Search

The brain lesions images of Alzheimer's disease (AD) patients are slightly different from the Magnetic Resonance Imaging of normal people, and the classification effect of general image recognition technology is not ideal. Alzheimer's datasets are small, making it difficult to train large-scale neural networks. In this paper, we propose a network model (WS-AMN) that fuses weak supervision and an attention mechanism. The weakly supervised data augmentation network is used as the basic model, the attention map generated by weakly supervised learning is used to guide the data augmentation, and an attention module with channel domain and spatial domain is embedded in the residual network to focus on the distinctive channels and spaces of images respectively. The location information enhances the corresponding features of related features and suppresses the influence of irrelevant features.The results show that the F1-score is 99.63%, the accuracy is 99.61%. Our model provides a high-performance solution for accurate classification of AD. [ABSTRACT FROM AUTHOR]

Cerebral micro-bleedings (CMBs) are small chronic brain hemorrhages that have many side effects. For example, CMBs can result in long-term disability, neurologic dysfunction, cognitive impairment and side effects from other medications and treatment. Therefore, it is important and essential to detect CMBs timely and in an early stage for prompt treatment. In this research, because of the limited labeled samples, it is hard to train a classifier to achieve high accuracy. Therefore, we proposed employing Densely connected neural network (DenseNet) as the basic algorithm for transfer learning to detect CMBs. To generate the subsamples for training and test, we used a sliding window to cover the whole original images from left to right and from top to bottom. Based on the central pixel of the subsamples, we could decide the target value. Considering the data imbalance, the cost matrix was also employed. Then, based on the new model, we tested the classification accuracy, and it achieved 97.71%, which provided better performance than the state of art methods. [ABSTRACT FROM AUTHOR]

Aim: This paper proposes a novel alcoholism identification approach that can assist radiologists in patient diagnosis. Method: AlexNet was used as the basic transfer learning model. The global learning rate was small, at 10−4, and the iteration epoch number was at 10. The learning rate factor of replaced layers was 10 times larger than that of the transferred layers. We tested five different replacement configurations of transfer learning. Results: The experiment shows that the best performance was achieved by replacing the final fully connected layer. Our method yielded a sensitivity of 97.44%± 1.15%, a specificity of 97.41 ± 1.51%, a precision of 97.34 ± 1.49%, an accuracy of 97.42 ± 0.95%, and an F1 score of 97.37 ± 0.97% on the test set. Conclusion: This method can assist radiologists in their routine alcoholism screening of brain magnetic resonance images. [ABSTRACT FROM AUTHOR]

Aim: Multiple sclerosis is a severe brain and/or spinal cord disease. It may lead to a wide range of symptoms. Hence, the early diagnosis and treatment is quite important. Method: This study proposed a 14-layer convolutional neural network, combined with three advanced techniques: batch normalization, dropout, and stochastic pooling. The output of the stochastic pooling was obtained via sampling from a multinomial distribution formed from the activations of each pooling region. In addition, we used data augmentation method to enhance the training set. In total 10 runs were implemented with the hold-out randomly set for each run. Results: The results showed that our 14-layer CNN secured a sensitivity of 98.77 ± 0.35%, a specificity of 98.76 ± 0.58%, and an accuracy of 98.77 ± 0.39%. Conclusion: Our results were compared with CNN using maximum pooling and average pooling. The comparison shows stochastic pooling gives better performance than other two pooling methods. Furthermore, we compared our proposed method with six state-of-the-art approaches, including five traditional artificial intelligence methods and one deep learning method. The comparison shows our method is superior to all other six state-of-the-art approaches. [ABSTRACT FROM AUTHOR]

Background: Capturing the segmentation of blood vessels by a fundus camera is crucial for the medical evaluation of various retinal vascular issues. However, due to the complicated vascular structure and unclear clinical criteria, the precise segmentation of blood arteries remains very challenging., Methods: To address this issue, we developed the upgraded multi-convolution block and squeeze and excitation based on the U-shape network (MCSE-U-net) model that segments retinal vessels using a U-shaped network. This model uses multi-convolution (MC) blocks, squeeze and excitation (SE) blocks, and squeeze blocks. First, the input image was processed using the luminance, chrominance-blue, chrominance-red (YCbCr) color conversion method to further improve visibility. Second, a MC module was added to increase the model's ability to accurately segment blood vessels. Third, SE blocks were added to enhance the network model's ability to segment fine blood vessels in medical images., Results: The suggested architecture was assessed using evaluation metrics, including the Dice coefficient, sensitivity (sen), specificity (spe), accuracy (acc), and mean intersection over union (mIoU), on an open-source Digital Retinal Images for Vessel Extraction (DRIVE) data set. The outcomes showed the effectiveness of the suggested approach, particularly in the extraction of peripheral vascular anatomy. Using the suggested architecture, the model had a Dice coefficient of 0.8430, a sen of 0.8752, a spe of 0.9902, an acc of 0.9725, and a mIoU of 0.8473 for the DRIVE data set. The Dice coefficient, sen, spe, acc, and mIoU of the MCSE-U-net increased by 3.08%, 6.22%, 0.62%, 0.61%, and 3.01%, respectively, compared to the original U-net, demonstrating the better all-around performance of the MCSE-U-net., Conclusions: The MCSE-U-net network performed and achieved more than the technologies already in use., Competing Interests: Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://qims.amegroups.com/article/view/10.21037/qims-23-1454/coif). The authors have no conflicts of interest to declare., (2024 Quantitative Imaging in Medicine and Surgery. All rights reserved.)

Background: Chemical exchange saturation transfer (CEST) is a promising method for the detection of biochemical alterations in cancers and neurological diseases. However, the sensitivity of the currently existing quantitative method for detecting ischemia needs further improvement., Methods: To further improve the quantification of the CEST signal and enhance the CEST detection for ischemia, we used a quantitative analysis method that combines an inverse Z-spectrum analysis and a 5-pool Lorentzian fitting. Specifically, a 5-pool Lorentzian simulation was conducted with the following brain tissue parameters: water, amide (3.5 ppm), amine (2.2 ppm), magnetization transfer (MT), and nuclear Overhauser enhancement (NOE; -3.5 ppm). The parameters were first calculated offline and stored as the initial value of the Z-spectrum fitting. Then, the measured Z-spectrum with the peak value set to 0 was fitted via the stored initial value, which yielded the reference Z-spectrum. Finally, the difference between the inverse of the Z-spectrum and the inverse of the reference Z-spectrum was used as the CEST definite spectrum., Results: The simulation results demonstrated that the Z-spectra of the rat brain were well simulated by a 5-pool Lorentzian fitting. Further, the proposed method detected a larger difference than did either the saturation transfer difference or the 5-pool Lorentzian fitting, as demonstrated by simulations. According to the results of the cerebral ischemia rat model, the proposed method provided the highest contrast-to-noise ratio (CNR) between the contralateral and the ipsilateral striatum under various acquisition conditions. The results indicated that the difference of fitted amplitudes generated with a 5-pool Lorentzian fitting in amide at 3.5 ppm (6.04%±0.39%; 6.86%±0.39%) was decreased in a stroke lesion compared to the contralateral normal tissue. Moreover, the difference of the residual of inversed Z-spectra in which 5-pool Lorentzian fitting was used to calculate the reference Z-spectra ( M T R R e x 5 L ) amplitudes in amide at 3.5 ppm (13.83%±2.20%, 15.69%±1.99%) was reduced in a stroke lesion compared to the contralateral normal tissue., Conclusions: M T R R e x 5 L is predominantly pH-sensitive and is suitable for detecting tissue acidosis following an acute stroke., Competing Interests: Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://qims.amegroups.com/article/view/10.21037/qims-22-420/coif). The authors have no conflicts of interest to declare., (2023 Quantitative Imaging in Medicine and Surgery. All rights reserved.)

By training different models and averaging their predictions, the performance of the machine-learning algorithm can be improved. The performance optimization of multiple models is supposed to generalize further data well. This requires the knowledge transfer of generalization information between models. In this article, a multiple kernel mutual learning method based on transfer learning of combined mid-level features is proposed for hyperspectral classification. Three-layer homogenous superpixels are computed on the image formed by PCA, which is used for computing mid-level features. The three mid-level features include: 1) the sparse reconstructed feature; 2) combined mean feature; and 3) uniqueness. The sparse reconstruction feature is obtained by a joint sparse representation model under the constraint of three-scale superpixels' boundaries and regions. The combined mean features are computed with average values of spectra in multilayer superpixels, and the uniqueness is obtained by the superposed manifold ranking values of multilayer superpixels. Next, three kernels of samples in different feature spaces are computed for mutual learning by minimizing the divergence. Then, a combined kernel is constructed to optimize the sample distance measurement and applied by employing SVM training to build classifiers. Experiments are performed on real hyperspectral datasets, and the corresponding results demonstrated that the proposed method can perform significantly better than several state-of-the-art competitive algorithms based on MKL and deep learning.

This paper presents an improved local ternary pattern (LTP) for automatic target recognition (ATR) in infrared imagery. Firstly, a robust LTP (RLTP) scheme is proposed to overcome the limitation of the original LTP for achieving the invariance with respect to the illumination transformation. Then, a soft concave-convex partition (SCCP) is introduced to add some flexibility to the original concave-convex partition (CCP) scheme. Referring to the orthogonal combination of local binary patterns (OC_LBP), the orthogonal combination of LTP (OC_LTP) is adopted to reduce the dimensionality of the LTP histogram. Further, a novel operator, called the soft concave-convex orthogonal combination of robust LTP (SCC_OC_RLTP), is proposed by combing RLTP, SCCP and OC_LTP. Finally, the new operator is used for ATR along with a blocking schedule to improve its discriminability and a feature selection technique to enhance its efficiency. Experimental results on infrared imagery show that the proposed features can achieve competitive ATR results compared with the state-of-the-art methods.