Unsupervised Bidirectional Cross-Modality Adaptation via Deeply Synergistic Image and Feature Alignment for Medical Image Segmentation

Learning generalizable visual representation via adaptive spectral random convolution for medical image segmentation

Medical image segmentation models often fail to generalize well when applied to new datasets, hindering their usage in clinical practice. Existing random-convolution-based domain generalization approaches, which involve randomizing the convolutional kernel weights in the initial layers of CNN models, have shown promise in improving model generalizability. Nevertheless, the indiscriminate introduction of high-frequency noise during early feature extraction may pollute the critical fine details and degrade the model's performance on new datasets. To mitigate this problem, we propose an adaptive spectral random convolution (ASRConv) module designed to selectively randomize low-frequency features while avoiding the introduction of high-frequency artifacts. Unlike prior arts, ASRConv dynamically generates convolution kernel weights, enabling more effective control over feature frequencies than randomized kernels. Specifically, ASRConv achieves this selective randomization through a novel weight generation module conditioned on random noise inputs. The adversarial domain augmentation strategy guides the weight generation module in adaptively suppressing high-frequency noise during training, allowing ASRConv to improve feature diversity and reduce overfitting to specific domains. Extensive experimental results show that our proposed ASRConv method consistently outperforms the state-of-the-art methods, with average DSC improvements of 3.07% and 1.18% on fundus and polyp datasets, respectively. We also qualitatively demonstrate the robustness of our model against domain distribution shifts. All these results validate the effectiveness of the proposed ASRConv in learning domain-invariant representations for robust medical image segmentation.

Generative appearance replay for continual unsupervised domain adaptation

Deep learning models can achieve high accuracy when trained on large amounts of labeled data. However, real-world scenarios often involve several challenges: Training data may become available in installments, may originate from multiple different domains, and may not contain labels for training. Certain settings, for instance medical applications, often involve further restrictions that prohibit retention of previously seen data due to privacy regulations. In this work, to address such challenges, we study unsupervised segmentation in continual learning scenarios that involve domain shift. To that end, we introduce GarDA (Generative Appearance Replay for continual Domain Adaptation), a generative-replay based approach that can adapt a segmentation model sequentially to new domains with unlabeled data. In contrast to single-step unsupervised domain adaptation (UDA), continual adaptation to a sequence of domains enables leveraging and consolidation of information from multiple domains. Unlike previous approaches in incremental UDA, our method does not require access to previously seen data, making it applicable in many practical scenarios. We evaluate GarDA on three datasets with different organs and modalities, where it substantially outperforms existing techniques. Our code is available at: https://github.com/histocartography/generative-appearance-replay.

Unsupervised deep consistency learning adaptation network for cardiac cross-modality structural segmentation

Deep neural networks have recently been succeessful in the field of medical image segmentation; however, they are typically subject to performance degradation problems when well-trained models are tested in another new domain with different data distributions. Given that annotated cross-domain images may inaccessible, unsupervised domain adaptation methods that transfer learnable information from annotated source domains to unannotated target domains with different distributions have attracted substantial attention. Many methods leverage image-level or pixel-level translation networks to align domain-invariant information and mitigate domain shift issues. However, These methods rarely perform well when there is a large domain gap. A new unsupervised deep consistency learning adaptation network, which adopts input space consistency learning and output space consistency learning to realize unsupervised domain adaptation and cardiac structural segmentation, is introduced in this paper The framework mainly includes a domain translation path and a cross-modality segmentation path. In domain translation path, a symmetric alignment generator network with attention to cross-modality features and anatomy is introduced to align bidirectional domain features. In the segmentation path, entropy map minimization, output probability map minimization and segmentation prediction minimization are leveraged to align the output space features. The model conducts supervised learning to extract source domain features and conducts unsupervised deep consistency learning to extract target domain features. Through experimental testing on two challenging cross-modality segmentation tasks, our method has robust performance compared to that of previous methods. Furthermore, ablation experiments are conducted to confirm the effectiveness of our framework.

EDRL: Entropy-guided disentangled representation learning for unsupervised domain adaptation in semantic segmentation

BACKGROUND AND OBJECTIVE

Deep learning-based approaches are excellent at learning from large amounts of data, but can be poor at generalizing the learned knowledge to testing datasets with domain shift, i.e., when there exists distribution discrepancy between the training dataset (source domain) and the testing dataset (target domain). In this paper, we investigate unsupervised domain adaptation (UDA) techniques to train a cross-domain segmentation method which is robust to domain shift, eliminating the requirement of any annotations on the target domain.

METHODS

To this end, we propose an Entropy-guided Disentangled Representation Learning, referred as EDRL, for UDA in semantic segmentation. Concretely, we synergistically integrate image alignment via disentangled representation learning with feature alignment via entropy-based adversarial learning into one network, which is trained end-to-end. We additionally introduce a dynamic feature selection mechanism via soft gating, which helps to further enhance the task-specific feature alignment. We validate the proposed method on two publicly available datasets: the CT-MR dataset and the multi-sequence cardiac MR (MS-CMR) dataset.

RESULTS

On both datasets, our method achieved better results than the state-of-the-art (SOTA) methods. Specifically, on the CT-MR dataset, our method achieved an average DSC of 84.8% when taking CT as the source domain and MR as the target domain, and an average DSC of 84.0% when taking MR as the source domain and CT as the target domain.

CONCLUSIONS

Results from comprehensive experiments demonstrate the efficacy of the proposed EDRL model for cross-domain medical image segmentation.

A modality-collaborative convolution and transformer hybrid network for unpaired multi-modal medical image segmentation with limited annotations

BACKGROUND

Multi-modal learning is widely adopted to learn the latent complementary information between different modalities in multi-modal medical image segmentation tasks. Nevertheless, the traditional multi-modal learning methods require spatially well-aligned and paired multi-modal images for supervised training, which cannot leverage unpaired multi-modal images with spatial misalignment and modality discrepancy. For training accurate multi-modal segmentation networks using easily accessible and low-cost unpaired multi-modal images in clinical practice, unpaired multi-modal learning has received comprehensive attention recently.

PURPOSE

Existing unpaired multi-modal learning methods usually focus on the intensity distribution gap but ignore the scale variation problem between different modalities. Besides, within existing methods, shared convolutional kernels are frequently employed to capture common patterns in all modalities, but they are typically inefficient at learning global contextual information. On the other hand, existing methods highly rely on a large number of labeled unpaired multi-modal scans for training, which ignores the practical scenario when labeled data is limited. To solve the above problems, we propose a modality-collaborative convolution and transformer hybrid network (MCTHNet) using semi-supervised learning for unpaired multi-modal segmentation with limited annotations, which not only collaboratively learns modality-specific and modality-invariant representations, but also could automatically leverage extensive unlabeled scans for improving performance.

METHODS

We make three main contributions to the proposed method. First, to alleviate the intensity distribution gap and scale variation problems across modalities, we develop a modality-specific scale-aware convolution (MSSC) module that can adaptively adjust the receptive field sizes and feature normalization parameters according to the input. Secondly, we propose a modality-invariant vision transformer (MIViT) module as the shared bottleneck layer for all modalities, which implicitly incorporates convolution-like local operations with the global processing of transformers for learning generalizable modality-invariant representations. Third, we design a multi-modal cross pseudo supervision (MCPS) method for semi-supervised learning, which enforces the consistency between the pseudo segmentation maps generated by two perturbed networks to acquire abundant annotation information from unlabeled unpaired multi-modal scans.

RESULTS

Extensive experiments are performed on two unpaired CT and MR segmentation datasets, including a cardiac substructure dataset derived from the MMWHS-2017 dataset and an abdominal multi-organ dataset consisting of the BTCV and CHAOS datasets. Experiment results show that our proposed method significantly outperforms other existing state-of-the-art methods under various labeling ratios, and achieves a comparable segmentation performance close to single-modal methods with fully labeled data by only leveraging a small portion of labeled data. Specifically, when the labeling ratio is 25%, our proposed method achieves overall mean DSC values of 78.56% and 76.18% in cardiac and abdominal segmentation, respectively, which significantly improves the average DSC value of two tasks by 12.84% compared to single-modal U-Net models.

CONCLUSIONS

Our proposed method is beneficial for reducing the annotation burden of unpaired multi-modal medical images in clinical applications.

3D Domain Adaptive Instance Segmentation via Cyclic Segmentation GANs

3D instance segmentation for unlabeled imaging modalities is a challenging but essential task as collecting expert annotation can be expensive and time-consuming. Existing works segment a new modality by either deploying pre-trained models optimized on diverse training data or sequentially conducting image translation and segmentation with two relatively independent networks. In this work, we propose a novel Cyclic Segmentation Generative Adversarial Network (CySGAN) that conducts image translation and instance segmentation simultaneously using a unified network with weight sharing. Since the image translation layer can be removed at inference time, our proposed model does not introduce additional computational cost upon a standard segmentation model. For optimizing CySGAN, besides the CycleGAN losses for image translation and supervised losses for the annotated source domain, we also utilize self-supervised and segmentation-based adversarial objectives to enhance the model performance by leveraging unlabeled target domain images. We benchmark our approach on the task of 3D neuronal nuclei segmentation with annotated electron microscopy (EM) images and unlabeled expansion microscopy (ExM) data. The proposed CySGAN outperforms pre-trained generalist models, feature-level domain adaptation models, and the baselines that conduct image translation and segmentation sequentially. Our implementation and the newly collected, densely annotated ExM zebrafish brain nuclei dataset, named NucExM, are publicly available at https://connectomics-bazaar.github.io/proj/CySGAN/index.html.

Deep-Learning-Based Metal Artefact Reduction With Unsupervised Domain Adaptation Regularization for Practical CT Images

CT metal artefact reduction (MAR) methods based on supervised deep learning are often troubled by domain gap between simulated training dataset and real-application dataset, i.e., methods trained on simulation cannot generalize well to practical data. Unsupervised MAR methods can be trained directly on practical data, but they learn MAR with indirect metrics and often perform unsatisfactorily. To tackle the domain gap problem, we propose a novel MAR method called UDAMAR based on unsupervised domain adaptation (UDA). Specifically, we introduce a UDA regularization loss into a typical image-domain supervised MAR method, which mitigates the domain discrepancy between simulated and practical artefacts by feature-space alignment. Our adversarial-based UDA focuses on a low-level feature space where the domain difference of metal artefacts mainly lies. UDAMAR can simultaneously learn MAR from simulated data with known labels and extract critical information from unlabeled practical data. Experiments on both clinical dental and torso datasets show the superiority of UDAMAR by outperforming its supervised backbone and two state-of-the-art unsupervised methods. We carefully analyze UDAMAR by both experiments on simulated metal artefacts and various ablation studies. On simulation, its close performance to the supervised methods and advantages over the unsupervised methods justify its efficacy. Ablation studies on the influence from the weight of UDA regularization loss, UDA feature layers, and the amount of practical data used for training further demonstrate the robustness of UDAMAR. UDAMAR provides a simple and clean design and is easy to implement. These advantages make it a very feasible solution for practical CT MAR.

A novel one-to-multiple unsupervised domain adaptation framework for abdominal organ segmentation

Abdominal multi-organ segmentation in multi-sequence magnetic resonance images (MRI) is of great significance in many clinical scenarios, e.g., MRI-oriented pre-operative treatment planning. Labeling multiple organs on a single MR sequence is a time-consuming and labor-intensive task, let alone manual labeling on multiple MR sequences. Training a model by one sequence and generalizing it to other domains is one way to reduce the burden of manual annotation, but the existence of domain gap often leads to poor generalization performance of such methods. Image translation-based unsupervised domain adaptation (UDA) is a common way to address this domain gap issue. However, existing methods focus less on keeping anatomical consistency and are limited by one-to-one domain adaptation, leading to low efficiency for adapting a model to multiple target domains. This work proposes a unified framework called OMUDA for one-to-multiple unsupervised domain-adaptive segmentation, where disentanglement between content and style is used to efficiently translate a source domain image into multiple target domains. Moreover, generator refactoring and style constraint are conducted in OMUDA for better maintaining cross-modality structural consistency and reducing domain aliasing. The average Dice Similarity Coefficients (DSCs) of OMUDA for multiple sequences and organs on the in-house test set, the AMOS22 dataset and the CHAOS dataset are 85.51%, 82.66% and 91.38%, respectively, which are slightly lower than those of CycleGAN(85.66% and 83.40%) in the first two data sets and slightly higher than CycleGAN(91.36%) in the last dataset. But compared with CycleGAN, OMUDA reduces floating-point calculations by about 87 percent in the training phase and about 30 percent in the inference stage respectively. The quantitative results in both segmentation performance and training efficiency demonstrate the usability of OMUDA in some practical scenes, such as the initial phase of product development.

Segmentation of Portal Vein in Multiphase CTA Image Based on Unsupervised Domain Transfer and Pseudo Label

BACKGROUND

Clinically, physicians diagnose portal vein diseases on abdominal CT angiography (CTA) images scanned in the hepatic arterial phase (H-phase), portal vein phase (P-phase) and equilibrium phase (E-phase) simultaneously. However, existing studies typically segment the portal vein on P-phase images without considering other phase images.

METHOD

We propose a method for segmenting portal veins on multiphase images based on unsupervised domain transfer and pseudo labels by using annotated P-phase images. Firstly, unsupervised domain transfer is performed to make the H-phase and E-phase images of the same patient approach the P-phase image in style, reducing the image differences caused by contrast media. Secondly, the H-phase (or E-phase) image and its style transferred image are input into the segmentation module together with the P-phase image. Under the constraints of pseudo labels, accurate prediction results are obtained.

RESULTS

This method was evaluated on the multiphase CTA images of 169 patients. The portal vein segmented from the H-phase and E-phase images achieved DSC values of 0.76 and 0.86 and Jaccard values of 0.61 and 0.76, respectively.

CONCLUSION

The method can automatically segment the portal vein on H-phase and E-phase images when only the portal vein on the P-phase CTA image is annotated, which greatly assists in clinical diagnosis.

Rethinking Disentanglement in Unsupervised Domain Adaptation for Medical Image Segmentation

Domain adaptation has become an important topic because the trained neural networks from the source domain generally perform poorly in the target domain due to domain shifts, especially for medical image analysis. Previous DA methods mainly focus on disentangling domain features. However, it is based on feature independence, which often can not be guaranteed in reality. In this work, we present a new DA approach called Dimension-based Disentangled Dilated Domain Adaptation (D4A) to disentangle the storage locations between the features to tackle the problem of domain shift for medical image segmentation tasks without the annotations of the target domain. We use Adaptive Instance Normalization (AdaIN) to encourage the content information to be stored in the spatial dimension, and the style information to be stored in the channel dimension. In addition, we apply dilated convolution to preserve anatomical information avoiding the loss of information due to downsampling. We validate the proposed method for cross-modality medical image segmentation tasks on two public datasets, and the comparison experiments and ablation studies demonstrate the effectiveness of our method, which outperforms the state-of-the-art methods.

Mind the Gap: Alleviating Local Imbalance for Unsupervised Cross-Modality Medical Image Segmentation

Unsupervised cross-modality medical image adaptation aims to alleviate the severe domain gap between different imaging modalities without using the target domain label. A key in this campaign relies upon aligning the distributions of source and target domain. One common attempt is to enforce the global alignment between two domains, which, however, ignores the fatal local-imbalance domain gap problem, i.e., some local features with larger domain gap are harder to transfer. Recently, some methods conduct alignment focusing on local regions to improve the efficiency of model learning. While this operation may cause a deficiency of critical information from contexts. To tackle this limitation, we propose a novel strategy to alleviate the domain gap imbalance considering the characteristics of medical images, namely Global-Local Union Alignment. Specifically, a feature-disentanglement style-transfer module first synthesizes the target-like source images to reduce the global domain gap. Then, a local feature mask is integrated to reduce the 'inter-gap' for local features by prioritizing those discriminative features with larger domain gap. This combination of global and local alignment can precisely localize the crucial regions in segmentation target while preserving the overall semantic consistency. We conduct a series of experiments with two cross-modality adaptation tasks, i,e. cardiac substructure and abdominal multi-organ segmentation. Experimental results indicate that our method achieves state-of-the-art performance in both tasks.

Single slice thigh CT muscle group segmentation with domain adaptation and self-training

Purpose

Thigh muscle group segmentation is important for assessing muscle anatomy, metabolic disease, and aging. Many efforts have been put into quantifying muscle tissues with magnetic resonance (MR) imaging, including manual annotation of individual muscles. However, leveraging publicly available annotations in MR images to achieve muscle group segmentation on single-slice computed tomography (CT) thigh images is challenging.

Approach

We propose an unsupervised domain adaptation pipeline with self-training to transfer labels from three-dimensional MR to single CT slices. First, we transform the image appearance from MR to CT with CycleGAN and feed the synthesized CT images to a segmenter simultaneously. Single CT slices are divided into hard and easy cohorts based on the entropy of pseudo-labels predicted by the segmenter. After refining easy cohort pseudo-labels based on anatomical assumption, self-training with easy and hard splits is applied to fine-tune the segmenter.

Results

On 152 withheld single CT thigh images, the proposed pipeline achieved a mean Dice of 0.888 (0.041) across all muscle groups, including gracilis, hamstrings, quadriceps femoris, and sartorius muscle.

Conclusions

To our best knowledge, this is the first pipeline to achieve domain adaptation from MR to CT for thigh images. The proposed pipeline effectively and robustly extracts muscle groups on two-dimensional single-slice CT thigh images. The container is available for public use in GitHub repository available at: https://github.com/MASILab/DA_CT_muscle_seg.

Between Generating Noise and Generating Images: Noise in the Correct Frequency Improves the Quality of Synthetic Histopathology Images for Digital Pathology

Artificial intelligence and machine learning techniques have the promise to revolutionize the field of digital pathology. However, these models demand considerable amounts of data, while the availability of unbiased training data is limited. Synthetic images can augment existing datasets, to improve and validate AI algorithms. Yet, controlling the exact distribution of cellular features within them is still challenging. One of the solutions is harnessing conditional generative adversarial networks that take a semantic mask as an input rather than a random noise. Unlike other domains, outlining the exact cellular structure of tissues is hard, and most of the input masks depict regions of cell types. This is also the case for non-small cell lung cancer, the most common type of lung cancer. Deciding whether a patient would receive immunotherapy depends on quantifying regions of stained cells. However, using polygon-based masks introduce inherent artifacts within the synthetic images - due to the mismatch between the polygon size and the single-cell size. In this work, we show that introducing random single-pixel noise with the appropriate spatial frequency into a polygon semantic mask can dramatically improve the quality of the synthetic images. We used our platform to generate synthetic images of immunohistochemistry-treated lung biopsies. We test the quality of the images using a three-fold validation procedure. First, we show that adding the appropriate noise frequency yields 87% of the similarity metrics improvement that is obtained by adding the actual single-cell features. Second, we show that the synthetic images pass the Turing test. Finally, we show that adding these synthetic images to the train set improves AI performance in terms of PD-L1 semantic segmentation performances. Our work suggests a simple and powerful approach for generating synthetic data on demand to unbias limited datasets to improve the algorithms' accuracy and validate their robustness.

Multiple sclerosis lesion segmentation: revisiting weighting mechanisms for federated learning

Background and introduction

Federated learning (FL) has been widely employed for medical image analysis to facilitate multi-client collaborative learning without sharing raw data. Despite great success, FL's applications remain suboptimal in neuroimage analysis tasks such as lesion segmentation in multiple sclerosis (MS), due to variance in lesion characteristics imparted by different scanners and acquisition parameters.

Methods

In this work, we propose the first FL MS lesion segmentation framework via two effective re-weighting mechanisms. Specifically, a learnable weight is assigned to each local node during the aggregation process, based on its segmentation performance. In addition, the segmentation loss function in each client is also re-weighted according to the lesion volume for the data during training.

Results

The proposed method has been validated on two FL MS segmentation scenarios using public and clinical datasets. Specifically, the case-wise and voxel-wise Dice score of the proposed method under the first public dataset is 65.20 and 74.30, respectively. On the second in-house dataset, the case-wise and voxel-wise Dice score is 53.66, and 62.31, respectively.

Discussions and conclusions

The Comparison experiments on two FL MS segmentation scenarios using public and clinical datasets have demonstrated the effectiveness of the proposed method by significantly outperforming other FL methods. Furthermore, the segmentation performance of FL incorporating our proposed aggregation mechanism can achieve comparable performance to that from centralized training with all the raw data.

Test-time bi-directional adaptation between image and model for robust segmentation

BACKGROUND AND OBJECTIVE

Deep learning models often suffer from performance degradations when deployed in real clinical environments due to appearance shifts between training and testing images. Most extant methods use training-time adaptation, which almost require target domain samples in the training phase. However, these solutions are limited by the training process and cannot guarantee the accurate prediction of test samples with unforeseen appearance shifts. Further, it is impractical to collect target samples in advance. In this paper, we provide a general method of making existing segmentation models robust to samples with unknown appearance shifts when deployed in daily clinical practice.

METHODS

Our proposed test-time bi-directional adaptation framework combines two complementary strategies. First, our image-to-model (I2M) adaptation strategy adapts appearance-agnostic test images to the learned segmentation model using a novel plug-and-play statistical alignment style transfer module during testing. Second, our model-to-image (M2I) adaptation strategy adapts the learned segmentation model to test images with unknown appearance shifts. This strategy applies an augmented self-supervised learning module to fine-tune the learned model with proxy labels that it generates. This innovative procedure can be adaptively constrained using our novel proxy consistency criterion. This complementary I2M and M2I framework demonstrably achieves robust segmentation against unknown appearance shifts using existing deep-learning models.

RESULTS

Extensive experiments on 10 datasets containing fetal ultrasound, chest X-ray, and retinal fundus images demonstrate that our proposed method achieves promising robustness and efficiency in segmenting images with unknown appearance shifts.

CONCLUSIONS

To address the appearance shift problem in clinically acquired medical images, we provide robust segmentation by using two complementary strategies. Our solution is general and amenable for deployment in clinical settings.

Unpaired, unsupervised domain adaptation assumes your domains are already similar

Unsupervised domain adaptation is a popular method in medical image analysis, but it can be tricky to make it work: without labels to link the domains, domains must be matched using feature distributions. If there is no additional information, this often leaves a choice between multiple possibilities to map the data that may be equally likely but not equally correct. In this paper we explore the fundamental problems that may arise in unsupervised domain adaptation, and discuss conditions that might still make it work. Focusing on medical image analysis, we argue that images from different domains may have similar class balance, similar intensities, similar spatial structure, or similar textures. We demonstrate how these implicit conditions can affect domain adaptation performance in experiments with synthetic data, MNIST digits, and medical images. We observe that practical success of unsupervised domain adaptation relies on existing similarities in the data, and is anything but guaranteed in the general case. Understanding these implicit assumptions is a key step in identifying potential problems in domain adaptation and improving the reliability of the results.

MouseGAN++: Unsupervised Disentanglement and Contrastive Representation for Multiple MRI Modalities Synthesis and Structural Segmentation of Mouse Brain

Segmenting the fine structure of the mouse brain on magnetic resonance (MR) images is critical for delineating morphological regions, analyzing brain function, and understanding their relationships. Compared to a single MRI modality, multimodal MRI data provide complementary tissue features that can be exploited by deep learning models, resulting in better segmentation results. However, multimodal mouse brain MRI data is often lacking, making automatic segmentation of mouse brain fine structure a very challenging task. To address this issue, it is necessary to fuse multimodal MRI data to produce distinguished contrasts in different brain structures. Hence, we propose a novel disentangled and contrastive GAN-based framework, named MouseGAN++, to synthesize multiple MR modalities from single ones in a structure-preserving manner, thus improving the segmentation performance by imputing missing modalities and multi-modality fusion. Our results demonstrate that the translation performance of our method outperforms the state-of-the-art methods. Using the subsequently learned modality-invariant information as well as the modality-translated images, MouseGAN++ can segment fine brain structures with averaged dice coefficients of 90.0% (T2w) and 87.9% (T1w), respectively, achieving around +10% performance improvement compared to the state-of-the-art algorithms. Our results demonstrate that MouseGAN++, as a simultaneous image synthesis and segmentation method, can be used to fuse cross-modality information in an unpaired manner and yield more robust performance in the absence of multimodal data. We release our method as a mouse brain structural segmentation tool for free academic usage at https://github.com/yu02019.

Reciprocal assistance of intravascular imaging in three-dimensional stent reconstruction: Using cross-modal translation based on disentanglement representation

BACKGROUND

Accurate and efficient 3-dimension (3D) reconstruction of coronary stents in intravascular imaging of optical coherence tomography (OCT) or intravascular ultrasound (IVUS) is important for optimization of complex percutaneous coronary interventions (PCI). Deep learning has been used to address this technical challenge. However, manual annotation of stent is strenuous, especially for IVUS images. To this end, we aim to explore whether the OCT and IVUS images can assist each other in stent 3D reconstruction when one of them is lack of labeled dataset.

METHODS

We firstly performed cross-modal translation between OCT and IVUS images, where disentangled representation was employed to generate synthetic images with good stent consistency. The reciprocal assistance of OCT and IVUS in stent 3D reconstruction was then conducted by applying unsupervised and semi-supervised learning with the aid of synthetic images. Stent consistency in synthetic images and reciprocal effectiveness in stent 3D reconstruction were quantitatively assessed by F1-Score (FS) on two datasets: OCT-High Definition IVUS (HD IVUS) and OCT-Conventional IVUS (IVUS).

RESULTS

The employment of disentangled representation achieved higher stent consistency in synthetic images (OCT to HD IVUS: FS=0.789 vs 0.684; HD IVUS to OCT: FS=0.766 vs 0.682; OCT to IVUS: FS=0.806 vs 0.664; IVUS to OCT: FS=0.724 vs 0.673). For stent 3D reconstruction, the assistance from synthetic images significantly promoted unsupervised adaptation across modalities (OCT to HD IVUS: FS=0.776 vs 0.109; HD IVUS to OCT: FS=0.826 vs 0.125; OCT to IVUS: FS=0.782 vs 0.068; IVUS to OCT: FS=0.815 vs 0.123), and improved performance in semi-supervised learning, especially when only limited labeled data was available.

CONCLUSION

The intravascular images of OCT and IVUS can provide reciprocal assistance to each other in stent 3D reconstruction by cross-modal translation, where the stent consistency in synthetic images was maintained by disentangled representation.

Self-training adversarial learning for cross-domain retinal OCT fluid segmentation

Accurate measurements of the size, shape and volume of macular edema can provide important biomarkers to jointly assess disease progression and treatment outcome. Although many deep learning-based segmentation algorithms have achieved remarkable success in semantic segmentation, these methods have difficulty obtaining satisfactory segmentation results in retinal optical coherence tomography (OCT) fluid segmentation tasks due to low contrast, blurred boundaries, and varied distribution. Moreover, directly applying a well-trained model on one device to test the images from other devices may cause the performance degradation in the joint analysis of multi-domain OCT images. In this paper, we propose a self-training adversarial learning framework for unsupervised domain adaptation in retinal OCT fluid segmentation tasks. Specifically, we develop an image style transfer module and a fine-grained feature transfer module to reduce discrepancies in the appearance and high-level features of images from different devices. Importantly, we transfer the target images to the appearance of source images to ensure that no image information of the source domain for supervised training is lost. To capture specific features of the target domain, we design a self-training module based on a discrepancy and similarity strategy to select the images with better segmentation results from the target domain and then introduce them into the source domain for the iterative training segmentation model. Extensive experiments on two challenging datasets demonstrate the effectiveness of our proposed method. In Particular, our proposed method achieves comparable results on cross-domain retinal OCT fluid segmentation compared with the state-of-the-art methods.

LE-UDA: Label-Efficient Unsupervised Domain Adaptation for Medical Image Segmentation

While deep learning methods hitherto have achieved considerable success in medical image segmentation, they are still hampered by two limitations: (i) reliance on large-scale well-labeled datasets, which are difficult to curate due to the expert-driven and time-consuming nature of pixel-level annotations in clinical practices, and (ii) failure to generalize from one domain to another, especially when the target domain is a different modality with severe domain shifts. Recent unsupervised domain adaptation (UDA) techniques leverage abundant labeled source data together with unlabeled target data to reduce the domain gap, but these methods degrade significantly with limited source annotations. In this study, we address this underexplored UDA problem, investigating a challenging but valuable realistic scenario, where the source domain not only exhibits domain shift w.r.t. the target domain but also suffers from label scarcity. In this regard, we propose a novel and generic framework called "Label-Efficient Unsupervised Domain Adaptation" (LE-UDA). In LE-UDA, we construct self-ensembling consistency for knowledge transfer between both domains, as well as a self-ensembling adversarial learning module to achieve better feature alignment for UDA. To assess the effectiveness of our method, we conduct extensive experiments on two different tasks for cross-modality segmentation between MRI and CT images. Experimental results demonstrate that the proposed LE-UDA can efficiently leverage limited source labels to improve cross-domain segmentation performance, outperforming state-of-the-art UDA approaches in the literature.

Deep learning based unpaired image-to-image translation applications for medical physics: a systematic review

Purpose. There is a growing number of publications on the application of unpaired image-to-image (I2I) translation in medical imaging. However, a systematic review covering the current state of this topic for medical physicists is lacking. The aim of this article is to provide a comprehensive review of current challenges and opportunities for medical physicists and engineers to apply I2I translation in practice.Methods and materials. The PubMed electronic database was searched using terms referring to unpaired (unsupervised), I2I translation, and medical imaging. This review has been reported in compliance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement. From each full-text article, we extracted information extracted regarding technical and clinical applications of methods, Transparent Reporting for Individual Prognosis Or Diagnosis (TRIPOD) study type, performance of algorithm and accessibility of source code and pre-trained models.Results. Among 461 unique records, 55 full-text articles were included in the review. The major technical applications described in the selected literature are segmentation (26 studies), unpaired domain adaptation (18 studies), and denoising (8 studies). In terms of clinical applications, unpaired I2I translation has been used for automatic contouring of regions of interest in MRI, CT, x-ray and ultrasound images, fast MRI or low dose CT imaging, CT or MRI only based radiotherapy planning, etc Only 5 studies validated their models using an independent test set and none were externally validated by independent researchers. Finally, 12 articles published their source code and only one study published their pre-trained models.Conclusion. I2I translation of medical images offers a range of valuable applications for medical physicists. However, the scarcity of external validation studies of I2I models and the shortage of publicly available pre-trained models limits the immediate applicability of the proposed methods in practice.

Deep learning for computational cytology: A survey

Computational cytology is a critical, rapid-developing, yet challenging topic in medical image computing concerned with analyzing digitized cytology images by computer-aided technologies for cancer screening. Recently, an increasing number of deep learning (DL) approaches have made significant achievements in medical image analysis, leading to boosting publications of cytological studies. In this article, we survey more than 120 publications of DL-based cytology image analysis to investigate the advanced methods and comprehensive applications. We first introduce various deep learning schemes, including fully supervised, weakly supervised, unsupervised, and transfer learning. Then, we systematically summarize public datasets, evaluation metrics, versatile cytology image analysis applications including cell classification, slide-level cancer screening, nuclei or cell detection and segmentation. Finally, we discuss current challenges and potential research directions of computational cytology.

Cross-Domain Echocardiography Segmentation with Multi-Space Joint Adaptation

The segmentation of the left ventricle endocardium (LVendo) and the left ventricle epicardium (LVepi) in echocardiography plays an important role in clinical diagnosis. Recently, deep neural networks have been the most commonly used approach for echocardiography segmentation. However, the performance of a well-trained segmentation network may degrade in unseen domain datasets due to the distribution shift of the data. Adaptation algorithms can improve the generalization of deep neural networks to different domains. In this paper, we present a multi-space adaptation-segmentation-joint framework, named MACS, for cross-domain echocardiography segmentation. It adopts a generative adversarial architecture; the generator fulfills the segmentation task and the multi-space discriminators align the two domains on both the feature space and output space. We evaluated the MACS method on two echocardiography datasets from different medical centers and vendors, the publicly available CAMUS dataset and our self-acquired dataset. The experimental results indicated that the MACS could handle unseen domain datasets well, without requirements for manual annotations, and improve the generalization performance by 2.2% in the Dice metric.

MITEA: A dataset for machine learning segmentation of the left ventricle in 3D echocardiography using subject-specific labels from cardiac magnetic resonance imaging

Segmentation of the left ventricle (LV) in echocardiography is an important task for the quantification of volume and mass in heart disease. Continuing advances in echocardiography have extended imaging capabilities into the 3D domain, subsequently overcoming the geometric assumptions associated with conventional 2D acquisitions. Nevertheless, the analysis of 3D echocardiography (3DE) poses several challenges associated with limited spatial resolution, poor contrast-to-noise ratio, complex noise characteristics, and image anisotropy. To develop automated methods for 3DE analysis, a sufficiently large, labeled dataset is typically required. However, ground truth segmentations have historically been difficult to obtain due to the high inter-observer variability associated with manual analysis. We address this lack of expert consensus by registering labels derived from higher-resolution subject-specific cardiac magnetic resonance (CMR) images, producing 536 annotated 3DE images from 143 human subjects (10 of which were excluded). This heterogeneous population consists of healthy controls and patients with cardiac disease, across a range of demographics. To demonstrate the utility of such a dataset, a state-of-the-art, self-configuring deep learning network for semantic segmentation was employed for automated 3DE analysis. Using the proposed dataset for training, the network produced measurement biases of -9 ± 16 ml, -1 ± 10 ml, -2 ± 5 %, and 5 ± 23 g, for end-diastolic volume, end-systolic volume, ejection fraction, and mass, respectively, outperforming an expert human observer in terms of accuracy as well as scan-rescan reproducibility. As part of the Cardiac Atlas Project, we present here a large, publicly available 3DE dataset with ground truth labels that leverage the higher resolution and contrast of CMR, to provide a new benchmark for automated 3DE analysis. Such an approach not only reduces the effect of observer-specific bias present in manual 3DE annotations, but also enables the development of analysis techniques which exhibit better agreement with CMR compared to conventional methods. This represents an important step for enabling more efficient and accurate diagnostic and prognostic information to be obtained from echocardiography.

Contrastive domain adaptation with consistency match for automated pneumonia diagnosis

Pneumonia can be difficult to diagnose since its symptoms are too variable, and the radiographic signs are often very similar to those seen in other illnesses such as a cold or influenza. Deep neural networks have shown promising performance in automated pneumonia diagnosis using chest X-ray radiography, allowing mass screening and early intervention to reduce the severe cases and death toll. However, they usually require many well-labelled chest X-ray images for training to achieve high diagnostic accuracy. To reduce the need for training data and annotation resources, we propose a novel method called Contrastive Domain Adaptation with Consistency Match (CDACM). It transfers the knowledge from different but relevant datasets to the unlabelled small-size target dataset and improves the semantic quality of the learnt representations. Specifically, we design a conditional domain adversarial network to exploit discriminative information conveyed in the predictions to mitigate the domain gap between the source and target datasets. Furthermore, due to the small scale of the target dataset, we construct a feature cloud for each target sample and leverage contrastive learning to extract more discriminative features. Lastly, we propose adaptive feature cloud expansion to push the decision boundary to a low-density area. Unlike most existing transfer learning methods that aim only to mitigate the domain gap, our method instead simultaneously considers the domain gap and the data deficiency problem of the target dataset. The conditional domain adaptation and the feature cloud generation of our method are learning jointly to extract discriminative features in an end-to-end manner. Besides, the adaptive feature cloud expansion improves the model's generalisation ability in the target domain. Extensive experiments on pneumonia and COVID-19 diagnosis tasks demonstrate that our method outperforms several state-of-the-art unsupervised domain adaptation approaches, which verifies the effectiveness of CDACM for automated pneumonia diagnosis using chest X-ray imaging.

CrossMoDA 2021 challenge: Benchmark of cross-modality domain adaptation techniques for vestibular schwannoma and cochlea segmentation

Domain Adaptation (DA) has recently been of strong interest in the medical imaging community. While a large variety of DA techniques have been proposed for image segmentation, most of these techniques have been validated either on private datasets or on small publicly available datasets. Moreover, these datasets mostly addressed single-class problems. To tackle these limitations, the Cross-Modality Domain Adaptation (crossMoDA) challenge was organised in conjunction with the 24th International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI 2021). CrossMoDA is the first large and multi-class benchmark for unsupervised cross-modality Domain Adaptation. The goal of the challenge is to segment two key brain structures involved in the follow-up and treatment planning of vestibular schwannoma (VS): the VS and the cochleas. Currently, the diagnosis and surveillance in patients with VS are commonly performed using contrast-enhanced T1 (ceT1) MR imaging. However, there is growing interest in using non-contrast imaging sequences such as high-resolution T2 (hrT2) imaging. For this reason, we established an unsupervised cross-modality segmentation benchmark. The training dataset provides annotated ceT1 scans (N=105) and unpaired non-annotated hrT2 scans (N=105). The aim was to automatically perform unilateral VS and bilateral cochlea segmentation on hrT2 scans as provided in the testing set (N=137). This problem is particularly challenging given the large intensity distribution gap across the modalities and the small volume of the structures. A total of 55 teams from 16 countries submitted predictions to the validation leaderboard. Among them, 16 teams from 9 different countries submitted their algorithm for the evaluation phase. The level of performance reached by the top-performing teams is strikingly high (best median Dice score - VS: 88.4%; Cochleas: 85.7%) and close to full supervision (median Dice score - VS: 92.5%; Cochleas: 87.7%). All top-performing methods made use of an image-to-image translation approach to transform the source-domain images into pseudo-target-domain images. A segmentation network was then trained using these generated images and the manual annotations provided for the source image.

Domain and Content Adaptive Convolution Based Multi-Source Domain Generalization for Medical Image Segmentation

The domain gap caused mainly by variable medical image quality renders a major obstacle on the path between training a segmentation model in the lab and applying the trained model to unseen clinical data. To address this issue, domain generalization methods have been proposed, which however usually use static convolutions and are less flexible. In this paper, we propose a multi-source domain generalization model based on the domain and content adaptive convolution (DCAC) for the segmentation of medical images across different modalities. Specifically, we design the domain adaptive convolution (DAC) module and content adaptive convolution (CAC) module and incorporate both into an encoder-decoder backbone. In the DAC module, a dynamic convolutional head is conditioned on the predicted domain code of the input to make our model adapt to the unseen target domain. In the CAC module, a dynamic convolutional head is conditioned on the global image features to make our model adapt to the test image. We evaluated the DCAC model against the baseline and four state-of-the-art domain generalization methods on the prostate segmentation, COVID-19 lesion segmentation, and optic cup/optic disc segmentation tasks. Our results not only indicate that the proposed DCAC model outperforms all competing methods on each segmentation task but also demonstrate the effectiveness of the DAC and CAC modules. Code is available at https://git.io/DCAC.

DLTTA: Dynamic Learning Rate for Test-Time Adaptation on Cross-Domain Medical Images

Test-time adaptation (TTA) has increasingly been an important topic to efficiently tackle the cross-domain distribution shift at test time for medical images from different institutions. Previous TTA methods have a common limitation of using a fixed learning rate for all the test samples. Such a practice would be sub-optimal for TTA, because test data may arrive sequentially therefore the scale of distribution shift would change frequently. To address this problem, we propose a novel dynamic learning rate adjustment method for test-time adaptation, called DLTTA, which dynamically modulates the amount of weights update for each test image to account for the differences in their distribution shift. Specifically, our DLTTA is equipped with a memory bank based estimation scheme to effectively measure the discrepancy of a given test sample. Based on this estimated discrepancy, a dynamic learning rate adjustment strategy is then developed to achieve a suitable degree of adaptation for each test sample. The effectiveness and general applicability of our DLTTA is extensively demonstrated on three tasks including retinal optical coherence tomography (OCT) segmentation, histopathological image classification, and prostate 3D MRI segmentation. Our method achieves effective and fast test-time adaptation with consistent performance improvement over current state-of-the-art test-time adaptation methods. Code is available at https://github.com/med-air/DLTTA.

Dlung: Unsupervised Few-Shot Diffeomorphic Respiratory Motion Modeling

Lung image registration plays an important role in lung analysis applications, such as respiratory motion modeling. Unsupervised learning-based image registration methods that can compute the deformation without the requirement of supervision attract much attention. However, it is noteworthy that they have two drawbacks: they do not handle the problem of limited data and do not guarantee diffeomorphic (topology-preserving) properties, especially when large deformation exists in lung scans. In this paper, we present an unsupervised few-shot learning-based diffeomorphic lung image registration, namely Dlung. We employ fine-tuning techniques to solve the problem of limited data and apply the scaling and squaring method to accomplish the diffeomorphic registration. Furthermore, atlas-based registration on spatio-temporal (4D) images is performed and thoroughly compared with baseline methods. Dlung achieves the highest accuracy with diffeomorphic properties. It constructs accurate and fast respiratory motion models with limited data. This research extends our knowledge of respiratory motion modeling.

Unsupervised Domain Adaptation with Shape Constraint and Triple Attention for Joint Optic Disc and Cup Segmentation

Currently, glaucoma has become an important cause of blindness. At present, although glaucoma cannot be cured, early treatment can prevent it from getting worse. A reliable way to detect glaucoma is to segment the optic disc and cup and then measure the cup-to-disc ratio (CDR). Many deep neural network models have been developed to autonomously segment the optic disc and the optic cup to help in diagnosis. However, their performance degrades when subjected to domain shift. While many domain-adaptation methods have been exploited to address this problem, they are apt to produce malformed segmentation results. In this study, it is suggested that the segmentation network be adjusted using a constrained formulation that embeds prior knowledge about the shape of the segmentation areas that is domain-invariant. Based on IOSUDA (i.e., Input and Output Space Unsupervised Domain Adaptation), a novel unsupervised joint optic cup-to-disc segmentation framework with shape constraints is proposed, called SCUDA (short for Shape-Constrained Unsupervised Domain Adaptation). A shape constrained loss function is novelly proposed in this paper which utilizes domain-invariant prior knowledge concerning the segmentation region of the joint optic cup-optical disc of fundus images to constrain the segmentation result during network training. In addition, a convolutional triple attention module is designed to improve the segmentation network, which captures cross-dimensional interactions and provides a rich feature representation to improve the segmentation accuracy. Experiments on the RIM-ONE_r3 and Drishti-GS datasets demonstrate that the algorithm outperforms existing approaches for segmenting optic discs and cups.

LMISA: A lightweight multi-modality image segmentation network via domain adaptation using gradient magnitude and shape constraint

In medical image segmentation, supervised machine learning models trained using one image modality (e.g. computed tomography (CT)) are often prone to failure when applied to another image modality (e.g. magnetic resonance imaging (MRI)) even for the same organ. This is due to the significant intensity variations of different image modalities. In this paper, we propose a novel end-to-end deep neural network to achieve multi-modality image segmentation, where image labels of only one modality (source domain) are available for model training and the image labels for the other modality (target domain) are not available. In our method, a multi-resolution locally normalized gradient magnitude approach is firstly applied to images of both domains for minimizing the intensity discrepancy. Subsequently, a dual task encoder-decoder network including image segmentation and reconstruction is utilized to effectively adapt a segmentation network to the unlabeled target domain. Additionally, a shape constraint is imposed by leveraging adversarial learning. Finally, images from the target domain are segmented, as the network learns a consistent latent feature representation with shape awareness from both domains. We implement both 2D and 3D versions of our method, in which we evaluate CT and MRI images for kidney and cardiac tissue segmentation. For kidney, a public CT dataset (KiTS19, MICCAI 2019) and a local MRI dataset were utilized. The cardiac dataset was from the Multi-Modality Whole Heart Segmentation (MMWHS) challenge 2017. Experimental results reveal that our proposed method achieves significantly higher performance with a much lower model complexity in comparison with other state-of-the-art methods. More importantly, our method is also capable of producing superior segmentation results than other methods for images of an unseen target domain without model retraining. The code is available at GitHub (https://github.com/MinaJf/LMISA) to encourage method comparison and further research.

Rethinking adversarial domain adaptation: Orthogonal decomposition for unsupervised domain adaptation in medical image segmentation

Medical image segmentation methods based on deep learning have made remarkable progress. However, such existing methods are sensitive to data distribution. Therefore, slight domain shifts will cause a decline of performance in practical applications. To relieve this problem, many domain adaptation methods learn domain-invariant representations by alignment or adversarial training whereas ignoring domain-specific representations. In response to this issue, this paper rethinks the traditional domain adaptation framework and proposes a novel orthogonal decomposition adversarial domain adaptation (ODADA) architecture for medical image segmentation. The main idea behind our proposed ODADA model is to decompose the input features into domain-invariant and domain-specific representations and then use the newly designed orthogonal loss function to encourage their independence. Furthermore, we propose a two-step optimization strategy to extract domain-invariant representations by separating domain-specific representations, fighting the performance degradation caused by domain shifts. Encouragingly, the proposed ODADA framework is plug-and-play and can replace the traditional adversarial domain adaptation module. The proposed method has consistently demonstrated effectiveness through comprehensive experiments on three publicly available datasets, including cross-site prostate segmentation dataset, cross-site COVID-19 lesion segmentation dataset, and cross-modality cardiac segmentation dataset. The source code is available at https://github.com/YonghengSun1997/ODADA.

Learning coronary artery calcium scoring in coronary CTA from non-contrast CT using unsupervised domain adaptation

Deep learning methods have demonstrated the ability to perform accurate coronary artery calcium (CAC) scoring. However, these methods require large and representative training data hampering applicability to diverse CT scans showing the heart and the coronary arteries. Training methods that accurately score CAC in cross-domain settings remains challenging. To address this, we present an unsupervised domain adaptation method that learns to perform CAC scoring in coronary CT angiography (CCTA) from non-contrast CT (NCCT). To address the domain shift between NCCT (source) domain and CCTA (target) domain, feature distributions are aligned between two domains using adversarial learning. A CAC scoring convolutional neural network is divided into a feature generator that maps input images to features in the latent space and a classifier that estimates predictions from the extracted features. For adversarial learning, a discriminator is used to distinguish the features between source and target domains. Hence, the feature generator aims to extract features with aligned distributions to fool the discriminator. The network is trained with adversarial loss as the objective function and a classification loss on the source domain as a constraint for adversarial learning. In the experiments, three data sets were used. The network is trained with 1,687 labeled chest NCCT scans from the National Lung Screening Trial. Furthermore, 200 labeled cardiac NCCT scans and 200 unlabeled CCTA scans were used to train the generator and the discriminator for unsupervised domain adaptation. Finally, a data set containing 313 manually labeled CCTA scans was used for testing. Directly applying the CAC scoring network trained on NCCT to CCTA led to a sensitivity of 0.41 and an average false positive volume 140 mm³/scan. The proposed method improved the sensitivity to 0.80 and reduced average false positive volume of 20 mm³/scan. The results indicate that the unsupervised domain adaptation approach enables automatic CAC scoring in contrast enhanced CT while learning from a large and diverse set of CT scans without contrast. This may allow for better utilization of existing annotated data sets and extend the applicability of automatic CAC scoring to contrast-enhanced CT scans without the need for additional manual annotations. The code is publicly available at https://github.com/qurAI-amsterdam/CACscoringUsingDomainAdaptation.

Learning From Synthetic CT Images via Test-Time Training for Liver Tumor Segmentation

Automatic liver tumor segmentation could offer assistance to radiologists in liver tumor diagnosis, and its performance has been significantly improved by recent deep learning based methods. These methods rely on large-scale well-annotated training datasets, but collecting such datasets is time-consuming and labor-intensive, which could hinder their performance in practical situations. Learning from synthetic data is an encouraging solution to address this problem. In our task, synthetic tumors can be injected to healthy images to form training pairs. However, directly applying the model trained using the synthetic tumor images on real test images performs poorly due to the domain shift problem. In this paper, we propose a novel approach, namely Synthetic-to-Real Test-Time Training (SR-TTT), to reduce the domain gap between synthetic training images and real test images. Specifically, we add a self-supervised auxiliary task, i.e., two-step reconstruction, which takes the output of the main segmentation task as its input to build an explicit connection between these two tasks. Moreover, we design a scheduled mixture strategy to avoid error accumulation and bias explosion in the training process. During test time, we adapt the segmentation model to each test image with self-supervision from the auxiliary task so as to improve the inference performance. The proposed method is extensively evaluated on two public datasets for liver tumor segmentation. The experimental results demonstrate that our proposed SR-TTT can effectively mitigate the synthetic-to-real domain shift problem in the liver tumor segmentation task, and is superior to existing state-of-the-art approaches.

A bidirectional multilayer contrastive adaptation network with anatomical structure preservation for unpaired cross-modality medical image segmentation

Multi-modal medical image segmentation has achieved great success through supervised deep learning networks. However, because of domain shift and limited annotation information, unpaired cross-modality segmentation tasks are still challenging. The unsupervised domain adaptation (UDA) methods can alleviate the segmentation degradation of cross-modality segmentation by knowledge transfer between different domains, but current methods still suffer from the problems of model collapse, adversarial training instability, and mismatch of anatomical structures. To tackle these issues, we propose a bidirectional multilayer contrastive adaptation network (BMCAN) for unpaired cross-modality segmentation. The shared encoder is first adopted for learning modality-invariant encoding representations in image synthesis and segmentation simultaneously. Secondly, to retain the anatomical structure consistency in cross-modality image synthesis, we present a structure-constrained cross-modality image translation approach for image alignment. Thirdly, we construct a bidirectional multilayer contrastive learning approach to preserve the anatomical structures and enhance encoding representations, which utilizes two groups of domain-specific multilayer perceptron (MLP) networks to learn modality-specific features. Finally, a semantic information adversarial learning approach is designed to learn structural similarities of semantic outputs for output space alignment. Our proposed method was tested on three different cross-modality segmentation tasks: brain tissue, brain tumor, and cardiac substructure segmentation. Compared with other UDA methods, experimental results show that our proposed BMCAN achieves state-of-the-art segmentation performance on the above three tasks, and it has fewer training components and better feature representations for overcoming overfitting and domain shift problems. Our proposed method can efficiently reduce the annotation burden of radiologists in cross-modality image analysis.

Cardiac MRI segmentation with sparse annotations: Ensembling deep learning uncertainty and shape priors

The performance of deep learning for cardiac magnetic resonance imaging (MRI) segmentation is oftentimes degraded when using small datasets and sparse annotations for training or adapting a pre-trained model to previously unseen datasets. Here, we developed and evaluated an approach to addressing some of these issues to facilitate broader use of deep learning for short-axis cardiac MRI segmentation. We developed a globally optimal label fusion (GOLF) algorithm that enforced spatial smoothness to generate consensus segmentation from segmentation predictions provided by a deep learning ensemble algorithm. The GOLF consensus was entered into an uncertainty-guided coupled continuous kernel cut (ugCCKC) algorithm that employed normalized cut, image-grid continuous regularization, and "nesting" and circular shape priors of the left ventricular myocardium (LVM) and cavity (LVC). In addition, the uncertainty measurements derived from the segmentation predictions were used to constrain the similarity of GOLF and final segmentation. We optimized ugCCKC through upper bound relaxation, for which we developed an efficient coupled continuous max-flow algorithm implemented in an iterative manner. We showed that GOLF yielded average symmetric surface distance (ASSD) 0.2-0.8 mm lower than an averaging method with higher or similar Dice similarity coefficient (DSC). We also demonstrated that ugCCKC incorporating the shape priors improved DSC by 0.01-0.05 and reduced ASSD by 0.1-0.9 mm. In addition, we integrated GOLF and ugCCKC into a deep learning ensemble algorithm by refining the segmentation of an unannotated dataset and using the refined segmentation to update the trained models. With the proposed framework, we demonstrated the capability of using relatively small datasets (5-10 subjects) with sparse (5-25% slices labeled) annotations to train a deep learning algorithm, while achieving DSC of 0.871-0.893 for LVM and 0.933-0.959 for LVC on the LVQuan dataset, and these were 0.844-0.871 for LVM and 0.923-0.931 for LVC on the ACDC dataset. Furthermore, we showed that the proposed approach can be adapted to substantially alleviate the domain shift issue. Moreover, we calculated a number of commonly used LV function measurements using the derived segmentation and observed strong correlations (Pearson r=0.77-1.00, p<0.001) between algorithm and manual LV function analyses. These results suggest that the developed approaches can be used to facilitate broader application of deep learning in research and clinical cardiac MR imaging workflow.

A multimodal domain adaptive segmentation framework for IDH genotype prediction

PURPOSE

The gene mutation status of isocitrate dehydrogenase (IDH) in gliomas leads to a different prognosis. It is challenging to perform automated tumor segmentation and genotype prediction directly using label-deprived multimodal magnetic resonance (MR) images. We propose a novel framework that employs a domain adaptive mechanism to address this issue.

METHODS

Multimodal domain adaptive segmentation (MDAS) framework was proposed to solve the gap issue in cross dataset model transfer. Image translation was used to adaptively align the multimodal data from two domains at the image level, and segmentation consistency loss was proposed to retain more pathological information through semantic constraints. The data distribution between the labeled public dataset and label-free target dataset was learned to achieve better unsupervised segmentation results on the target dataset. Then, the segmented tumor foci were used as a mask to extract the radiomics and deep features. And the subsequent prediction of IDH gene mutation status was conducted by training a random forest classifier. The prediction model does not need any expert segmented labels.

RESULTS

We implemented our method on the public BraTS 2019 dataset and 110 astrocytoma cases of grade II-IV brain tumors from our hospital. We obtained a Dice score of 77.41% for unsupervised tumor segmentation, a genotype prediction accuracy (ACC) of 0.7639 and an area under curve (AUC) of 0.8600. Experimental results demonstrate that our domain adaptive approach outperforms the methods utilizing direct transfer learning. The model using hybrid features gives better results than the model using radiomics or deep features alone.

CONCLUSIONS

Domain adaptation enables the segmentation network to achieve better performance, and the extraction of mixed features at multiple levels on the segmented region of interest ensures effective prediction of the IDH gene mutation status.

MSGAN: Multi-Stage Generative Adversarial Networks for Cross-Modality Domain Adaptation

Domain adaptation has become an important topic because the trained neural networks from the source domain generally perform poorly in the target domain due to domain shifts, especially for cross-modality medical images. In this work, we present a new unsupervised domain adaptation approach called Multi-Stage GAN (MSGAN) to tackle the problem of domain shift for CT and MRI segmentation tasks. We adopt the multi-stage strategy in parallel to avoid information loss and transfer rough styles on low-resolution feature maps to the detailed textures on high-resolution feature maps. In detail, the style layers map the learnt style codes from the Gaussian noise to the input features in order to synthesize images with different styles. We validate the proposed method for cross-modality medical image segmentation tasks on two public datasets, and the results demonstrate the effectiveness of our method. Clinical relevance- This technique paves the way to translate cross-modality images (MRI and CT) and it can also mitigate the performance degradation when applying deep neural networks in a cross-domain scenario.

Deep Learning in Neuroimaging: Overcoming Challenges With Emerging Approaches

Deep learning (DL) is of great interest in psychiatry due its potential yet largely untapped ability to utilize multidimensional datasets (such as fMRI data) to predict clinical outcomes. Typical DL methods, however, have strong assumptions, such as large datasets and underlying model opaqueness, that are suitable for natural image prediction problems but not medical imaging. Here we describe three relatively novel DL approaches that may help accelerate its incorporation into mainstream psychiatry research and ultimately bring it into the clinic as a prognostic tool. We first introduce two methods that can reduce the amount of training data required to develop accurate models. These may prove invaluable for fMRI-based DL given the time and monetary expense required to acquire neuroimaging data. These methods are (1) transfer learning - the ability of deep learners to incorporate knowledge learned from one data source (e.g., fMRI data from one site) and apply it toward learning from a second data source (e.g., data from another site), and (2) data augmentation (via Mixup) - a self-supervised learning technique in which "virtual" instances are created. We then discuss explainable artificial intelligence (XAI), i.e., tools that reveal what features (and in what combinations) deep learners use to make decisions. XAI can be used to solve the "black box" criticism common in DL and reveal mechanisms that ultimately produce clinical outcomes. We expect these techniques to greatly enhance the applicability of DL in psychiatric research and help reveal novel mechanisms and potential pathways for therapeutic intervention in mental illness.

A dual meta-learning framework based on idle data for enhancing segmentation of pancreatic cancer

Automated segmentation of pancreatic cancer is vital for clinical diagnosis and treatment. However, the small size and inconspicuous boundaries limit the segmentation performance, which is further exacerbated for deep learning techniques with the few training samples due to the high threshold of image acquisition and annotation. To alleviate this issue caused by the small-scale dataset, we collect idle multi-parametric MRIs of pancreatic cancer from different studies to construct a relatively large dataset for enhancing the CT pancreatic cancer segmentation. Therefore, we propose a deep learning segmentation model with the dual meta-learning framework for pancreatic cancer. It can integrate the common knowledge of tumors obtained from idle MRIs and salient knowledge from CT images, making high-level features more discriminative. Specifically, the random intermediate modalities between MRIs and CT are first generated to smoothly fill in the gaps in visual appearance and provide rich intermediate representations for ensuing meta-learning scheme. Subsequently, we employ intermediate modalities-based model-agnostic meta-learning to capture and transfer commonalities. At last, a meta-optimizer is utilized to adaptively learn the salient features within CT data, thus alleviating the interference due to internal differences. Comprehensive experimental results demonstrated our method achieved the promising segmentation performance, with a max Dice score of 64.94% on our private dataset, and outperformed state-of-the-art methods on a public pancreatic cancer CT dataset. The proposed method is an effective pancreatic cancer segmentation framework, which can be easily integrated into other segmentation networks and thus promises to be a potential paradigm for alleviating data scarcity challenges using idle data.

Domain Adaptation Meets Zero-Shot Learning: An Annotation-Efficient Approach to Multi-Modality Medical Image Segmentation

Due to the lack of properly annotated medical data, exploring the generalization capability of the deep model is becoming a public concern. Zero-shot learning (ZSL) has emerged in recent years to equip the deep model with the ability to recognize unseen classes. However, existing studies mainly focus on natural images, which utilize linguistic models to extract auxiliary information for ZSL. It is impractical to apply the natural image ZSL solutions directly to medical images, since the medical terminology is very domain-specific, and it is not easy to acquire linguistic models for the medical terminology. In this work, we propose a new paradigm of ZSL specifically for medical images utilizing cross-modality information. We make three main contributions with the proposed paradigm. First, we extract the prior knowledge about the segmentation targets, called relation prototypes, from the prior model and then propose a cross-modality adaptation module to inherit the prototypes to the zero-shot model. Second, we propose a relation prototype awareness module to make the zero-shot model aware of information contained in the prototypes. Last but not least, we develop an inheritance attention module to recalibrate the relation prototypes to enhance the inheritance process. The proposed framework is evaluated on two public cross-modality datasets including a cardiac dataset and an abdominal dataset. Extensive experiments show that the proposed framework significantly outperforms the state of the arts.

Unsupervised domain adaptation with hyperbolic graph convolution network for segmentation of X-ray breast mass

Deep learning has been widely used in medical image segmentation, such as breast tumor segmentation, prostate MR image segmentation, and so on. However, the labeling of the data set takes a lot of time. Although the emergence of unsupervised domain adaptation fills the technical gap, the existing domain adaptation methods for breast segmentation do not consider the alignment of the source domain and target domain breast mass structure. This paper proposes a hyperbolic graph convolutional network architecture. First, a hyperbolic graph convolutional network is used to make the source and target domains structurally aligned. Secondly, we adopt a hyperbolic space mapping model that has better expressive ability than Euclidean space in a graph structure. In particular, when constructing the graph structure, we added the completion adjacency matrix, so that the graph structure can be changed after each feature mapping, which can better improve the segmentation accuracy. Extensive comparative and ablation experiments were performed on two common breast datasets(CBIS-DDSM and INbreast). Experiments show that the method in this paper is better than the most advanced model. When CBIS-DDSM and INbreast are used as the source domain, the segmentation accuracy reaches 89.1% and 80.7%.

From Accuracy to Reliability and Robustness in Cardiac Magnetic Resonance Image Segmentation: A Review

Since the rise of deep learning (DL) in the mid-2010s, cardiac magnetic resonance (CMR) image segmentation has achieved state-of-the-art performance. Despite achieving inter-observer variability in terms of different accuracy performance measures, visual inspections reveal errors in most segmentation results, indicating a lack of reliability and robustness of DL segmentation models, which can be critical if a model was to be deployed into clinical practice. In this work, we aim to bring attention to reliability and robustness, two unmet needs of cardiac image segmentation methods, which are hampering their translation into practice. To this end, we first study the performance accuracy evolution of CMR segmentation, illustrate the improvements brought by DL algorithms and highlight the symptoms of performance stagnation. Afterwards, we provide formal definitions of reliability and robustness. Based on the two definitions, we identify the factors that limit the reliability and robustness of state-of-the-art deep learning CMR segmentation techniques. Finally, we give an overview of the current set of works that focus on improving the reliability and robustness of CMR segmentation, and we categorize them into two families of methods: quality control methods and model improvement techniques. The first category corresponds to simpler strategies that only aim to flag situations where a model may be incurring poor reliability or robustness. The second one, instead, directly tackles the problem by bringing improvements into different aspects of the CMR segmentation model development process. We aim to bring the attention of more researchers towards these emerging trends regarding the development of reliable and robust CMR segmentation frameworks, which can guarantee the safe use of DL in clinical routines and studies.