evaluate the custom models. On the thyroid dataset, the best performance was achieved by the 3D vision transformer model with a Matthews correlation coefficient (MCC) of 0.79 (accuracy = 0.90) for the normal-versus-abnormal classification. On the open-access datasets, the custom models achieved the best performance (MCC > 0.88, accuracy > 0.96). Results obtained for the normal-versus-abnormal classification suggest OCT, complemented with deep learning-based analysis, as a tool for real-time automatic diseased tissue identification in thyroid surgery., Funding: Ake Wiberg Stiftelse [M19-0455 M20-0034 M21-0083]; Forskningsradet i Sydostra Sverige [931466]; VINNOVA via Medtech4Health; AIDA(1908) [2017-02447]; Vetenskapsradet-Swedish Research Council [2018-05250]

Intraoperative guidance tools for thyroid surgery based on optical coherence tomography (OCT) could aid distinguish between normal and diseased tissue. However, OCT images are difficult to interpret, thus, real-time automatic analysis could support the clinical decision-making. In this study, several deep learning models were investigated for thyroid disease classification on 2D and 3D OCT data obtained from ex vivo specimens of 22 patients undergoing surgery and diagnosed with several thyroid pathologies. Additionally, two open-access datasets were used to evaluate the custom models. On the thyroid dataset, the best performance was achieved by the 3D vision transformer model with a Matthews correlation coefficient (MCC) of 0.79 (accuracy = 0.90) for the normal-versus-abnormal classification. On the open-access datasets, the custom models achieved the best performance (MCC > 0.88, accuracy > 0.96). Results obtained for the normal-versus-abnormal classification suggest OCT, complemented with deep learning-based analysis, as a tool for real-time automatic diseased tissue identification in thyroid surgery., Funding: Ake Wiberg Stiftelse [M19-0455 M20-0034 M21-0083]; Forskningsradet i Sydostra Sverige [931466]; VINNOVA via Medtech4Health; AIDA(1908) [2017-02447]; Vetenskapsradet-Swedish Research Council [2018-05250]

Intraoperative guidance tools for thyroid surgery based on optical coherence tomography (OCT) could aid distinguish between normal and diseased tissue. However, OCT images are difficult to interpret, thus, real-time automatic analysis could support the clinical decision-making. In this study, several deep learning models were investigated for thyroid disease classification on 2D and 3D OCT data obtained from ex vivo specimens of 22 patients undergoing surgery and diagnosed with several thyroid pathologies. Additionally, two open-access datasets were used to evaluate the custom models. On the thyroid dataset, the best performance was achieved by the 3D vision transformer model with a Matthews correlation coefficient (MCC) of 0.79 (accuracy = 0.90) for the normal-versus-abnormal classification. On the open-access datasets, the custom models achieved the best performance (MCC > 0.88, accuracy > 0.96). Results obtained for the normal-versus-abnormal classification suggest OCT, complemented with deep learning-based analysis, as a tool for real-time automatic diseased tissue identification in thyroid surgery., Funding: Ake Wiberg Stiftelse [M19-0455 M20-0034 M21-0083]; Forskningsradet i Sydostra Sverige [931466]; VINNOVA via Medtech4Health; AIDA(1908) [2017-02447]; Vetenskapsradet-Swedish Research Council [2018-05250]

t-distributed stochastic neighbor embedding (t-SNE) is a method for reducing high-dimensional data to a low-dimensional representation, and is mostly used for visualizing data. In parametric t-SNE, a neural network learns to reproduce this mapping. When used for EEG analysis, the data are usually first transformed into a set of features, but it is not known which features are optimal. The principle of t-SNE was used to train convolutional neural network (CNN) encoders to learn to produce both a high- and a low-dimensional representation, eliminating the need for feature engineering. To evaluate the method, the Temple University EEG Corpus was used to create three datasets with distinct EEG characters: (1) wakefulness and sleep; (2) interictal epileptiform discharges; and (3) seizure activity. The CNN encoders produced low-dimensional representations of the datasets with a structure that conformed well to the EEG characters and generalized to new data. Compared to parametric t-SNE for either a short-time Fourier transform or wavelet representation of the datasets, the developed CNN encoders performed equally well in separating categories, as assessed by support vector machines. The CNN encoders generally produced a higher degree of clustering, both visually and in the number of clusters detected by k-means clustering. The developed principle is promising and could be further developed to create general tools for exploring relations in EEG data., Funding: Linkoping University; University Hospital of Linkoeping; ALF of Region OEstergoetland [LIO-936176, ROE-941359]; ITEA3/VINNOVA

t-distributed stochastic neighbor embedding (t-SNE) is a method for reducing high-dimensional data to a low-dimensional representation, and is mostly used for visualizing data. In parametric t-SNE, a neural network learns to reproduce this mapping. When used for EEG analysis, the data are usually first transformed into a set of features, but it is not known which features are optimal. The principle of t-SNE was used to train convolutional neural network (CNN) encoders to learn to produce both a high- and a low-dimensional representation, eliminating the need for feature engineering. To evaluate the method, the Temple University EEG Corpus was used to create three datasets with distinct EEG characters: (1) wakefulness and sleep; (2) interictal epileptiform discharges; and (3) seizure activity. The CNN encoders produced low-dimensional representations of the datasets with a structure that conformed well to the EEG characters and generalized to new data. Compared to parametric t-SNE for either a short-time Fourier transform or wavelet representation of the datasets, the developed CNN encoders performed equally well in separating categories, as assessed by support vector machines. The CNN encoders generally produced a higher degree of clustering, both visually and in the number of clusters detected by k-means clustering. The developed principle is promising and could be further developed to create general tools for exploring relations in EEG data., Funding: Linkoping University; University Hospital of Linkoeping; ALF of Region OEstergoetland [LIO-936176, ROE-941359]; ITEA3/VINNOVA

Background: Stereotactic radiosurgery (SRS) can be an effective primary or adjuvant treatment option for intracranial tumors. However, it carries risks of various radiation toxicities, which can lead to functional deficits for the patients. Current inverse planning algorithms for SRS provide an efficient way for sparing organs at risk (OARs) by setting maximum radiation dose constraints in the treatment planning process.Purpose: We propose using activation maps from functional MRI (fMRI) to map the eloquent regions of the brain and define functional OARs (fOARs) for Gamma Knife SRS treatment planning.Methods: We implemented a pipeline for analyzing patient fMRI data, generating fOARs from the resulting activation maps, and loading them onto the GammaPlan treatment planning software. We used the Lightning inverse planner to generate multiple treatment plans from open MRI data of five subjects, and evaluated the effects of incorporating the proposed fOARs.Results: The Lightning optimizer designs treatment plans with high conformity to the specified parameters. Setting maximum dose constraints on fOARs successfully limits the radiation dose incident on them, but can have a negative impact on treatment plan quality metrics. By masking out fOAR voxels surrounding the tumor target it is possible to achieve high quality treatment plans while controlling the radiation dose on fOARs.Conclusions: The proposed method can effectively reduce the radiation dose incident on the eloquent brain areas during Gamma Knife SRS of brain tumors., Funding: Centrum foer Industriell Informationsteknologi, Linkoepings Universitet; Vinnova [2018-02230, 2021-01954]

The infiltrative nature of malignant gliomas results in active tumor spreading into the peritumoral edema, which is not visible in conventional magnetic resonance imaging (cMRI) even after contrast injection. MR relaxometry (qMRI) measures relaxation rates dependent on tissue properties, and can offer additional contrast mechanisms to highlight the non-enhancing infiltrative tumor. To investigate if qMRI data provides additional information compared to cMRI sequences when considering deep learning-based brain tumor detection and segmentation, preoperative conventional (T1-w per- and post-contrast, T2-w and FLAIR) and quantitative (pre- and post-contrast R1, R2 and proton density) MR data was obtained from 23 patients with typical radiological findings suggestive of a high-grade malignant glioma. 2D deep learning models were trained on transversal slices (n=528) for tumor detection and segmentation using either conventional or quantitative data. Moreover, trends in quantitative R1 and R2 rates of regions identified as relevant for tumor detection by model explainability methods were qualitatively analyzed. Tumor detection and segmentation performance for models trained with a combination of qMRI pre- and post-contrast was the highest (detection MCC=0.72, segmentation DSC=0.90), however, the difference compared to cMRI was not statistically significant. Overall analysis of the relevant regions identified using model explainability showed no differences between models trained on cMRI or qMRI. When looking at the individual cases, relaxation rates of brain regions outside the annotation and identified as relevant for tumor detection exhibited changes after contrast injection similar to region inside the annotation in the majority of cases. In conclusion, models trained on qMRI data obtained similar detection and segmentation performance to those trained on cMRI data, with the advantage of quantitatively measuring brain tissue properties within similar scan time. When co, Funding: CENIIT at Linkoeping University, ITEA3 / VINNOVA funded project Intelligence based iMprovement of Personalized treatment And Clinical workflow supporT (IMPACT); ITEA4 / VINNOVA funded project Automation, Surgery Support and Intuitive 3D visualization to optimize workflow in IGT SysTems (ASSIST) [2021-01954]; Cancer Strength Area at Linkoeping University, VINOVA project via the Analytic Imaging Diagnostics Arena (AIDA) [2017-02447]; Medical Research Council of Southeast Sweden [FORSS-234551]; Swedish Research Council [2018-05250]

The infiltrative nature of malignant gliomas results in active tumor spreading into the peritumoral edema, which is not visible in conventional magnetic resonance imaging (cMRI) even after contrast injection. MR relaxometry (qMRI) measures relaxation rates dependent on tissue properties, and can offer additional contrast mechanisms to highlight the non-enhancing infiltrative tumor. To investigate if qMRI data provides additional information compared to cMRI sequences when considering deep learning-based brain tumor detection and segmentation, preoperative conventional (T1-w per- and post-contrast, T2-w and FLAIR) and quantitative (pre- and post-contrast R1, R2 and proton density) MR data was obtained from 23 patients with typical radiological findings suggestive of a high-grade malignant glioma. 2D deep learning models were trained on transversal slices (n=528) for tumor detection and segmentation using either conventional or quantitative data. Moreover, trends in quantitative R1 and R2 rates of regions identified as relevant for tumor detection by model explainability methods were qualitatively analyzed. Tumor detection and segmentation performance for models trained with a combination of qMRI pre- and post-contrast was the highest (detection MCC=0.72, segmentation DSC=0.90), however, the difference compared to cMRI was not statistically significant. Overall analysis of the relevant regions identified using model explainability showed no differences between models trained on cMRI or qMRI. When looking at the individual cases, relaxation rates of brain regions outside the annotation and identified as relevant for tumor detection exhibited changes after contrast injection similar to region inside the annotation in the majority of cases. In conclusion, models trained on qMRI data obtained similar detection and segmentation performance to those trained on cMRI data, with the advantage of quantitatively measuring brain tissue properties within similar scan time. When co, Funding: CENIIT at Linkoeping University, ITEA3 / VINNOVA funded project Intelligence based iMprovement of Personalized treatment And Clinical workflow supporT (IMPACT); ITEA4 / VINNOVA funded project Automation, Surgery Support and Intuitive 3D visualization to optimize workflow in IGT SysTems (ASSIST) [2021-01954]; Cancer Strength Area at Linkoeping University, VINOVA project via the Analytic Imaging Diagnostics Arena (AIDA) [2017-02447]; Medical Research Council of Southeast Sweden [FORSS-234551]; Swedish Research Council [2018-05250]

Background: Stereotactic radiosurgery (SRS) can be an effective primary or adjuvant treatment option for intracranial tumors. However, it carries risks of various radiation toxicities, which can lead to functional deficits for the patients. Current inverse planning algorithms for SRS provide an efficient way for sparing organs at risk (OARs) by setting maximum radiation dose constraints in the treatment planning process.Purpose: We propose using activation maps from functional MRI (fMRI) to map the eloquent regions of the brain and define functional OARs (fOARs) for Gamma Knife SRS treatment planning.Methods: We implemented a pipeline for analyzing patient fMRI data, generating fOARs from the resulting activation maps, and loading them onto the GammaPlan treatment planning software. We used the Lightning inverse planner to generate multiple treatment plans from open MRI data of five subjects, and evaluated the effects of incorporating the proposed fOARs.Results: The Lightning optimizer designs treatment plans with high conformity to the specified parameters. Setting maximum dose constraints on fOARs successfully limits the radiation dose incident on them, but can have a negative impact on treatment plan quality metrics. By masking out fOAR voxels surrounding the tumor target it is possible to achieve high quality treatment plans while controlling the radiation dose on fOARs.Conclusions: The proposed method can effectively reduce the radiation dose incident on the eloquent brain areas during Gamma Knife SRS of brain tumors., Funding: Centrum foer Industriell Informationsteknologi, Linkoepings Universitet; Vinnova [2018-02230, 2021-01954]

There is mounting evidence of the long-term effects of COVID-19 on the central nervous system, with patients experiencing diverse symptoms, often suggesting brain involvement. Conventional brain MRI of these patients shows unspecific patterns, with no clear connection of the symptomatology to brain tissue abnormalities, whereas diffusion tensor studies and volumetric analyses detect measurable changes in the brain after COVID-19. Diffusion MRI exploits the random motion of water molecules to achieve unique sensitivity to structures at the microscopic level, and new sequences employing generalized diffusion encoding provide structural information which are sensitive to intravoxel features. In this observational study, a total of 32 persons were investigated: 16 patients previously hospitalized for COVID-19 with persisting symptoms of post-COVID condition (mean age 60 years: range 41–79, all male) at 7-month follow-up and 16 matched controls, not previously hospitalized for COVID-19, with no post-COVID symptoms (mean age 58 years, range 46–69, 11 males). Standard MRI and generalized diffusion encoding MRI were employed to examine the brain white matter of the subjects. To detect possible group differences, several tissue microstructure descriptors obtainable with the employed diffusion sequence, the fractional anisotropy, mean diffusivity, axial diffusivity, radial diffusivity, microscopic anisotropy, orientational coherence (Cc) and variance in compartment’s size (CMD) were analysed using the tract-based spatial statistics framework. The tract-based spatial statistics analysis showed widespread statistically significant differences (P < 0.05, corrected for multiple comparisons using the familywise error rate) in all the considered metrics in the white matter of the patients compared to the controls. Fractional anisotropy, microscopic anisotropy and Cc were lower in the patient group, while axial diffusivity, radial diffusivity, mean diffusivity and CMD were higher, Funding: Analytic Imaging Diagnostic Arena (AIDA), a Medtech4Health initiative; ITEA/ VINNOVA (The Swedish Innovation Agency) project ASSIST (Automation, Surgery Support and Intuitive 3D visualization to optimize workflow in IGT SysTems) [2021-01954]; Wallenberg Center for Molecular Medicine

Using 3D CNNs on high-resolution medical volumes is very computationally demanding, especially for large datasets like UK Biobank, which aims to scan 100,000 subjects. Here, we demonstrate that using 2D CNNs on a few 2D projections (representing mean and standard deviation across axial, sagittal and coronal slices) of 3D volumes leads to reasonable test accuracy (mean absolute error of about 3.5 years) when predicting age from brain volumes. Using our approach, one training epoch with 20,324 subjects takes 20–50 s using a single GPU, which is two orders of magnitude faster than a small 3D CNN. This speedup is explained by the fact that 3D brain volumes contain a lot of redundant information, which can be efficiently compressed using 2D projections. These results are important for researchers who do not have access to expensive GPU hardware for 3D CNNs., Funding: ITEA/VINNOVA [2021-01954]; Ake Wiberg foundation

Brain age prediction from 3D MRI volumes using deep learning has recently become a popular research topic, as brain age has been shown to be an important biomarker. Training deep networks can be very computationally demanding for large datasets like the U.K. Biobank (currently 29,035 subjects). In our previous work, it was demonstrated that using a few 2D projections (mean and standard deviation along three axes) instead of each full 3D volume leads to much faster training at the cost of a reduction in prediction accuracy. Here, we investigated if another set of 2D projections, based on higher-order statistical central moments and eigenslices, leads to a higher accuracy. Our results show that higher-order moments do not lead to a higher accuracy, but that eigenslices provide a small improvement. We also show that an ensemble of such models provides further improvement., Funding: ITEA/VINNOVA

Deep learning models have achieved prominent performance in digital pathology, with the potential to provide healthcare professionals with accurate decision-making assistance in their workflow. In this study, ViT and CNN models were implemented and compared for patch-level classification of four major glioblastoma tissue structures in histology images. A subset of the IvyGAP dataset (41 subjects, 123 images) was used, stain-normalised and patches of size 256x256 pixels were extracted. A per-subject split approach was applied to obtain training, validation and testing sets. Three models were implemented, a ViT and a CNN trained from scratch, and a ViT pre-trained on a different brain tumour histology dataset. The models' performance was assessed using a range of metrics, including accuracy and Matthew's correlation coefficient (MCC). In addition, calibration experiments were conducted and evaluated to align the models with the ground truth, utilising the temperature scaling technique. The models' uncertainty was estimated using the Monte Carlo dropout method. Lastly, the models were compared using the Wilcoxon signed-rank statistical significance test with Bonferroni correction. Among the models, the scratch-trained ViT obtained the highest test accuracy of 67% and an MCC of 0.45. The scratch-trained CNN reached a test accuracy of 49% and an MCC of 0.15, and the pre-trained ViT only achieved a test accuracy of 28% and an MCC of 0.034. Comparing the reliability graphs and metrics before and after applying temperature scaling, the subsequent experiments proceeded with the uncalibrated ViTs and the calibrated CNN. The calibrated CNN demonstrated moderate to high uncertainty across classes, and the ViTs had an overall high uncertainty. Statistically, there was no difference among the models at a significance level of 0.017. In conclusion, the scratch-trained ViT model considerably outperformed the scratch-trained CNN and the pre-trained ViT in classification. However, t

Classifying subjects as healthy or diseased using neuroimaging data has gained a lot of attention during the last 10 years, and recently, different deep learning approaches have been used. Despite this fact, there has not been any investigation regarding how 3D augmentation can help to create larger datasets, required to train deep networks with millions of parameters. In this study, deep learning was applied to derivatives from resting state functional MRI data, to investigate how different 3D augmentation techniques affect the test accuracy. Specifically, resting state derivatives from 1112 subjects in ABIDE (Autism Brain Imaging Data Exchange) preprocessed were used to train a 3D convolutional neural network (CNN) to classify each subject according to presence or absence of autism spectrum disorder. The results show that augmentation only provide minor improvements to the test accuracy., Funding: Swedish research council [2017-04889]; ITEA/VINNOVA [2021-01954]; Ake Wiberg foundation [M22-0088]

Background: Stereotactic radiosurgery (SRS) can be an effective primary or adjuvant treatment option for intracranial tumors. However, it carries risks of various radiation toxicities, which can lead to functional deficits for the patients. Current inverse planning algorithms for SRS provide an efficient way for sparing organs at risk (OARs) by setting maximum radiation dose constraints in the treatment planning process.Purpose: We propose using activation maps from functional MRI (fMRI) to map the eloquent regions of the brain and define functional OARs (fOARs) for Gamma Knife SRS treatment planning.Methods: We implemented a pipeline for analyzing patient fMRI data, generating fOARs from the resulting activation maps, and loading them onto the GammaPlan treatment planning software. We used the Lightning inverse planner to generate multiple treatment plans from open MRI data of five subjects, and evaluated the effects of incorporating the proposed fOARs.Results: The Lightning optimizer designs treatment plans with high conformity to the specified parameters. Setting maximum dose constraints on fOARs successfully limits the radiation dose incident on them, but can have a negative impact on treatment plan quality metrics. By masking out fOAR voxels surrounding the tumor target it is possible to achieve high quality treatment plans while controlling the radiation dose on fOARs.Conclusions: The proposed method can effectively reduce the radiation dose incident on the eloquent brain areas during Gamma Knife SRS of brain tumors., Funding: Centrum foer Industriell Informationsteknologi, Linkoepings Universitet; Vinnova [2018-02230, 2021-01954]

The infiltrative nature of malignant gliomas results in active tumor spreading into the peritumoral edema, which is not visible in conventional magnetic resonance imaging (cMRI) even after contrast injection. MR relaxometry (qMRI) measures relaxation rates dependent on tissue properties, and can offer additional contrast mechanisms to highlight the non-enhancing infiltrative tumor. To investigate if qMRI data provides additional information compared to cMRI sequences when considering deep learning-based brain tumor detection and segmentation, preoperative conventional (T1-w per- and post-contrast, T2-w and FLAIR) and quantitative (pre- and post-contrast R1, R2 and proton density) MR data was obtained from 23 patients with typical radiological findings suggestive of a high-grade malignant glioma. 2D deep learning models were trained on transversal slices (n=528) for tumor detection and segmentation using either conventional or quantitative data. Moreover, trends in quantitative R1 and R2 rates of regions identified as relevant for tumor detection by model explainability methods were qualitatively analyzed. Tumor detection and segmentation performance for models trained with a combination of qMRI pre- and post-contrast was the highest (detection MCC=0.72, segmentation DSC=0.90), however, the difference compared to cMRI was not statistically significant. Overall analysis of the relevant regions identified using model explainability showed no differences between models trained on cMRI or qMRI. When looking at the individual cases, relaxation rates of brain regions outside the annotation and identified as relevant for tumor detection exhibited changes after contrast injection similar to region inside the annotation in the majority of cases. In conclusion, models trained on qMRI data obtained similar detection and segmentation performance to those trained on cMRI data, with the advantage of quantitatively measuring brain tissue properties within similar scan time. When co, Funding: CENIIT at Linkoeping University, ITEA3 / VINNOVA funded project Intelligence based iMprovement of Personalized treatment And Clinical workflow supporT (IMPACT); ITEA4 / VINNOVA funded project Automation, Surgery Support and Intuitive 3D visualization to optimize workflow in IGT SysTems (ASSIST) [2021-01954]; Cancer Strength Area at Linkoeping University, VINOVA project via the Analytic Imaging Diagnostics Arena (AIDA) [2017-02447]; Medical Research Council of Southeast Sweden [FORSS-234551]; Swedish Research Council [2018-05250]

t-distributed stochastic neighbor embedding (t-SNE) is a method for reducing high-dimensional data to a low-dimensional representation, and is mostly used for visualizing data. In parametric t-SNE, a neural network learns to reproduce this mapping. When used for EEG analysis, the data are usually first transformed into a set of features, but it is not known which features are optimal. The principle of t-SNE was used to train convolutional neural network (CNN) encoders to learn to produce both a high- and a low-dimensional representation, eliminating the need for feature engineering. To evaluate the method, the Temple University EEG Corpus was used to create three datasets with distinct EEG characters: (1) wakefulness and sleep; (2) interictal epileptiform discharges; and (3) seizure activity. The CNN encoders produced low-dimensional representations of the datasets with a structure that conformed well to the EEG characters and generalized to new data. Compared to parametric t-SNE for either a short-time Fourier transform or wavelet representation of the datasets, the developed CNN encoders performed equally well in separating categories, as assessed by support vector machines. The CNN encoders generally produced a higher degree of clustering, both visually and in the number of clusters detected by k-means clustering. The developed principle is promising and could be further developed to create general tools for exploring relations in EEG data., Funding: Linkoping University; University Hospital of Linkoeping; ALF of Region OEstergoetland [LIO-936176, ROE-941359]; ITEA3/VINNOVA

t-distributed stochastic neighbor embedding (t-SNE) is a method for reducing high-dimensional data to a low-dimensional representation, and is mostly used for visualizing data. In parametric t-SNE, a neural network learns to reproduce this mapping. When used for EEG analysis, the data are usually first transformed into a set of features, but it is not known which features are optimal. The principle of t-SNE was used to train convolutional neural network (CNN) encoders to learn to produce both a high- and a low-dimensional representation, eliminating the need for feature engineering. To evaluate the method, the Temple University EEG Corpus was used to create three datasets with distinct EEG characters: (1) wakefulness and sleep; (2) interictal epileptiform discharges; and (3) seizure activity. The CNN encoders produced low-dimensional representations of the datasets with a structure that conformed well to the EEG characters and generalized to new data. Compared to parametric t-SNE for either a short-time Fourier transform or wavelet representation of the datasets, the developed CNN encoders performed equally well in separating categories, as assessed by support vector machines. The CNN encoders generally produced a higher degree of clustering, both visually and in the number of clusters detected by k-means clustering. The developed principle is promising and could be further developed to create general tools for exploring relations in EEG data., Funding: Linkoping University; University Hospital of Linkoeping; ALF of Region OEstergoetland [LIO-936176, ROE-941359]; ITEA3/VINNOVA

Intraoperative guidance tools for thyroid surgery based on optical coherence tomography (OCT) could aid distinguish between normal and diseased tissue. However, OCT images are difficult to interpret, thus, real-time automatic analysis could support the clinical decision-making. In this study, several deep learning models were investigated for thyroid disease classification on 2D and 3D OCT data obtained from ex vivo specimens of 22 patients undergoing surgery and diagnosed with several thyroid pathologies. Additionally, two open-access datasets were used to evaluate the custom models. On the thyroid dataset, the best performance was achieved by the 3D vision transformer model with a Matthews correlation coefficient (MCC) of 0.79 (accuracy = 0.90) for the normal-versus-abnormal classification. On the open-access datasets, the custom models achieved the best performance (MCC > 0.88, accuracy > 0.96). Results obtained for the normal-versus-abnormal classification suggest OCT, complemented with deep learning-based analysis, as a tool for real-time automatic diseased tissue identification in thyroid surgery., Funding: Ake Wiberg Stiftelse [M19-0455 M20-0034 M21-0083]; Forskningsradet i Sydostra Sverige [931466]; VINNOVA via Medtech4Health; AIDA(1908) [2017-02447]; Vetenskapsradet-Swedish Research Council [2018-05250]