Johan W. Verjans

Minh Hieu Phan, Zhibin Liao, Johan W. Verjans et al.

Medical image synthesis is a challenging task due to the scarcity of paired data. Several methods have applied CycleGAN to leverage unpaired data, but they often generate inaccurate mappings that shift the anatomy. This problem is further exacerbated when the images from the source and target modalities are heavily misaligned. Recently, current methods have aimed to address this issue by incorporating a supplementary segmentation network. Unfortunately, this strategy requires costly and time-consuming pixel-level annotations. To overcome this problem, this paper proposes MaskGAN, a novel and cost-effective framework that enforces structural consistency by utilizing automatically extracted coarse masks. Our approach employs a mask generator to outline anatomical structures and a content generator to synthesize CT contents that align with these structures. Extensive experiments demonstrate that MaskGAN outperforms state-of-the-art synthesis methods on a challenging pediatric dataset, where MR and CT scans are heavily misaligned due to rapid growth in children. Specifically, MaskGAN excels in preserving anatomical structures without the need for expert annotations. The code for this paper can be found at https://github.com/HieuPhan33/MaskGAN.

Zhibin Liao, Kewen Liao, Haifeng Shen et al.

Convolutional neural networks (CNNs) have gained significant popularity in orthopedic imaging in recent years due to their ability to solve fracture classification problems. A common criticism of CNNs is their opaque learning and reasoning process, making it difficult to trust machine diagnosis and the subsequent adoption of such algorithms in clinical setting. This is especially true when the CNN is trained with limited amount of medical data, which is a common issue as curating sufficiently large amount of annotated medical imaging data is a long and costly process. While interest has been devoted to explaining CNN learnt knowledge by visualizing network attention, the utilization of the visualized attention to improve network learning has been rarely investigated. This paper explores the effectiveness of regularizing CNN network with human-provided attention guidance on where in the image the network should look for answering clues. On two orthopedics radiographic fracture classification datasets, through extensive experiments we demonstrate that explicit human-guided attention indeed can direct correct network attention and consequently significantly improve classification performance. The development code for the proposed attention guidance is publicly available on GitHub.

Townim F. Chowdhury, Vu Minh Hieu Phan, Kewen Liao et al.

The integration of vision-language models such as CLIP and Concept Bottleneck Models (CBMs) offers a promising approach to explaining deep neural network (DNN) decisions using concepts understandable by humans, addressing the black-box concern of DNNs. While CLIP provides both explainability and zero-shot classification capability, its pre-training on generic image and text data may limit its classification accuracy and applicability to medical image diagnostic tasks, creating a transfer learning problem. To maintain explainability and address transfer learning needs, CBM methods commonly design post-processing modules after the bottleneck module. However, this way has been ineffective. This paper takes an unconventional approach by re-examining the CBM framework through the lens of its geometrical representation as a simple linear classification system. The analysis uncovers that post-CBM fine-tuning modules merely rescale and shift the classification outcome of the system, failing to fully leverage the system's learning potential. We introduce an adaptive module strategically positioned between CLIP and CBM to bridge the gap between source and downstream domains. This simple yet effective approach enhances classification performance while preserving the explainability afforded by the framework. Our work offers a comprehensive solution that encompasses the entire process, from concept discovery to model training, providing a holistic recipe for leveraging the strengths of GPT, CLIP, and CBM.

Vu Minh Hieu Phan, Yutong Xie, Yuankai Qi et al.

Medical vision language pre-training (VLP) has emerged as a frontier of research, enabling zero-shot pathological recognition by comparing the query image with the textual descriptions for each disease. Due to the complex semantics of biomedical texts, current methods struggle to align medical images with key pathological findings in unstructured reports. This leads to the misalignment with the target disease's textual representation. In this paper, we introduce a novel VLP framework designed to dissect disease descriptions into their fundamental aspects, leveraging prior knowledge about the visual manifestations of pathologies. This is achieved by consulting a large language model and medical experts. Integrating a Transformer module, our approach aligns an input image with the diverse elements of a disease, generating aspect-centric image representations. By consolidating the matches from each aspect, we improve the compatibility between an image and its associated disease. Additionally, capitalizing on the aspect-oriented representations, we present a dual-head Transformer tailored to process known and unknown diseases, optimizing the comprehensive detection efficacy. Conducting experiments on seven downstream datasets, ours improves the accuracy of recent methods by up to 8.56% and 17.26% for seen and unseen categories, respectively. Our code is released at https://github.com/HieuPhan33/MAVL.

Ta Duc Huy, Sen Kim Tran, Phan Nguyen et al.

The ability to interpret and intervene model decisions is important for the adoption of computer-aided diagnosis methods in clinical workflows. Recent concept-based methods link the model predictions with interpretable concepts and modify their activation scores to interact with the model. However, these concepts are at the image level, which hinders the model from pinpointing the exact patches the concepts are activated. Alternatively, prototype-based methods learn representations from training image patches and compare these with test image patches, using the similarity scores for final class prediction. However, interpreting the underlying concepts of these patches can be challenging and often necessitates post-hoc guesswork. To address this issue, this paper introduces the novel Concept-based Similarity Reasoning network (CSR), which offers (i) patch-level prototype with intrinsic concept interpretation, and (ii) spatial interactivity. First, the proposed CSR provides localized explanation by grounding prototypes of each concept on image regions. Second, our model introduces novel spatial-level interaction, allowing doctors to engage directly with specific image areas, making it an intuitive and transparent tool for medical imaging. CSR improves upon prior state-of-the-art interpretable methods by up to 4.5\% across three biomedical datasets. Our code is released at https://github.com/tadeephuy/InteractCSR.

Townim Faisal Chowdhury, Kewen Liao, Vu Minh Hieu Phan et al.

Deep Neural Networks (DNNs) are widely used for visual classification tasks, but their complex computation process and black-box nature hinder decision transparency and interpretability. Class activation maps (CAMs) and recent variants provide ways to visually explain the DNN decision-making process by displaying 'attention' heatmaps of the DNNs. Nevertheless, the CAM explanation only offers relative attention information, that is, on an attention heatmap, we can interpret which image region is more or less important than the others. However, these regions cannot be meaningfully compared across classes, and the contribution of each region to the model's class prediction is not revealed. To address these challenges that ultimately lead to better DNN Interpretation, in this paper, we propose CAPE, a novel reformulation of CAM that provides a unified and probabilistically meaningful assessment of the contributions of image regions. We quantitatively and qualitatively compare CAPE with state-of-the-art CAM methods on CUB and ImageNet benchmark datasets to demonstrate enhanced interpretability. We also test on a cytology imaging dataset depicting a challenging Chronic Myelomonocytic Leukemia (CMML) diagnosis problem. Code is available at: https://github.com/AIML-MED/CAPE.

Vu Minh Hieu Phan, Yutong Xie, Bowen Zhang et al.

Unpaired medical image synthesis aims to provide complementary information for an accurate clinical diagnostics, and address challenges in obtaining aligned multi-modal medical scans. Transformer-based models excel in imaging translation tasks thanks to their ability to capture long-range dependencies. Although effective in supervised training settings, their performance falters in unpaired image synthesis, particularly in synthesizing structural details. This paper empirically demonstrates that, lacking strong inductive biases, Transformer can converge to non-optimal solutions in the absence of paired data. To address this, we introduce UNet Structured Transformer (UNest), a novel architecture incorporating structural inductive biases for unpaired medical image synthesis. We leverage the foundational Segment-Anything Model to precisely extract the foreground structure and perform structural attention within the main anatomy. This guides the model to learn key anatomical regions, thus improving structural synthesis under the lack of supervision in unpaired training. Evaluated on two public datasets, spanning three modalities, i.e., MR, CT, and PET, UNest improves recent methods by up to 19.30% across six medical image synthesis tasks. Our code is released at https://github.com/HieuPhan33/MICCAI2024-UNest.

Yu Tian, Fengbei Liu, Guansong Pang et al.

Unsupervised anomaly detection (UAD) methods are trained with normal (or healthy) images only, but during testing, they are able to classify normal and abnormal (or disease) images. UAD is an important medical image analysis (MIA) method to be applied in disease screening problems because the training sets available for those problems usually contain only normal images. However, the exclusive reliance on normal images may result in the learning of ineffective low-dimensional image representations that are not sensitive enough to detect and segment unseen abnormal lesions of varying size, appearance, and shape. Pre-training UAD methods with self-supervised learning, based on computer vision techniques, can mitigate this challenge, but they are sub-optimal because they do not explore domain knowledge for designing the pretext tasks, and their contrastive learning losses do not try to cluster the normal training images, which may result in a sparse distribution of normal images that is ineffective for anomaly detection. In this paper, we propose a new self-supervised pre-training method for MIA UAD applications, named Pseudo Multi-class Strong Augmentation via Contrastive Learning (PMSACL). PMSACL consists of a novel optimisation method that contrasts a normal image class from multiple pseudo classes of synthesised abnormal images, with each class enforced to form a dense cluster in the feature space. In the experiments, we show that our PMSACL pre-training improves the accuracy of SOTA UAD methods on many MIA benchmarks using colonoscopy, fundus screening and Covid-19 Chest X-ray datasets. The code is made publicly available via https://github.com/tianyu0207/PMSACL.

Yu Tian, Guansong Pang, Fengbei Liu et al.

Unsupervised anomaly detection (UAD) learns one-class classifiers exclusively with normal (i.e., healthy) images to detect any abnormal (i.e., unhealthy) samples that do not conform to the expected normal patterns. UAD has two main advantages over its fully supervised counterpart. Firstly, it is able to directly leverage large datasets available from health screening programs that contain mostly normal image samples, avoiding the costly manual labelling of abnormal samples and the subsequent issues involved in training with extremely class-imbalanced data. Further, UAD approaches can potentially detect and localise any type of lesions that deviate from the normal patterns. One significant challenge faced by UAD methods is how to learn effective low-dimensional image representations to detect and localise subtle abnormalities, generally consisting of small lesions. To address this challenge, we propose a novel self-supervised representation learning method, called Constrained Contrastive Distribution learning for anomaly detection (CCD), which learns fine-grained feature representations by simultaneously predicting the distribution of augmented data and image contexts using contrastive learning with pretext constraints. The learned representations can be leveraged to train more anomaly-sensitive detection models. Extensive experiment results show that our method outperforms current state-of-the-art UAD approaches on three different colonoscopy and fundus screening datasets. Our code is available at https://github.com/tianyu0207/CCD.

Yu Tian, Guansong Pang, Yuanhong Chen et al.

Anomaly detection with weakly supervised video-level labels is typically formulated as a multiple instance learning (MIL) problem, in which we aim to identify snippets containing abnormal events, with each video represented as a bag of video snippets. Although current methods show effective detection performance, their recognition of the positive instances, i.e., rare abnormal snippets in the abnormal videos, is largely biased by the dominant negative instances, especially when the abnormal events are subtle anomalies that exhibit only small differences compared with normal events. This issue is exacerbated in many methods that ignore important video temporal dependencies. To address this issue, we introduce a novel and theoretically sound method, named Robust Temporal Feature Magnitude learning (RTFM), which trains a feature magnitude learning function to effectively recognise the positive instances, substantially improving the robustness of the MIL approach to the negative instances from abnormal videos. RTFM also adapts dilated convolutions and self-attention mechanisms to capture long- and short-range temporal dependencies to learn the feature magnitude more faithfully. Extensive experiments show that the RTFM-enabled MIL model (i) outperforms several state-of-the-art methods by a large margin on four benchmark data sets (ShanghaiTech, UCF-Crime, XD-Violence and UCSD-Peds) and (ii) achieves significantly improved subtle anomaly discriminability and sample efficiency. Code is available at https://github.com/tianyu0207/RTFM.

Ta Duc Huy, Duy Anh Huynh, Yutong Xie et al.

Visual grounding (VG) is the capability to identify the specific regions in an image associated with a particular text description. In medical imaging, VG enhances interpretability by highlighting relevant pathological features corresponding to textual descriptions, improving model transparency and trustworthiness for wider adoption of deep learning models in clinical practice. Current models struggle to associate textual descriptions with disease regions due to inefficient attention mechanisms and a lack of fine-grained token representations. In this paper, we empirically demonstrate two key observations. First, current VLMs assign high norms to background tokens, diverting the model's attention from regions of disease. Second, the global tokens used for cross-modal learning are not representative of local disease tokens. This hampers identifying correlations between the text and disease tokens. To address this, we introduce simple, yet effective Disease-Aware Prompting (DAP) process, which uses the explainability map of a VLM to identify the appropriate image features. This simple strategy amplifies disease-relevant regions while suppressing background interference. Without any additional pixel-level annotations, DAP improves visual grounding accuracy by 20.74% compared to state-of-the-art methods across three major chest X-ray datasets.

Tuan Dung Nguyen, Minh Khoi Ho, Qi Chen et al.

Large vision-language models (LVLMs) achieve strong performance on visual reasoning tasks but remain highly susceptible to hallucination. Existing detection methods predominantly rely on coarse, whole-image measures of how an object token relates to the input image. This global strategy is limited: hallucinated tokens may exhibit weak but widely scattered correlations across many local regions, which aggregate into deceptively high overall relevance, thus evading the current global hallucination detectors. We begin with a simple yet critical observation: a faithful object token must be strongly grounded in a specific image region. Building on this insight, we introduce a patch-level hallucination detection framework that examines fine-grained token-level interactions across model layers. Our analysis uncovers two characteristic signatures of hallucinated tokens: (i) they yield diffuse, non-localized attention patterns, in contrast to the compact, well-focused attention seen in faithful tokens; and (ii) they fail to exhibit meaningful semantic alignment with any visual region. Guided by these findings, we develop a lightweight and interpretable detection method that leverages patch-level statistical features, combined with hidden-layer representations. Our approach achieves up to 90% accuracy in token-level hallucination detection, demonstrating the superiority of fine-grained structural analysis for detecting hallucinations.

Abin Shoby, Ta Duc Huy, Tuan Dung Nguyen et al.

Vision Language models (VLMs) often hallucinate non-existent objects. Detecting hallucination is analogous to detecting deception: a single final statement is insufficient, one must examine the underlying reasoning process. Yet existing detectors rely mostly on final-layer signals. Attention-based methods assume hallucinated tokens exhibit low attention, while entropy-based ones use final-step uncertainty. Our analysis reveals the opposite: hallucinated objects can exhibit peaked attention due to contextual priors; and models often express high confidence because intermediate layers have already converged to an incorrect hypothesis. We show that the key to hallucination detection lies within the model's thought process, not its final output. By probing decoder layers, we uncover a previously overlooked behavior, overthinking: models repeatedly revise object hypotheses across layers before committing to an incorrect answer. Once the model latches onto a confounded hypothesis, it can propagate through subsequent layers, ultimately causing hallucination. To capture this behavior, we introduce the Overthinking Score, a metric to measure how many competing hypotheses the model entertains and how unstable these hypotheses are across layers. This score significantly improves hallucination detection: 78.9% F1 on MSCOCO and 71.58% on AMBER.

Yu Tian, Leonardo Zorron Cheng Tao Pu, Yuyuan Liu et al.

In this paper, we propose and analyse a system that can automatically detect, localise and classify polyps from colonoscopy videos. The detection of frames with polyps is formulated as a few-shot anomaly classification problem, where the training set is highly imbalanced with the large majority of frames consisting of normal images and a small minority comprising frames with polyps. Colonoscopy videos may contain blurry images and frames displaying feces and water jet sprays to clean the colon -- such frames can mistakenly be detected as anomalies, so we have implemented a classifier to reject these two types of frames before polyp detection takes place. Next, given a frame containing a polyp, our method localises (with a bounding box around the polyp) and classifies it into five different classes. Furthermore, we study a method to improve the reliability and interpretability of the classification result using uncertainty estimation and classification calibration. Classification uncertainty and calibration not only help improve classification accuracy by rejecting low-confidence and high-uncertain results, but can be used by doctors to decide how to decide on the classification of a polyp. All the proposed detection, localisation and classification methods are tested using large data sets and compared with relevant baseline approaches.

Yu Tian, Gabriel Maicas, Leonardo Zorron Cheng Tao Pu et al.

Anomaly detection methods generally target the learning of a normal image distribution (i.e., inliers showing healthy cases) and during testing, samples relatively far from the learned distribution are classified as anomalies (i.e., outliers showing disease cases). These approaches tend to be sensitive to outliers that lie relatively close to inliers (e.g., a colonoscopy image with a small polyp). In this paper, we address the inappropriate sensitivity to outliers by also learning from inliers. We propose a new few-shot anomaly detection method based on an encoder trained to maximise the mutual information between feature embeddings and normal images, followed by a few-shot score inference network, trained with a large set of inliers and a substantially smaller set of outliers. We evaluate our proposed method on the clinical problem of detecting frames containing polyps from colonoscopy video sequences, where the training set has 13350 normal images (i.e., without polyps) and less than 100 abnormal images (i.e., with polyps). The results of our proposed model on this data set reveal a state-of-the-art detection result, while the performance based on different number of anomaly samples is relatively stable after approximately 40 abnormal training images.

Yuyuan Liu, Yu Tian, Gabriel Maicas et al.

The automatic detection of frames containing polyps from a colonoscopy video sequence is an important first step for a fully automated colonoscopy analysis tool. Typically, such detection system is built using a large annotated data set of frames with and without polyps, which is expensive to be obtained. In this paper, we introduce a new system that detects frames containing polyps as anomalies from a distribution of frames from exams that do not contain any polyps. The system is trained using a one-class training set consisting of colonoscopy frames without polyps -- such training set is considerably less expensive to obtain, compared to the 2-class data set mentioned above. During inference, the system is only able to reconstruct frames without polyps, and when it tries to reconstruct a frame with polyp, it automatically removes (i.e., photoshop) it from the frame -- the difference between the input and reconstructed frames is used to detect frames with polyps. We name our proposed model as anomaly detection generative adversarial network (ADGAN), comprising a dual GAN with two generators and two discriminators. We show that our proposed approach achieves the state-of-the-art result on this data set, compared with recently proposed anomaly detection systems.