Search

Acquisition of high quality manual annotations is vital for the development of segmentation algorithms. However, to create them we require a substantial amount of expert time and knowledge. Large numbers of labels are required to train convolutional neural networks due to the vast number of parameters that must be learned in the optimisation process. Here, we develop the STAMP algorithm to allow the simultaneous training and pruning of a UNet architecture for medical image segmentation with targeted channelwise dropout to make the network robust to the pruning. We demonstrate the technique across segmentation tasks and imaging modalities. It is then shown that, through online pruning, we are able to train networks to have much higher performance than the equivalent standard UNet models while reducing their size by more than 85% in terms of parameters. This has the potential to allow networks to be directly trained on datasets where very low numbers of labels are available.

Accurate topology is key when performing meaningful anatomical segmentations, however, it is often overlooked in traditional deep learning methods. In this work we propose TEDS-Net: a novel segmentation method that guarantees accurate topology. Our method is built upon a continuous diffeomorphic framework, which enforces topology preservation. However, in practice, diffeomorphic fields are represented using a finite number of parameters and sampled using methods such as linear interpolation, violating the theoretical guarantees. We therefore introduce additional modifications to more strictly enforce it. Our network learns how to warp a binary prior, with the desired topological characteristics, to complete the segmentation task. We tested our method on myocardium segmentation from an open-source 2D heart dataset. TEDS-Net preserved topology in 100% of the cases, compared to 90% from the U-Net, without sacrificing on Hausdorff Distance or Dice performance. Code will be made available at: www.github.com/mwyburd/TEDS-Net, International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI) 2021

Segmentation of anatomical structures from Cardiac Magnetic Resonance (CMR) is central to the non-invasive quantitative assessment of cardiac function and structure. Anatomical variability, imaging heterogeneity and cardiac dynamics challenge the automation of this task. Deep learning (DL) approaches have taken over the field of automatic segmentation in recent years, however they are limited by data availability and the additional variability introduced by differences in scanners and protocols. In this work, we propose a 2-step fully automated pipeline to segment CMR images, based on DL encoder-decoder frameworks, and we explore two domain adaptation techniques, domain adversarial training and iterative domain unlearning, to overcome the imaging heterogeneity limitations. We evaluate our methods on the MICCAI 2020 Multi-Centre, Multi-Vendor & Multi-Disease Cardiac Image Segmentation Challenge training and validation datasets. The results show the improvement in performance produced by domain adaptation models, especially among the seen vendors. Finally, we build an ensemble of baseline and domain adapted networks, that reported state-of-art mean Dice scores of 0.912, 0.857 and 0.861 for left ventricle (LV) cavity, LV myocardium and right ventricle cavity, respectively, on the externally validated Challenge dataset, including several unseen vendors, centers and cardiac pathologies. Keywords: Segmentation, Cardiac Magnetic Resonance, Deep Learning, Domain Adaptation, Data harmonization

The combination of datasets is vital for providing increased statistical power, and is especially important for neurological conditions where limited data is available. However, our ability to combine datasets is limited by the addition of variance caused by factors such as differences in acquisition protocol and hardware. We aim to create scanner-invariant features using an iterative training scheme based on domain adaptation techniques, whilst simultaneously completing the desired segmentation task. We demonstrate the technique using an encoder-decoder architecture similar to the U-Net but expect that the proposed training scheme would be applicable to any feedforward network and task. We show that the network can be used to harmonise two datasets and also show that the network is applicable in the common scenario of limited available training data, meaning that the network should be applicable for real-world segmentation problems.

Highlights • We explored various domain adaptation methods for robust WM lesion segmentation. • We used a triplanar U-net ensemble network (TrUE-Net) as our baseline model. • Transfer learning: fine-tuning from the coarsest encoder layer gave good results. • Semi-supervised domain adversarial training of NNs (DANN) performed the best. • Among unsupervised methods, DANN performed better than domain unlearning., Graphical abstract, Robust automated segmentation of white matter hyperintensities (WMHs) in different datasets (domains) is highly challenging due to differences in acquisition (scanner, sequence), population (WMH amount and location) and limited availability of manual segmentations to train supervised algorithms. In this work we explore various domain adaptation techniques such as transfer learning and domain adversarial learning methods, including domain adversarial neural networks and domain unlearning, to improve the generalisability of our recently proposed triplanar ensemble network, which is our baseline model. We used datasets with variations in intensity profile, lesion characteristics and acquired using different scanners. For the source domain, we considered a dataset consisting of data acquired from 3 different scanners, while the target domain consisted of 2 datasets. We evaluated the domain adaptation techniques on the target domain datasets, and additionally evaluated the performance on the source domain test dataset for the adversarial techniques. For transfer learning, we also studied various training options such as minimal number of unfrozen layers and subjects required for fine-tuning in the target domain. On comparing the performance of different techniques on the target dataset, domain adversarial training of neural network gave the best performance, making the technique promising for robust WMH segmentation.

Combining datasets is vital for increased statistical power, especially for neurological conditions where limited data is available. However, variance due to differences in acquisition protocol and hardware limits our ability to combine datasets. We propose an iterative training scheme based on domain adaptation techniques, aiming to create scanner-invariant features while simultaneously maintaining overall performance on the main task. We demonstrate this on age prediction, but expect that our proposed training scheme will be applicable to any feedforward network and classification or regression task. We show that not only can we harmonise three MRI datasets from different studies, but can also successfully adapt the training to work with very biased datasets. The training scheme should, therefore, be applicable to most real-world data scenarios, enabling harmonisation for the task of interest.

Accurate segmentations of neuroanatomical structures are essential for volumetric and morphological assessment, but manual segmentation is time-consuming and error-prone. We propose a convolutional neural network for structural segmentation based on deformation of an example mask that is disease-state agnostic, which we apply to the hippocampus. The hippocampus is one of the first subcortical structures affected by Alzheimer’s disease, atrophying as the disease progresses. As the disease state may be unknown, and due to the varying degrees of atrophy, an accurate shape prior is not always available. The proposed network is based on an adapted spatial transformer network that learns a deformation field to resample an initial binary mask, to create an output segmentation. This segmentation is learnt by the network from the input T1-weighted MRI in an end-to-end manner. Experiments on the HarP dataset show that the network outperforms other segmentation methods and is consistent across disease states, independent of the degree of disease-related atrophy. We also explore the effect of the initial binary mask on the segmentation, showing that the it is insensitive to the size and initial location of the binary mask.

Highlights • We demonstrate a flexible deep-learning-based harmonisation framework. • Applied to age prediction and segmentation tasks in a range of datasets. • Scanner information is removed, maintaining performance and improving generalisability. • The framework can be used with any feedforward network architecture. • It successfully removes additional confounds and works with varied distributions., Increasingly large MRI neuroimaging datasets are becoming available, including many highly multi-site multi-scanner datasets. Combining the data from the different scanners is vital for increased statistical power; however, this leads to an increase in variance due to nonbiological factors such as the differences in acquisition protocols and hardware, which can mask signals of interest. We propose a deep learning based training scheme, inspired by domain adaptation techniques, which uses an iterative update approach to aim to create scanner-invariant features while simultaneously maintaining performance on the main task of interest, thus reducing the influence of scanner on network predictions. We demonstrate the framework for regression, classification and segmentation tasks with two different network architectures. We show that not only can the framework harmonise many-site datasets but it can also adapt to many data scenarios, including biased datasets and limited training labels. Finally, we show that the framework can be extended for the removal of other known confounds in addition to scanner. The overall framework is therefore flexible and should be applicable to a wide range of neuroimaging studies.

Both normal ageing and neurodegenerative diseases cause morphological changes to the brain. Age-related brain changes are subtle, nonlinear, and spatially and temporally heterogenous, both within a subject and across a population. Machine learning models are particularly suited to capture these patterns and can produce a model that is sensitive to changes of interest, despite the large variety in healthy brain appearance. In this paper, the power of convolutional neural networks (CNNs) and the rich UK Biobank dataset, the largest database currently available, are harnessed to address the problem of predicting brain age. We developed a 3D CNN architecture to predict chronological age, using a training dataset of 12, 802 T1-weighted MRI images and a further 6, 885 images for testing. The proposed method shows competitive performance on age prediction, but, most importantly, the CNN prediction errors ΔBrainAge = AgePredicted − AgeTrue correlated significantly with many clinical measurements from the UK Biobank in the female and male groups. In addition, having used images from only one imaging modality in this experiment, we examined the relationship between ΔBrainAge and the image-derived phenotypes (IDPs) from all other imaging modalities in the UK Biobank, showing correlations consistent with known patterns of ageing. Furthermore, we show that the use of nonlinearly registered images to train CNNs can lead to the network being driven by artefacts of the registration process and missing subtle indicators of ageing, limiting the clinical relevance. Due to the longitudinal aspect of the UK Biobank study, in the future it will be possible to explore whether the ΔBrainAge from models such as this network were predictive of any health outcomes.HighlightsBrain age is estimated using a 3D CNN from 12,802 full T1-weighted images.Regions used to drive predictions are different for linearly and nonlinearly registered data.Linear registrations utilise a greater diversity of biologically meaningful areas.Correlations with IDPs and non-imaging variables are consistent with other publications.Excluding subjects with various health conditions had minimal impact on main correlations.