Multi-task learning approach for volumetric segmentation and reconstruction in 3D OCT images

The choroid is the vascular layer of the eye that supplies photoreceptors with oxygen. Changes in the choroid are associated with many pathologies including myopia where the choroid progressively thins due to axial elongation. To quantize these changes, there is a need to automatically and accurately segment the choroidal layer from optical coherence tomography (OCT) images. In this paper, we propose a multi-task learning approach to segment the choroid from three-dimensional OCT images. Our proposed architecture aggregates the spatial context from adjacent cross-sectional slices to reconstruct the central slice. Spatial context learned by this reconstruction mechanism is then fused with a U-Net based architecture for segmentation. The proposed approach was evaluated on volumetric OCT scans of 166 myopic eyes acquired with a commercial OCT system, and achieved a cross-validation Intersection over Union (IoU) score of 94.69% which significantly outperformed (p<0.001) the other state-of-the-art methods on the same data set. Choroidal thickness maps generated by our approach also achieved a better structural similarity index (SSIM) of 72.11% with respect to the groundtruth. In particular, our approach performs well for highly challenging eyes with thinner choroids. Compared to other methods, our proposed approach also requires lesser processing time and has lower computational requirements. The results suggest that our proposed approach could potentially be used as a fast and reliable method for automated choroidal segmentation.     

Multi-task learning for segmentation and classification of tumors in 3D automated breast ultrasound images

Tumor classification and segmentation are two important tasks for computer-aided diagnosis (CAD) using 3D automated breast ultrasound (ABUS) images. However, they are challenging due to the significant shape variation of breast tumors and the fuzzy nature of ultrasound images (e.g., low contrast and signal to noise ratio). Considering the correlation between tumor classification and segmentation, we argue that learning these two tasks jointly is able to improve the outcomes of both tasks. In this paper, we propose a novel multi-task learning framework for joint segmentation and classification of tumors in ABUS images. The proposed framework consists of two sub-networks: an encoder-decoder network for segmentation and a light-weight multi-scale network for classification. To account for the fuzzy boundaries of tumors in ABUS images, our framework uses an iterative training strategy to refine feature maps with the help of probability maps obtained from previous iterations. Experimental results based on a clinical dataset of 170 3D ABUS volumes collected from 107 patients indicate that the proposed multi-task framework improves tumor segmentation and classification over the single-task learning counterparts.     

The predictive value of segmentation metrics on dosimetry in organs at risk of the brain

BackgroundFully automatic medical image segmentation has been a long pursuit in radiotherapy (RT). Recent developments involving deep learning show promising results yielding consistent and time efficient contours. In order to train and validate these systems, several geometric based metrics, such as Dice Similarity Coefficient (DSC), Hausdorff, and other related metrics are currently the standard in automated medical image segmentation challenges. However, the relevance of these metrics in RT is questionable. The quality of automated segmentation results needs to reflect clinical relevant treatment outcomes, such as dosimetry and related tumor control and toxicity. In this study, we present results investigating the correlation between popular geometric segmentation metrics and dose parameters for Organs-At-Risk (OAR) in brain tumor patients, and investigate properties that might be predictive for dose changes in brain radiotherapy.MethodsA retrospective database of glioblastoma multiforme patients was stratified for planning difficulty, from which 12 cases were selected and reference sets of OARs and radiation targets were defined. In order to assess the relation between segmentation quality -as measured by standard segmentation assessment metrics- and quality of RT plans, clinically realistic, yet alternative contours for each OAR of the selected cases were obtained through three methods: (i) Manual contours by two additional human raters. (ii) Realistic manual manipulations of reference contours. (iii) Through deep learning based segmentation results. On the reference structure set a reference plan was generated that was re-optimized for each corresponding alternative contour set. The correlation between segmentation metrics, and dosimetric changes was obtained and analyzed for each OAR, by means of the mean dose and maximum dose to 1% of the volume (Dmax 1%). Furthermore, we conducted specific experiments to investigate the dosimetric effect of alternative OAR contours with respect to the proximity to the target, size, particular shape and relative location to the target.ResultsWe found a low correlation between the DSC, reflecting the alternative OAR contours, and dosimetric changes. The Pearson correlation coefficient between the mean OAR dose effect and the Dice was -0.11. For Dmax 1%, we found a correlation of -0.13. Similar low correlations were found for 22 other segmentation metrics. The organ based analysis showed that there is a better correlation for the larger OARs (i.e. brainstem and eyes) as for the smaller OARs (i.e. optic nerves and chiasm). Furthermore, we found that proximity to the target does not make contour variations more susceptible to the dose effect. However, the direction of the contour variation with respect to the relative location of the target seems to have a strong correlation with the dose effect.ConclusionsThis study shows a low correlation between segmentation metrics and dosimetric changes for OARs in brain tumor patients. Results suggest that the current metrics for image segmentation in RT, as well as deep learning systems employing such metrics, need to be revisited towards clinically oriented metrics that better reflect how segmentation quality affects dose distribution and related tumor control and toxicity.     

Multi-perspective label based deep learning framework for cerebral vasculature segmentation in whole-brain fluorescence images

The popularity of fluorescent labelling and mesoscopic optical imaging techniques enable the acquisition of whole mammalian brain vasculature images at capillary resolution. Segmentation of the cerebrovascular network is essential for analyzing the cerebrovascular structure and revealing the pathogenesis of brain diseases. Existing deep learning methods use a single type of annotated labels with the same pixel weight to train the neural network and segment vessels. Due to the variation in the shape, density and brightness of vessels in whole-brain fluorescence images, it is difficult for a neural network trained with a single type of label to segment all vessels accurately. To address this problem, we proposed a deep learning cerebral vasculature segmentation framework based on multi-perspective labels. First, the pixels in the central region of thick vessels and the skeleton region of vessels were extracted separately using morphological operations based on the binary annotated labels to generate two different labels. Then, we designed a three-stage 3D convolutional neural network containing three sub-networks, namely thick-vessel enhancement network, vessel skeleton enhancement network and multi-channel fusion segmentation network. The first two sub-networks were trained by the two labels generated in the previous step, respectively, and pre-segmented the vessels. The third sub-network was responsible for fusing the pre-segmented results to precisely segment the vessels. We validated our method on two mouse cerebral vascular datasets generated by different fluorescence imaging modalities. The results showed that our method outperforms the state-of-the-art methods, and the proposed method can be applied to segment the vasculature on large-scale volumes.     

Calibrating segmentation networks with margin-based label smoothing

Despite the undeniable progress in visual recognition tasks fueled by deep neural networks, there exists recent evidence showing that these models are poorly calibrated, resulting in over-confident predictions. The standard practices of minimizing the cross-entropy loss during training promote the predicted softmax probabilities to match the one-hot label assignments. Nevertheless, this yields a pre-softmax activation of the correct class that is significantly larger than the remaining activations, which exacerbates the miscalibration problem. Recent observations from the classification literature suggest that loss functions that embed implicit or explicit maximization of the entropy of predictions yield state-of-the-art calibration performances. Despite these findings, the impact of these losses in the relevant task of calibrating medical image segmentation networks remains unexplored. In this work, we provide a unifying constrained-optimization perspective of current state-of-the-art calibration losses. Specifically, these losses could be viewed as approximations of a linear penalty (or a Lagrangian term) imposing equality constraints on logit distances. This points to an important limitation of such underlying equality constraints, whose ensuing gradients constantly push towards a non-informative solution, which might prevent from reaching the best compromise between the discriminative performance and calibration of the model during gradient-based optimization. Following our observations, we propose a simple and flexible generalization based on inequality constraints, which imposes a controllable margin on logit distances. Comprehensive experiments on a variety of public medical image segmentation benchmarks demonstrate that our method sets novel state-of-the-art results on these tasks in terms of network calibration, whereas the discriminative performance is also improved. The code is available at https://github.com/Bala93/MarginLoss     

Non-local statistical label fusion for multi-atlas segmentation

Multi-atlas segmentation provides a general purpose, fully-automated approach for transferring spatial information from an existing dataset (“atlases”) to a previously unseen context (“target”) through image registration. The method to resolve voxelwise label conflicts between the registered atlases (“label fusion”) has a substantial impact on segmentation quality. Ideally, statistical fusion algorithms (e.g., STAPLE) would result in accurate segmentations as they provide a framework to elegantly integrate models of rater performance. The accuracy of statistical fusion hinges upon accurately modeling the underlying process of how raters err. Despite success on human raters, current approaches inaccurately model multi-atlas behavior as they fail to seamlessly incorporate exogenous intensity information into the estimation process. As a result, locally weighted voting algorithms represent the de facto standard fusion approach in clinical applications. Moreover, regardless of the approach, fusion algorithms are generally dependent upon large atlas sets and highly accurate registration as they implicitly assume that the registered atlases form a collectively unbiased representation of the target. Herein, we propose a novel statistical fusion algorithm, Non-Local STAPLE (NLS). NLS reformulates the STAPLE framework from a non-local means perspective in order to learn what label an atlas would have observed, given perfect correspondence. Through this reformulation, NLS (1) seamlessly integrates intensity into the estimation process, (2) provides a theoretically consistent model of multi-atlas observation error, and (3) largely diminishes the need for large atlas sets and very high-quality registrations. We assess the sensitivity and optimality of the approach and demonstrate significant improvement in two empirical multi-atlas experiments.     

One model is all you need: Multi-task learning enables simultaneous histology image segmentation and classification

The recent surge in performance for image analysis of digitised pathology slides can largely be attributed to the advances in deep learning. Deep models can be used to initially localise various structures in the tissue and hence facilitate the extraction of interpretable features for biomarker discovery. However, these models are typically trained for a single task and therefore scale poorly as we wish to adapt the model for an increasing number of different tasks. Also, supervised deep learning models are very data hungry and therefore rely on large amounts of training data to perform well. In this paper, we present a multi-task learning approach for segmentation and classification of nuclei, glands, lumina and different tissue regions that leverages data from multiple independent data sources. While ensuring that our tasks are aligned by the same tissue type and resolution, we enable meaningful simultaneous prediction with a single network. As a result of feature sharing, we also show that the learned representation can be used to improve the performance of additional tasks via transfer learning, including nuclear classification and signet ring cell detection. As part of this work, we train our developed Cerberus model on a huge amount of data, consisting of over 600 thousand objects for segmentation and 440 thousand patches for classification. We use our approach to process 599 colorectal whole-slide images from TCGA, where we localise 377 million, 900 thousand and 2.1 million nuclei, glands and lumina respectively. We make this resource available to remove a major barrier in the development of explainable models for computational pathology.     

Multi-task multi-modal learning for joint diagnosis and prognosis of human cancers

With the tremendous development of artificial intelligence, many machine learning algorithms have been applied to the diagnosis of human cancers. Recently, rather than predicting categorical variables (e.g., stages and subtypes) as in cancer diagnosis, several prognosis prediction models basing on patients’ survival information have been adopted to estimate the clinical outcome of cancer patients. However, most existing studies treat the diagnosis and prognosis tasks separately. In fact, the diagnosis information (e.g., TNM Stages) indicates the extent of the disease severity that is highly correlated with the patients’ survival. While the diagnosis is largely made based on histopathological images, recent studies have also demonstrated that integrative analysis of histopathological images and genomic data can hold great promise for improving the diagnosis and prognosis of cancers. However, direct combination of these two types of data may bring redundant features that will negatively affect the prediction performance. Therefore, it is necessary to select informative features from the derived multi-modal data. Based on the above considerations, we propose a multi-task multi-modal feature selection method for joint diagnosis and prognosis of cancers. Specifically, we make use of the task relationship learning framework to automatically discover the relationships between the diagnosis and prognosis tasks, through which we can identify important image and genomics features for both tasks. In addition, we add a regularization term to ensure that the correlation within the multi-modal data can be captured. We evaluate our method on three cancer datasets from The Cancer Genome Atlas project, and the experimental results verify that our method can achieve better performance on both diagnosis and prognosis tasks than the related methods.     

MIM软件自动勾画前列腺癌危及器官

目的 观察MIM软件自动勾画前列腺癌危及器官的效果，筛选最佳数据库病例数和匹配数。方法 随机选取111例前列腺癌CT资料，以1例为Atlas模板、以100例建立5个数据库(n=20、40、60、80、100),其余10例为测试集；根据戴斯相似系数(DSC)、雅卡尔相似系数(JSC)、豪斯多夫距离(HD)及平均最小距离(MDA)评价MIM软件于不同匹配数(5、9)下对各数据库自动勾画危及器官的效果。结果 MIM软件勾画双侧股骨头、膀胱及脊髓效果较好，勾画小肠和直肠效果一般。匹配数为5时，MIM软件勾画各数据库各危及器官效果差异均无统计学意义(P均>0.05);匹配数为9时，其勾画20例数据库小肠的DSC、JSC优于其他数据库(P均<0.05)。MIM软件勾画100例数据库直肠的DSC优于40例数据库(P<0.05),勾画40例数据库直肠的JSC优于100例数据库(P<0.05)。结论 MIM软件自动勾画前列腺癌危及器官较好；以40例为模板数据库、选择匹配数5可提高勾画小肠和直肠效果。     

Integration of light scattering with machine learning for label free cell detection

Light scattering has been used for label-free cell detection. The angular light scattering patterns from the cells are unique to them based on the cell size, nucleus size, number of mitochondria, and cell surface roughness. The patterns collected from the cells can then be classified based on different image characteristics. We have also developed a machine learning (ML) method to classify these cell light scattering patterns. As a case study we have used this light scattering technique integrated with the machine learning to analyze staurosporine-treated SH-SY5Y neuroblastoma cells and compare them to non-treated control cells. Experimental results show that the ML technique can provide a classification accuracy (treated versus non-treated) of over 90%. The predicted percentage of the treated cells in a mixed solution is within 5% of the reference (ground-truth) value and the technique has the potential to be a viable method for real-time detection and diagnosis.     

Cross-dimensional transfer learning in medical image segmentation with deep learning

Over the last decade, convolutional neural networks have emerged and advanced the state-of-the-art in various image analysis and computer vision applications. The performance of 2D image classification networks is constantly improving and being trained on databases made of millions of natural images. Conversely, in the field of medical image analysis, the progress is also remarkable but has mainly slowed down due to the relative lack of annotated data and besides, the inherent constraints related to the acquisition process. These limitations are even more pronounced given the volumetry of medical imaging data. In this paper, we introduce an efficient way to transfer the efficiency of a 2D classification network trained on natural images to 2D, 3D uni- and multi-modal medical image segmentation applications. In this direction, we designed novel architectures based on two key principles: weight transfer by embedding a 2D pre-trained encoder into a higher dimensional U-Net, and dimensional transfer by expanding a 2D segmentation network into a higher dimension one. The proposed networks were tested on benchmarks comprising different modalities: MR, CT, and ultrasound images. Our 2D network ranked first on the CAMUS challenge dedicated to echo-cardiographic data segmentation and surpassed the state-of-the-art. Regarding 2D/3D MR and CT abdominal images from the CHAOS challenge, our approach largely outperformed the other 2D-based methods described in the challenge paper on Dice, RAVD, ASSD, and MSSD scores and ranked third on the online evaluation platform. Our 3D network applied to the BraTS 2022 competition also achieved promising results, reaching an average Dice score of 91.69% (91.22%) for the whole tumor, 83.23% (84.77%) for the tumor core and 81.75% (83.88%) for enhanced tumor using the approach based on weight (dimensional) transfer. Experimental and qualitative results illustrate the effectiveness of our methods for multi-dimensional medical image segmentation.     

Diverse data augmentation for learning image segmentation with cross-modality annotations

The dearth of annotated data is a major hurdle in building reliable image segmentation models. Manual annotation of medical images is tedious, time-consuming, and significantly variable across imaging modalities. The need for annotation can be ameliorated by leveraging an annotation-rich source modality in learning a segmentation model for an annotation-poor target modality. In this paper, we introduce a diverse data augmentation generative adversarial network (DDA-GAN) to train a segmentation model for an unannotated target image domain by borrowing information from an annotated source image domain. This is achieved by generating diverse augmented data for the target domain by one-to-many source-to-target translation. The DDA-GAN uses unpaired images from the source and target domains and is an end-to-end convolutional neural network that (i) explicitly disentangles domain-invariant structural features related to segmentation from domain-specific appearance features, (ii) combines structural features from the source domain with appearance features randomly sampled from the target domain for data augmentation, and (iii) train the segmentation model with the augmented data in the target domain and the annotations from the source domain. The effectiveness of our method is demonstrated both qualitatively and quantitatively in comparison with the state of the art for segmentation of craniomaxillofacial bony structures via MRI and cardiac substructures via CT.     

Multi-task learning based structured sparse canonical correlation analysis for brain imaging genetics

The advances in technologies for acquiring brain imaging and high-throughput genetic data allow the researcher to access a large amount of multi-modal data. Although the sparse canonical correlation analysis is a powerful bi-multivariate association analysis technique for feature selection, we are still facing major challenges in integrating multi-modal imaging genetic data and yielding biologically meaningful interpretation of imaging genetic findings. In this study, we propose a novel multi-task learning based structured sparse canonical correlation analysis (MTS2CCA) to deliver interpretable results and improve integration in imaging genetics studies. We perform comparative studies with state-of-the-art competing methods on both simulation and real imaging genetic data. On the simulation data, our proposed model has achieved the best performance in terms of canonical correlation coefficients, estimation accuracy, and feature selection accuracy. On the real imaging genetic data, our proposed model has revealed promising features of single-nucleotide polymorphisms and brain regions related to sleep. The identified features can be used to improve clinical score prediction using promising imaging genetic biomarkers. An interesting future direction is to apply our model to additional neurological or psychiatric cohorts such as patients with Alzheimer’s or Parkinson’s disease to demonstrate the generalizability of our method.     

Distributed contrastive learning for medical image segmentation

Supervised deep learning needs a large amount of labeled data to achieve high performance. However, in medical imaging analysis, each site may only have a limited amount of data and labels, which makes learning ineffective. Federated learning (FL) can learn a shared model from decentralized data. But traditional FL requires fully-labeled data for training, which is very expensive to obtain. Self-supervised contrastive learning (CL) can learn from unlabeled data for pre-training, followed by fine-tuning with limited annotations. However, when adopting CL in FL, the limited data diversity on each site makes federated contrastive learning (FCL) ineffective. In this work, we propose two federated self-supervised learning frameworks for volumetric medical image segmentation with limited annotations. The first one features high accuracy and fits high-performance servers with high-speed connections. The second one features lower communication costs, suitable for mobile devices. In the first framework, features are exchanged during FCL to provide diverse contrastive data to each site for effective local CL while keeping raw data private. Global structural matching aligns local and remote features for a unified feature space among different sites. In the second framework, to reduce the communication cost for feature exchanging, we propose an optimized method FCLOpt that does not rely on negative samples. To reduce the communications of model download, we propose the predictive target network update (PTNU) that predicts the parameters of the target network. Based on PTNU, we propose the distance prediction (DP) to remove most of the uploads of the target network. Experiments on a cardiac MRI dataset show the proposed two frameworks substantially improve the segmentation and generalization performance compared with state-of-the-art techniques.     

A generative probability model of joint label fusion for multi-atlas based brain segmentation

Automated labeling of anatomical structures in medical images is very important in many neuroscience studies. Recently, patch-based labeling has been widely investigated to alleviate the possible mis-alignment when registering atlases to the target image. However, the weights used for label fusion from the registered atlases are generally computed independently and thus lack the capability of preventing the ambiguous atlas patches from contributing to the label fusion. More critically, these weights are often calculated based only on the simple patch similarity, thus not necessarily providing optimal solution for label fusion. To address these limitations, we propose a generative probability model to describe the procedure of label fusion in a multi-atlas scenario, for the goal of labeling each point in the target image by the best representative atlas patches that also have the largest labeling unanimity in labeling the underlying point correctly. Specifically, sparsity constraint is imposed upon label fusion weights, in order to select a small number of atlas patches that best represent the underlying target patch, thus reducing the risks of including the misleading atlas patches. The labeling unanimity among atlas patches is achieved by exploring their dependencies, where we model these dependencies as the joint probability of each pair of atlas patches in correctly predicting the labels, by analyzing the correlation of their morphological error patterns and also the labeling consensus among atlases. The patch dependencies will be further recursively updated based on the latest labeling results to correct the possible labeling errors, which falls to the Expectation Maximization (EM) framework. To demonstrate the labeling performance, we have comprehensively evaluated our patch-based labeling method on the whole brain parcellation and hippocampus segmentation. Promising labeling results have been achieved with comparison to the conventional patch-based labeling method, indicating the potential application of the proposed method in the future clinical studies.     

Expert-driven label fusion in multi-atlas-based segmentation of the prostate using weighted atlases

Purpose

Automated segmentation is required for radiotherapy treatment planning, and multi-atlas methods are frequently used for this purpose. The combination of multiple intermediate results from multi-atlas segmentation into a single segmentation map can be achieved by label fusion. A method that includes expert knowledge in the label fusion phase of multi-atlas-based segmentation was developed. The method was tested by application to prostate segmentation, and the accuracy was compared to standard techniques.

Methods

The selective and iterative method for performance level estimation (SIMPLE) algorithm for label fusion was modified with a weight map given by an expert that indicates the importance of each region in the evaluation of segmentation results. Voxel-based weights specified by an expert when performing the label fusion step in atlas-based segmentation were introduced into the modified SIMPLE algorithm. These weights incorporate expert knowledge on accuracy requirements in different regions of a segmentation. Using this knowledge, segmentation accuracy in regions known to be important can be improved by sacrificing segmentation accuracy in less important regions. Contextual information such as the presence of vulnerable tissue is then used in the segmentation process. This method using weight maps to fine-tune the result of multi-atlas-based segmentation was tested using a set of 146 atlas images consisting of an MR image of the lower abdomen and a prostate segmentation. Each image served as a target in a set of leave-one-out experiments. These experiments were repeated for a weight map derived from the clinical practice in our hospital.

Results

The segmentation accuracy increased 6 % in regions that border vulnerable tissue using expert-specified voxel-based weight maps. This was achieved at the cost of a 4 % decrease in accuracy in less clinically relevant regions.

Conclusion

The inclusion of expert knowledge in a multi-atlas-based segmentation procedure was shown to be feasible for prostate segmentation. This method allows an expert to ensure that automatic segmentation is most accurate in critical regions. This improved local accuracy can increase the practical value of automatic segmentation.     

Mutual consistency learning for semi-supervised medical image segmentation

In this paper, we propose a novel mutual consistency network (MC-Net+) to effectively exploit the unlabeled data for semi-supervised medical image segmentation. The MC-Net+ model is motivated by the observation that deep models trained with limited annotations are prone to output highly uncertain and easily mis-classified predictions in the ambiguous regions (e.g., adhesive edges or thin branches) for medical image segmentation. Leveraging these challenging samples can make the semi-supervised segmentation model training more effective. Therefore, our proposed MC-Net+ model consists of two new designs. First, the model contains one shared encoder and multiple slightly different decoders (i.e., using different up-sampling strategies). The statistical discrepancy of multiple decoders’ outputs is computed to denote the model’s uncertainty, which indicates the unlabeled hard regions. Second, we apply a novel mutual consistency constraint between one decoder’s probability output and other decoders’ soft pseudo labels. In this way, we minimize the discrepancy of multiple outputs (i.e., the model uncertainty) during training and force the model to generate invariant results in such challenging regions, aiming at regularizing the model training. We compared the segmentation results of our MC-Net+ model with five state-of-the-art semi-supervised approaches on three public medical datasets. Extension experiments with two standard semi-supervised settings demonstrate the superior performance of our model over other methods, which sets a new state of the art for semi-supervised medical image segmentation. Our code is released publicly at https://github.com/ycwu1997/MC-Net.