CyCMIS: Cycle-consistent Cross-domain Medical Image Segmentation via diverse image augmentation

Domain shift, a phenomenon when there exists distribution discrepancy between training dataset (source domain) and test dataset (target domain), is very common in practical applications and may cause significant performance degradation, which hinders the effective deployment of deep learning models to clinical settings. Adaptation algorithms to improve the model generalizability from source domain to target domain has significant practical value. In this paper, we investigate unsupervised domain adaptation (UDA) technique to train a cross-domain segmentation method which is robust to domain shift, and which does not require any annotations on the test domain. To this end, we propose Cycle-consistent Cross-domain Medical Image Segmentation, referred as CyCMIS, integrating online diverse image translation via disentangled representation learning and semantic consistency regularization into one network. Different from learning one-to-one mapping, our method characterizes the complex relationship between domains as many-to-many mapping. A novel diverse inter-domain semantic consistency loss is then proposed to regularize the cross-domain segmentation process. We additionally introduce an intra-domain semantic consistency loss to encourage the segmentation consistency between the original input and the image after cross-cycle reconstruction. We conduct comprehensive experiments on two publicly available datasets to evaluate the effectiveness of the proposed method. Results demonstrate the efficacy of the present approach.     

Memory consistent unsupervised off-the-shelf model adaptation for source-relaxed medical image segmentation

Unsupervised domain adaptation (UDA) has been a vital protocol for migrating information learned from a labeled source domain to facilitate the implementation in an unlabeled heterogeneous target domain. Although UDA is typically jointly trained on data from both domains, accessing the labeled source domain data is often restricted, due to concerns over patient data privacy or intellectual property. To sidestep this, we propose “off-the-shelf (OS)” UDA (OSUDA), aimed at image segmentation, by adapting an OS segmentor trained in a source domain to a target domain, in the absence of source domain data in adaptation. Toward this goal, we aim to develop a novel batch-wise normalization (BN) statistics adaptation framework. In particular, we gradually adapt the domain-specific low-order BN statistics, e.g., mean and variance, through an exponential momentum decay strategy, while explicitly enforcing the consistency of the domain shareable high-order BN statistics, e.g., scaling and shifting factors, via our optimization objective. We also adaptively quantify the channel-wise transferability to gauge the importance of each channel, via both low-order statistics divergence and a scaling factor. Furthermore, we incorporate unsupervised self-entropy minimization into our framework to boost performance alongside a novel queued, memory-consistent self-training strategy to utilize the reliable pseudo label for stable and efficient unsupervised adaptation. We evaluated our OSUDA-based framework on both cross-modality and cross-subtype brain tumor segmentation and cardiac MR to CT segmentation tasks. Our experimental results showed that our memory consistent OSUDA performs better than existing source-relaxed UDA methods and yields similar performance to UDA methods with source data.     

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.     

Source free domain adaptation for medical image segmentation with fourier style mining

Unsupervised domain adaptation (UDA) aims to exploit the knowledge learned from a labeled source dataset to solve similar tasks in a new unlabeled target domain. Existing UDA techniques typically assume that samples from source and target domains are freely accessible during the training. However, it may be impractical to access source images due to privacy concerns, especially in medical imaging scenarios with the patient information. To tackle this issue, we devise a novel source free domain adaptation framework with fourier style mining, where only a well-trained source segmentation model is available for the adaptation to the target domain. Our framework is composed of two stages: a generation stage and an adaptation stage. In the generation stage, we design a Fourier Style Mining (FSM) generator to inverse source-like images through statistic information of the pretrained source model and mutual Fourier Transform. These generated source-like images can provide source data distribution and benefit the domain alignment. In the adaptation stage, we design a Contrastive Domain Distillation (CDD) module to achieve feature-level adaptation, including a domain distillation loss to transfer relation knowledge and a domain contrastive loss to narrow down the domain gap by a self-supervised paradigm. Besides, a Compact-Aware Domain Consistency (CADC) module is proposed to enhance consistency learning by filtering out noisy pseudo labels with shape compactness metric, thus achieving output-level adaptation. Extensive experiments on cross-device and cross-centre datasets are conducted for polyp and prostate segmentation, and our method delivers impressive performance compared with state-of-the-art domain adaptation methods. The source code is available at https://github.com/CityU-AIM-Group/SFDA-FSM.     

Attentive continuous generative self-training for unsupervised domain adaptive medical image translation

Self-training is an important class of unsupervised domain adaptation (UDA) approaches that are used to mitigate the problem of domain shift, when applying knowledge learned from a labeled source domain to unlabeled and heterogeneous target domains. While self-training-based UDA has shown considerable promise on discriminative tasks, including classification and segmentation, through reliable pseudo-label filtering based on the maximum softmax probability, there is a paucity of prior work on self-training-based UDA for generative tasks, including image modality translation. To fill this gap, in this work, we seek to develop a generative self-training (GST) framework for domain adaptive image translation with continuous value prediction and regression objectives. Specifically, we quantify both aleatoric and epistemic uncertainties within our GST using variational Bayes learning to measure the reliability of synthesized data. We also introduce a self-attention scheme that de-emphasizes the background region to prevent it from dominating the training process. The adaptation is then carried out by an alternating optimization scheme with target domain supervision that focuses attention on the regions with reliable pseudo-labels. We evaluated our framework on two cross-scanner/center, inter-subject translation tasks, including tagged-to-cine magnetic resonance (MR) image translation and T1-weighted MR-to-fractional anisotropy translation. Extensive validations with unpaired target domain data showed that our GST yielded superior synthesis performance in comparison to adversarial training UDA methods.     

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.     

Uncertainty-aware multi-view co-training for semi-supervised medical image segmentation and domain adaptation

Although having achieved great success in medical image segmentation, deep learning-based approaches usually require large amounts of well-annotated data, which can be extremely expensive in the field of medical image analysis. Unlabeled data, on the other hand, is much easier to acquire. Semi-supervised learning and unsupervised domain adaptation both take the advantage of unlabeled data, and they are closely related to each other. In this paper, we propose uncertainty-aware multi-view co-training (UMCT), a unified framework that addresses these two tasks for volumetric medical image segmentation. Our framework is capable of efficiently utilizing unlabeled data for better performance. We firstly rotate and permute the 3D volumes into multiple views and train a 3D deep network on each view. We then apply co-training by enforcing multi-view consistency on unlabeled data, where an uncertainty estimation of each view is utilized to achieve accurate labeling. Experiments on the NIH pancreas segmentation dataset and a multi-organ segmentation dataset show state-of-the-art performance of the proposed framework on semi-supervised medical image segmentation. Under unsupervised domain adaptation settings, we validate the effectiveness of this work by adapting our multi-organ segmentation model to two pathological organs from the Medical Segmentation Decathlon Datasets. Additionally, we show that our UMCT-DA model can even effectively handle the challenging situation where labeled source data is inaccessible, demonstrating strong potentials for real-world applications.     

HAL-IA: A Hybrid Active Learning framework using Interactive Annotation for medical image segmentation

High performance of deep learning models on medical image segmentation greatly relies on large amount of pixel-wise annotated data, yet annotations are costly to collect. How to obtain high accuracy segmentation labels of medical images with limited cost (e.g. time) becomes an urgent problem. Active learning can reduce the annotation cost of image segmentation, but it faces three challenges: the cold start problem, an effective sample selection strategy for segmentation task and the burden of manual annotation. In this work, we propose a Hybrid Active Learning framework using Interactive Annotation (HAL-IA) for medical image segmentation, which reduces the annotation cost both in decreasing the amount of the annotated images and simplifying the annotation process. Specifically, we propose a novel hybrid sample selection strategy to select the most valuable samples for segmentation model performance improvement. This strategy combines pixel entropy, regional consistency and image diversity to ensure that the selected samples have high uncertainty and diversity. In addition, we propose a warm-start initialization strategy to build the initial annotated dataset to avoid the cold-start problem. To simplify the manual annotation process, we propose an interactive annotation module with suggested superpixels to obtain pixel-wise label with several clicks. We validate our proposed framework with extensive segmentation experiments on four medical image datasets. Experimental results showed that the proposed framework achieves high accuracy pixel-wise annotations and models with less labeled data and fewer interactions, outperforming other state-of-the-art methods. Our method can help physicians efficiently obtain accurate medical image segmentation results for clinical analysis and diagnosis.     

Semi-supervised medical image segmentation via a tripled-uncertainty guided mean teacher model with contrastive learning

Due to the difficulty in accessing a large amount of labeled data, semi-supervised learning is becoming an attractive solution in medical image segmentation. To make use of unlabeled data, current popular semi-supervised methods (e.g., temporal ensembling, mean teacher) mainly impose data-level and model-level consistency on unlabeled data. In this paper, we argue that in addition to these strategies, we could further utilize auxiliary tasks and consider task-level consistency to better excavate effective representations from unlabeled data for segmentation. Specifically, we introduce two auxiliary tasks, i.e., a foreground and background reconstruction task for capturing semantic information and a signed distance field (SDF) prediction task for imposing shape constraint, and explore the mutual promotion effect between the two auxiliary and the segmentation tasks based on mean teacher architecture. Moreover, to handle the potential bias of the teacher model caused by annotation scarcity, we develop a tripled-uncertainty guided framework to encourage the three tasks in the student model to learn more reliable knowledge from the teacher. When calculating uncertainty, we propose an uncertainty weighted integration (UWI) strategy for yielding the segmentation predictions of the teacher. In addition, following the advance of unsupervised learning in leveraging the unlabeled data, we also incorporate a contrastive learning based constraint to help the encoders extract more distinct representations to promote the medical image segmentation performance. Extensive experiments on the public 2017 ACDC dataset and the PROMISE12 dataset have demonstrated the effectiveness of our method.     

Exploring uncertainty measures in deep networks for Multiple sclerosis lesion detection and segmentation

Deep learning networks have recently been shown to outperform other segmentation methods on various public, medical-image challenge datasets, particularly on metrics focused on large pathologies. For diseases such as Multiple Sclerosis (MS), however, monitoring all the focal lesions visible on MRI sequences, even very small ones, is essential for disease staging, prognosis, and evaluating treatment efficacy. Small lesion segmentation presents significant challenges to popular deep learning models. This, coupled with their deterministic predictions, hinders their clinical adoption. Uncertainty estimates for these predictions would permit subsequent revision by clinicians. We present the first exploration of multiple uncertainty estimates based on Monte Carlo (MC) dropout (Gal and Ghahramani, 2016) in the context of deep networks for lesion detection and segmentation in medical images. Specifically, we develop a 3D MS lesion segmentation CNN, augmented to provide four different voxel-based uncertainty measures based on MC dropout. We train the network on a proprietary, large-scale, multi-site, multi-scanner, clinical MS dataset, and compute lesion-wise uncertainties by accumulating evidence from voxel-wise uncertainties within detected lesions. We analyze the performance of voxel-based segmentation and lesion-level detection by choosing operating points based on the uncertainty. Uncertainty filtering improves both voxel and lesion-wise TPR and FDR on remaining, certain predictions compared to sigmoid-based TPR/FDR curves. Small lesions and lesion-boundaries are the most uncertain regions, which is consistent with human-rater variability.     

Suggestive annotation of brain MR images with gradient-guided sampling

Machine learning has been widely adopted for medical image analysis in recent years given its promising performance in image segmentation and classification tasks. The success of machine learning, in particular supervised learning, depends on the availability of manually annotated datasets. For medical imaging applications, such annotated datasets are not easy to acquire, it takes a substantial amount of time and resource to curate an annotated medical image set. In this paper, we propose an efficient annotation framework for brain MR images that can suggest informative sample images for human experts to annotate. We evaluate the framework on two different brain image analysis tasks, namely brain tumour segmentation and whole brain segmentation. Experiments show that for brain tumour segmentation task on the BraTS 2019 dataset, training a segmentation model with only 7% suggestively annotated image samples can achieve a performance comparable to that of training on the full dataset. For whole brain segmentation on the MALC dataset, training with 42% suggestively annotated image samples can achieve a comparable performance to training on the full dataset. The proposed framework demonstrates a promising way to save manual annotation cost and improve data efficiency in medical imaging applications.     

FocusNetv2: Imbalanced large and small organ segmentation with adversarial shape constraint for head and neck CT images

Radiotherapy is a treatment where radiation is used to eliminate cancer cells. The delineation of organs-at-risk (OARs) is a vital step in radiotherapy treatment planning to avoid damage to healthy organs. For nasopharyngeal cancer, more than 20 OARs are needed to be precisely segmented in advance. The challenge of this task lies in complex anatomical structure, low-contrast organ contours, and the extremely imbalanced size between large and small organs. Common segmentation methods that treat them equally would generally lead to inaccurate small-organ labeling. We propose a novel two-stage deep neural network, FocusNetv2, to solve this challenging problem by automatically locating, ROI-pooling, and segmenting small organs with specifically designed small-organ localization and segmentation sub-networks while maintaining the accuracy of large organ segmentation. In addition to our original FocusNet, we employ a novel adversarial shape constraint on small organs to ensure the consistency between estimated small-organ shapes and organ shape prior knowledge. Our proposed framework is extensively tested on both self-collected dataset of 1,164 CT scans and the MICCAI Head and Neck Auto Segmentation Challenge 2015 dataset, which shows superior performance compared with state-of-the-art head and neck OAR segmentation methods.     

Commensal correlation network between segmentation and direct area estimation for bi-ventricle quantification

Accurate and automated cardiac bi-ventricle quantification based on cardiac magnetic resonance (CMR) image is a very crucial procedure for clinical cardiac disease diagnosis. Two traditional and commensal tasks, i.e., bi-ventricle segmentation and direct ventricle function index estimation, are always independently devoting to address ventricle quantification problem. However, because of inherent difficulties from the variable CMR imaging conditions, these two tasks are still open challenging. In this paper, we proposed a unified bi-ventricle quantification framework based on commensal correlation between the bi-ventricle segmentation and direct area estimation. Firstly, we proposed the area commensal correlation between the two traditional cardiac quantification tasks for the first time, and designed a novel deep commensal network (DCN) to join these two commensal tasks into a unified framework based on the proposed commensal correlation loss. Secondly, we proposed an differentiable area operator to model the proposed area commensal correlation and made the proposed model continuously differentiable. Thirdly, we proposed a high-efficiency and novel uncertainty estimation method through one-time inference based on cross-task output variability. And finally DCN achieved end-to-end optimization and fast convergence as well as uncertainty estimation with one-time inference. Experiments on the four open accessible short-axis CMR benchmark datasets (i.e., Sunnybrook, STACOM 2011, RVSC, and ACDC) showed that the proposed method achieves best bi-ventricle quantification accuracy and optimization performance. Hence, the proposed method has big potential to be extended to other medical image analysis tasks and has clinical application value.     

Multi-task learning for the segmentation of organs at risk with label dependence

Automatic segmentation of organs at risk is crucial to aid diagnoses and remains a challenging task in medical image analysis domain. To perform the segmentation, we use multi-task learning (MTL) to accurately determine the contour of organs at risk in CT images. We train an encoder-decoder network for two tasks in parallel. The main task is the segmentation of organs, entailing a pixel-level classification in the CT images, and the auxiliary task is the multi-label classification of organs, entailing an image-level multi-label classification of the CT images. To boost the performance of the multi-label classification, we propose a weighted mean cross entropy loss function for the network training, where the weights are the global conditional probability between two organs. Based on MTL, we optimize the false positive filtering (FPF) algorithm to decrease the number of falsely segmented organ pixels in the CT images. Specifically, we propose a dynamic threshold selection (DTS) strategy to prevent true positive rates from decreasing when using the FPF algorithm. We validate these methods on the public ISBI 2019 segmentation of thoracic organs at risk (SegTHOR) challenge dataset and a private medical organ dataset. The experimental results show that networks using our proposed methods outperform basic encoder-decoder networks without increasing the training time complexity.     

Learning joint segmentation of tissues and brain lesions from task-specific hetero-modal domain-shifted datasets

Brain tissue segmentation from multimodal MRI is a key building block of many neuroimaging analysis pipelines. Established tissue segmentation approaches have, however, not been developed to cope with large anatomical changes resulting from pathology, such as white matter lesions or tumours, and often fail in these cases. In the meantime, with the advent of deep neural networks (DNNs), segmentation of brain lesions has matured significantly. However, few existing approaches allow for the joint segmentation of normal tissue and brain lesions. Developing a DNN for such a joint task is currently hampered by the fact that annotated datasets typically address only one specific task and rely on task-specific imaging protocols including a task-specific set of imaging modalities. In this work, we propose a novel approach to build a joint tissue and lesion segmentation model from aggregated task-specific hetero-modal domain-shifted and partially-annotated datasets. Starting from a variational formulation of the joint problem, we show how the expected risk can be decomposed and optimised empirically. We exploit an upper bound of the risk to deal with heterogeneous imaging modalities across datasets. To deal with potential domain shift, we integrated and tested three conventional techniques based on data augmentation, adversarial learning and pseudo-healthy generation. For each individual task, our joint approach reaches comparable performance to task-specific and fully-supervised models. The proposed framework is assessed on two different types of brain lesions: White matter lesions and gliomas. In the latter case, lacking a joint ground-truth for quantitative assessment purposes, we propose and use a novel clinically-relevant qualitative assessment methodology.     

Semi-supervised body parsing and pose estimation for enhancing infant general movement assessment

General movement assessment (GMA) of infant movement videos (IMVs) is an effective method for early detection of cerebral palsy (CP) in infants. We demonstrate in this paper that end-to-end trainable neural networks for image sequence recognition can be applied to achieve good results in GMA, and more importantly, augmenting raw video with infant body parsing and pose estimation information can significantly improve performance. To solve the problem of efficiently utilizing partially labeled IMVs for body parsing, we propose a semi-supervised model, termed SiamParseNet (SPN), which consists of two branches, one for intra-frame body parts segmentation and another for inter-frame label propagation. During training, the two branches are jointly trained by alternating between using input pairs of only labeled frames and input of both labeled and unlabeled frames. We also investigate training data augmentation by proposing a factorized video generative adversarial network (FVGAN) to synthesize novel labeled frames for training. FVGAN decouples foreground and background generation which allows for generating multiple labeled frames from one real labeled frame. When testing, we employ a multi-source inference mechanism, where the final result for a test frame is either obtained via the segmentation branch or via propagation from a nearby key frame. We conduct extensive experiments for body parsing using SPN on two infant movement video datasets; on these partially labeled IMVs, we show that SPN coupled with FVGAN achieves state-of-the-art performance. We further demonstrate that our proposed SPN can be easily adapted to the infant pose estimation task with superior performance. Last but not least, we explore the clinical application of our method for GMA. We collected a new clinical IMV dataset with GMA annotations, and our experiments show that our SPN models for body parsing and pose estimation trained on the first two datasets generalize well to the new clinical dataset and their results can significantly boost the convolutional recurrent neural network (CRNN) based GMA prediction performance when combined with raw video inputs.     

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.     

Discriminative confidence estimation for probabilistic multi-atlas label fusion

Quantitative neuroimaging analyses often rely on the accurate segmentation of anatomical brain structures. In contrast to manual segmentation, automatic methods offer reproducible outputs and provide scalability to study large databases. Among existing approaches, multi-atlas segmentation has recently shown to yield state-of-the-art performance in automatic segmentation of brain images. It consists in propagating the labelmaps from a set of atlases to the anatomy of a target image using image registration, and then fusing these multiple warped labelmaps into a consensus segmentation on the target image. Accurately estimating the contribution of each atlas labelmap to the final segmentation is a critical step for the success of multi-atlas segmentation. Common approaches to label fusion either rely on local patch similarity, probabilistic statistical frameworks or a combination of both. In this work, we propose a probabilistic label fusion framework based on atlas label confidences computed at each voxel of the structure of interest. Maximum likelihood atlas confidences are estimated using a supervised approach, explicitly modeling the relationship between local image appearances and segmentation errors produced by each of the atlases. We evaluate different spatial pooling strategies for modeling local segmentation errors. We also present a novel type of label-dependent appearance features based on atlas labelmaps that are used during confidence estimation to increase the accuracy of our label fusion. Our approach is evaluated on the segmentation of seven subcortical brain structures from the MICCAI 2013 SATA Challenge dataset and the hippocampi from the ADNI dataset. Overall, our results indicate that the proposed label fusion framework achieves superior performance to state-of-the-art approaches in the majority of the evaluated brain structures and shows more robustness to registration errors.