Pseudo-healthy synthesis with pathology disentanglement and adversarial learning

Lesion region inpainting: an approach for pseudo-healthy image synthesis in intracranial infection imaging

The synthesis of pseudo-healthy images, involving the generation of healthy counterparts for pathological images, is crucial for data augmentation, clinical disease diagnosis, and understanding pathology-induced changes. Recently, Generative Adversarial Networks (GANs) have shown substantial promise in this domain. However, the heterogeneity of intracranial infection symptoms caused by various infections complicates the model's ability to accurately differentiate between pathological and healthy regions, leading to the loss of critical information in healthy areas and impairing the precise preservation of the subject's identity. Moreover, for images with extensive lesion areas, the pseudo-healthy images generated by these methods often lack distinct organ and tissue structures. To address these challenges, we propose a three-stage method (localization, inpainting, synthesis) that achieves nearly perfect preservation of the subject's identity through precise pseudo-healthy synthesis of the lesion region and its surroundings. The process begins with a Segmentor, which identifies the lesion areas and differentiates them from healthy regions. Subsequently, a Vague-Filler fills the lesion areas to construct a healthy outline, thereby preventing structural loss in cases of extensive lesions. Finally, leveraging this healthy outline, a Generative Adversarial Network integrated with a contextual residual attention module generates a more realistic and clearer image. Our method was validated through extensive experiments across different modalities within the BraTS2021 dataset, achieving a healthiness score of 0.957. The visual quality of the generated images markedly exceeded those produced by competing methods, with enhanced capabilities in repairing large lesion areas. Further testing on the COVID-19-20 dataset showed that our model could effectively partially reconstruct images of other organs.

Sketch-based semantic retrieval of medical images

The volume of medical images stored in hospitals is rapidly increasing; however, the utilization of these accumulated medical images remains limited. Existing content-based medical image retrieval (CBMIR) systems typically require example images, leading to practical limitations, such as the lack of customizable, fine-grained image retrieval, the inability to search without example images, and difficulty in retrieving rare cases. In this paper, we introduce a sketch-based medical image retrieval (SBMIR) system that enables users to find images of interest without the need for example images. The key concept is feature decomposition of medical images, which allows the entire feature of a medical image to be decomposed into and reconstructed from normal and abnormal features. Building on this concept, our SBMIR system provides an easy-to-use two-step graphical user interface: users first select a template image to specify a normal feature and then draw a semantic sketch of the disease on the template image to represent an abnormal feature. The system integrates both types of input to construct a query vector and retrieves reference images. For evaluation, ten healthcare professionals participated in a user test using two datasets. Consequently, our SBMIR system enabled users to overcome previous challenges, including image retrieval based on fine-grained image characteristics, image retrieval without example images, and image retrieval for rare cases. Our SBMIR system provides on-demand, customizable medical image retrieval, thereby expanding the utility of medical image databases.

Interpretable Diabetic Retinopathy Diagnosis Based on Biomarker Activation Map

OBJECTIVE

Deep learning classifiers provide the most accurate means of automatically diagnosing diabetic retinopathy (DR) based on optical coherence tomography (OCT) and its angiography (OCTA). The power of these models is attributable in part to the inclusion of hidden layers that provide the complexity required to achieve a desired task. However, hidden layers also render algorithm outputs difficult to interpret. Here we introduce a novel biomarker activation map (BAM) framework based on generative adversarial learning that allows clinicians to verify and understand classifiers' decision-making.

METHODS

A data set including 456 macular scans were graded as non-referable or referable DR based on current clinical standards. A DR classifier that was used to evaluate our BAM was first trained based on this data set. The BAM generation framework was designed by combing two U-shaped generators to provide meaningful interpretability to this classifier. The main generator was trained to take referable scans as input and produce an output that would be classified by the classifier as non-referable. The BAM is then constructed as the difference image between the output and input of the main generator. To ensure that the BAM only highlights classifier-utilized biomarkers an assistant generator was trained to do the opposite, producing scans that would be classified as referable by the classifier from non-referable scans.

RESULTS

The generated BAMs highlighted known pathologic features including nonperfusion area and retinal fluid.

CONCLUSION/SIGNIFICANCE

A fully interpretable classifier based on these highlights could help clinicians better utilize and verify automated DR diagnosis.

A large public dataset of annotated clinical MRIs and metadata of patients with acute stroke

To extract meaningful and reproducible models of brain function from stroke images, for both clinical and research proposes, is a daunting task severely hindered by the great variability of lesion frequency and patterns. Large datasets are therefore imperative, as well as fully automated image post-processing tools to analyze them. The development of such tools, particularly with artificial intelligence, is highly dependent on the availability of large datasets to model training and testing. We present a public dataset of 2,888 multimodal clinical MRIs of patients with acute and early subacute stroke, with manual lesion segmentation, and metadata. The dataset provides high quality, large scale, human-supervised knowledge to feed artificial intelligence models and enable further development of tools to automate several tasks that currently rely on human labor, such as lesion segmentation, labeling, calculation of disease-relevant scores, and lesion-based studies relating function to frequency lesion maps.

Adversarial counterfactual augmentation: application in Alzheimer's disease classification

Due to the limited availability of medical data, deep learning approaches for medical image analysis tend to generalise poorly to unseen data. Augmenting data during training with random transformations has been shown to help and became a ubiquitous technique for training neural networks. Here, we propose a novel adversarial counterfactual augmentation scheme that aims at finding the most effective synthesised images to improve downstream tasks, given a pre-trained generative model. Specifically, we construct an adversarial game where we update the input conditional factor of the generator and the downstream classifier with gradient backpropagation alternatively and iteratively. This can be viewed as finding the 'weakness' of the classifier and purposely forcing it to overcome its weakness via the generative model. To demonstrate the effectiveness of the proposed approach, we validate the method with the classification of Alzheimer's Disease (AD) as a downstream task. The pre-trained generative model synthesises brain images using age as conditional factor. Extensive experiments and ablation studies have been performed to show that the proposed approach improves classification performance and has potential to alleviate spurious correlations and catastrophic forgetting. Code: https://github.com/xiat0616/adversarial_counterfactual_augmentation.

D2-Net: Dual Disentanglement Network for Brain Tumor Segmentation With Missing Modalities

Multi-modal Magnetic Resonance Imaging (MRI) can provide complementary information for automatic brain tumor segmentation, which is crucial for diagnosis and prognosis. While missing modality data is common in clinical practice and it can result in the collapse of most previous methods relying on complete modality data. Current state-of-the-art approaches cope with the situations of missing modalities by fusing multi-modal images and features to learn shared representations of tumor regions, which often ignore explicitly capturing the correlations among modalities and tumor regions. Inspired by the fact that modality information plays distinct roles to segment different tumor regions, we aim to explicitly exploit the correlations among various modality-specific information and tumor-specific knowledge for segmentation. To this end, we propose a Dual Disentanglement Network (D²-Net) for brain tumor segmentation with missing modalities, which consists of a modality disentanglement stage (MD-Stage) and a tumor-region disentanglement stage (TD-Stage). In the MD-Stage, a spatial-frequency joint modality contrastive learning scheme is designed to directly decouple the modality-specific information from MRI data. To decompose tumor-specific representations and extract discriminative holistic features, we propose an affinity-guided dense tumor-region knowledge distillation mechanism in the TD-Stage through aligning the features of a disentangled binary teacher network with a holistic student network. By explicitly discovering relations among modalities and tumor regions, our model can learn sufficient information for segmentation even if some modalities are missing. Extensive experiments on the public BraTS-2018 database demonstrate the superiority of our framework over state-of-the-art methods in missing modalities situations. Codes are available at https://github.com/CityU-AIM-Group/D2Net.

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.

Harmonizing Pathological and Normal Pixels for Pseudo-Healthy Synthesis

Synthesizing a subject-specific pathology-free image from a pathological image is valuable for algorithm development and clinical practice. In recent years, several approaches based on the Generative Adversarial Network (GAN) have achieved promising results in pseudo-healthy synthesis. However, the discriminator (i.e., a classifier) in the GAN cannot accurately identify lesions and further hampers from generating admirable pseudo-healthy images. To address this problem, we present a new type of discriminator, the segmentor, to accurately locate the lesions and improve the visual quality of pseudo-healthy images. Then, we apply the generated images into medical image enhancement and utilize the enhanced results to cope with the low contrast problem existing in medical image segmentation. Furthermore, a reliable metric is proposed by utilizing two attributes of label noise to measure the health of synthetic images. Comprehensive experiments on the T2 modality of BraTS demonstrate that the proposed method substantially outperforms the state-of-the-art methods. The method achieves better performance than the existing methods with only 30% of the training data. The effectiveness of the proposed method is also demonstrated on the LiTS and the T1 modality of BraTS. The code and the pre-trained model of this study are publicly available at https://github.com/Au3C2/Generator-Versus-Segmentor.

Learning disentangled representations in the imaging domain

Disentangled representation learning has been proposed as an approach to learning general representations even in the absence of, or with limited, supervision. A good general representation can be fine-tuned for new target tasks using modest amounts of data, or used directly in unseen domains achieving remarkable performance in the corresponding task. This alleviation of the data and annotation requirements offers tantalising prospects for applications in computer vision and healthcare. In this tutorial paper, we motivate the need for disentangled representations, revisit key concepts, and describe practical building blocks and criteria for learning such representations. We survey applications in medical imaging emphasising choices made in exemplar key works, and then discuss links to computer vision applications. We conclude by presenting limitations, challenges, and opportunities.

When medical images meet generative adversarial network: recent development and research opportunities

Deep learning techniques have promoted the rise of artificial intelligence (AI) and performed well in computer vision. Medical image analysis is an important application of deep learning, which is expected to greatly reduce the workload of doctors, contributing to more sustainable health systems. However, most current AI methods for medical image analysis are based on supervised learning, which requires a lot of annotated data. The number of medical images available is usually small and the acquisition of medical image annotations is an expensive process. Generative adversarial network (GAN), an unsupervised method that has become very popular in recent years, can simulate the distribution of real data and reconstruct approximate real data. GAN opens some exciting new ways for medical image generation, expanding the number of medical images available for deep learning methods. Generated data can solve the problem of insufficient data or imbalanced data categories. Adversarial training is another contribution of GAN to medical imaging that has been applied to many tasks, such as classification, segmentation, or detection. This paper investigates the research status of GAN in medical images and analyzes several GAN methods commonly applied in this area. The study addresses GAN application for both medical image synthesis and adversarial learning for other medical image tasks. The open challenges and future research directions are also discussed.

Unsupervised anomaly detection in MR images using multicontrast information

PURPOSE

Anomaly detection in magnetic resonance imaging (MRI) is to distinguish the relevant biomarkers of diseases from those of normal tissues. In this paper, an unsupervised algorithm is proposed for pixel-level anomaly detection in multicontrast MRI.

METHODS

A deep neural network is developed, which uses only normal MR images as training data. The network has the two stages of feature generation and density estimation. For feature generation, relevant features are extracted from multicontrast MR images by performing contrast translation and dimension reduction. For density estimation, the distributions of the extracted features are estimated by using Gaussian mixture model (GMM). The two processes are trained to estimate normative distributions well presenting large normal datasets. In test phases, the proposed method can detect anomalies by measuring log-likelihood that a test sample belongs to the estimated normative distributions.

RESULTS

The proposed method and its variants were applied to detect glioblastoma and ischemic stroke lesion. Comparison studies with six previous anomaly detection algorithms demonstrated that the proposed method achieved relevant improvements in quantitative and qualitative evaluations. Ablation studies by removing each module from the proposed framework validated the effectiveness of each proposed module.

CONCLUSION

The proposed deep learning framework is an effective tool to detect anomalies in multicontrast MRI. The unsupervised approaches would have great potentials in detecting various lesions where annotated lesion data collection is limited.

Decomposing normal and abnormal features of medical images for content-based image retrieval of glioma imaging

In medical imaging, the characteristics purely derived from a disease should reflect the extent to which abnormal findings deviate from the normal features. Indeed, physicians often need corresponding images without abnormal findings of interest or, conversely, images that contain similar abnormal findings regardless of normal anatomical context. This is called comparative diagnostic reading of medical images, which is essential for a correct diagnosis. To support comparative diagnostic reading, content-based image retrieval (CBIR) that can selectively utilize normal and abnormal features in medical images as two separable semantic components will be useful. In this study, we propose a neural network architecture to decompose the semantic components of medical images into two latent codes: normal anatomy code and abnormal anatomy code. The normal anatomy code represents counterfactual normal anatomies that should have existed if the sample is healthy, whereas the abnormal anatomy code attributes to abnormal changes that reflect deviation from the normal baseline. By calculating the similarity based on either normal or abnormal anatomy codes or the combination of the two codes, our algorithm can retrieve images according to the selected semantic component from a dataset consisting of brain magnetic resonance images of gliomas. Moreover, it can utilize a synthetic query vector combining normal and abnormal anatomy codes from two different query images. To evaluate whether the retrieved images are acquired according to the targeted semantic component, the overlap of the ground-truth labels is calculated as metrics of the semantic consistency. Our algorithm provides a flexible CBIR framework by handling the decomposed features with qualitatively and quantitatively remarkable results.

Adversarial attack vulnerability of medical image analysis systems: Unexplored factors

Adversarial attacks are considered a potentially serious security threat for machine learning systems. Medical image analysis (MedIA) systems have recently been argued to be vulnerable to adversarial attacks due to strong financial incentives and the associated technological infrastructure. In this paper, we study previously unexplored factors affecting adversarial attack vulnerability of deep learning MedIA systems in three medical domains: ophthalmology, radiology, and pathology. We focus on adversarial black-box settings, in which the attacker does not have full access to the target model and usually uses another model, commonly referred to as surrogate model, to craft adversarial examples that are then transferred to the target model. We consider this to be the most realistic scenario for MedIA systems. Firstly, we study the effect of weight initialization (pre-training on ImageNet or random initialization) on the transferability of adversarial attacks from the surrogate model to the target model, i.e., how effective attacks crafted using the surrogate model are on the target model. Secondly, we study the influence of differences in development (training and validation) data between target and surrogate models. We further study the interaction of weight initialization and data differences with differences in model architecture. All experiments were done with a perturbation degree tuned to ensure maximal transferability at minimal visual perceptibility of the attacks. Our experiments show that pre-training may dramatically increase the transferability of adversarial examples, even when the target and surrogate's architectures are different: the larger the performance gain using pre-training, the larger the transferability. Differences in the development data between target and surrogate models considerably decrease the performance of the attack; this decrease is further amplified by difference in the model architecture. We believe these factors should be considered when developing security-critical MedIA systems planned to be deployed in clinical practice. We recommend avoiding using only standard components, such as pre-trained architectures and publicly available datasets, as well as disclosure of design specifications, in addition to using adversarial defense methods. When evaluating the vulnerability of MedIA systems to adversarial attacks, various attack scenarios and target-surrogate differences should be simulated to achieve realistic robustness estimates. The code and all trained models used in our experiments are publicly available.³.

Lung Lesion Localization of COVID-19 From Chest CT Image: A Novel Weakly Supervised Learning Method

Chest computed tomography (CT) image data is necessary for early diagnosis, treatment, and prognosis of Coronavirus Disease 2019 (COVID-19). Artificial intelligence has been tried to help clinicians in improving the diagnostic accuracy and working efficiency of CT. Whereas, existing supervised approaches on CT image of COVID-19 pneumonia require voxel-based annotations for training, which take a lot of time and effort. This paper proposed a weakly-supervised method for COVID-19 lesion localization based on generative adversarial network (GAN) with image-level labels only. We first introduced a GAN-based framework to generate normal-looking CT slices from CT slices with COVID-19 lesions. We then developed a novel feature match strategy to improve the reality of generated images by guiding the generator to capture the complex texture of chest CT images. Finally, the localization map of lesions can be easily obtained by subtracting the output image from its corresponding input image. By adding a classifier branch to the GAN-based framework to classify localization maps, we can further develop a diagnosis system with improved classification accuracy. Three CT datasets from hospitals of Sao Paulo, Italian Society of Medical and Interventional Radiology, and China Medical University about COVID-19 were collected in this article for evaluation. Our weakly supervised learning method obtained AUC of 0.883, dice coefficient of 0.575, accuracy of 0.884, sensitivity of 0.647, specificity of 0.929, and F1-score of 0.640, which exceeded other widely used weakly supervised object localization methods by a significant margin. We also compared the proposed method with fully supervised learning methods in COVID-19 lesion segmentation task, the proposed weakly supervised method still leads to a competitive result with dice coefficient of 0.575. Furthermore, we also analyzed the association between illness severity and visual score, we found that the common severity cohort had the largest sample size as well as the highest visual score which suggests our method can help rapid diagnosis of COVID-19 patients, especially in massive common severity cohort. In conclusion, we proposed this novel method can serve as an accurate and efficient tool to alleviate the bottleneck of expert annotation cost and advance the progress of computer-aided COVID-19 diagnosis.

Anomaly detection for the individual analysis of brain PET images

Purpose: In clinical practice, positron emission tomography (PET) images are mostly analyzed visually, but the sensitivity and specificity of this approach greatly depend on the observer's experience. Quantitative analysis of PET images would alleviate this problem by helping define an objective limit between normal and pathological findings. We present an anomaly detection framework for the individual analysis of PET images. Approach: We created subject-specific abnormality maps that summarize the pathology's topographical distribution in the brain by comparing the subject's PET image to a model of healthy PET appearance that is specific to the subject under investigation. This model was generated from demographically and morphologically matched PET scans from a control dataset. Results: We generated abnormality maps for healthy controls, patients at different stages of Alzheimer's disease and with different frontotemporal dementia syndromes. We showed that no anomalies were detected for the healthy controls and that the anomalies detected from the patients with dementia coincided with the regions where abnormal uptake was expected. We also validated the proposed framework using the abnormality maps as inputs of a classifier and obtained higher classification accuracies than when using the PET images themselves as inputs. Conclusions: The proposed method was able to automatically locate and characterize the areas characteristic of dementia from PET images. The abnormality maps are expected to (i) help clinicians in their diagnosis by highlighting, in a data-driven fashion, the pathological areas, and (ii) improve the interpretability of subsequent analyses, such as computer-aided diagnosis or spatiotemporal modeling.