The deployment of large, black-box foundation models for medical image classification is often hindered by the high cost of acquiring large, task-specific labeled datasets for fine-tuning. While active learning (AL) presents a promising solution, many state-of-the-art AL methods are computationally expensive or require full access to internal model parameters. We present VALIANT (Visual Adaptation and Learning Integration for Active learNing Tasks), a new active learning framework designed to efficiently adapt black-box foundation models by overcoming these limitations. VALIANT introduces a lightweight Visual Prompt Decoder (VIPD), trained via unsupervised Zero-Order Optimization (ZOO), to generate task-specific visual prompts without internal model access. Our core contribution is a perturbation-based ranking strategy that leverages this VIPD to formulate a computationally efficient, gradient-aware informativeness metric. This metric, which we term prompt instability, identifies the most impactful samples for the labeling budget. VALIANT further enhances this process by incorporating anatomical information from unsupervised segmentation maps to generate more discriminative visual prompts. Extensive evaluations on multiple medical datasets demonstrate VALIANT’s superior performance and significant reduction in labeling costs compared to a range of existing active learning techniques, positioning it as a scalable and practical solution for medical image analysis.

Pretraining foundation models for chest X-rays have improved by leveraging paired radiographs and radiology reports, sometimes augmented with medical definitions. However, current approaches often fail to encode rich, granular medical knowledge: radiology reports capture current disease manifestations, while abstract definitions remain overly generalized. To address this, DeViDe integrates open-source radiographic disease descriptions from the web (detailing general visual characteristics of diseases) alongside radiology reports and abstract definitions to form a comprehensive knowledge snapshot. DeViDe introduces three key novelties: (i) web-derived radiographic descriptions and their augmentation using large language models; (ii) a pipeline to enrich raw radiology reports with knowledge from these descriptions; and (iii) a multi-label alignment strategy to effectively align each image with multiple descriptions. Extensive experiments show that in zero-shot settings, DeViDe performs competitively with fully supervised models on external datasets and achieves state-of-the-art results on four large-scale benchmarks. Furthermore, fine-tuning on five classification and three segmentation tasks confirms its superior performance across diverse data distributions.

Deep learning-based brain tumor segmentation on multi-sequence magnetic resonance imaging (MRI) has gained widespread attention due to its great potential in supporting brain disease diagnosis. Although, compared to single-sequence images, more information is available from multi-sequence MR images, noise and artifacts on any given MR sequence can result in significant performance degradations. As in clinical routine, it is not always possible to maintain high imaging quality across all MR sequences (e.g., foreign bodies, ventricular drainage, shunts, involuntary patient motion, etc.), ensuring robustness of brain tumor segmentation from multi-sequence MR images is of great importance in clinical practice, but rarely explored. Accordingly, in this paper, we propose a robust brain tumor segmentation framework to mitigate the performance degradation caused by noise and artifacts on multi-sequence MR images. Specifically, based on semantic affinity, we propose a unique cross-sequence semantic interaction module (CSSI) to exploit inter-sequence correlations and extract noise-resilient features. In addition, we incorporate a batch-level covariance mechanism to suppress the redundant background information and improve the semantic enhancement effect of the CSSI module. In order to further improve segmentation performance, we also incorporate a sequence-level variance regularization mechanism to exploit sequence-specific features. To validate the robustness of ROXSI, brain tumor segmentation performance was evaluated under the existence of four common artifacts, at five different perturbation levels. We further performed a blinded qualitative clinical evaluation with two experienced neuro-radiologists, evaluating results from ROXSI and other popular CNN and Transformer-based segmentation models. Experimental results on two benchmark datasets demonstrate the superior robustness of ROXSI over other state-of-the-art segmentation methods.

In fully supervised learning-based medical image classification, the robustness of a trained model is influenced by its exposure to the range of candidate disease classes. Generalized Zero Shot Learning (GZSL) aims to correctly predict seen and novel unseen classes. Current GZSL approaches have focused mostly on the single-label case. However, it is common for chest X-rays to be labelled with multiple disease classes. We propose a novel multi-modal multi-label GZSL approach that leverages feature disentanglement andmulti-modal information to synthesize features of unseen classes. Disease labels are processed through a pre-trained BioBert model to obtain text embeddings that are used to create a dictionary encoding similarity among different labels. We then use disentangled features and graph aggregation to learn a second dictionary of inter-label similarities. A subsequent clustering step helps to identify representative vectors for each class. The multi-modal multi-label dictionaries and the class representative vectors are used to guide the feature synthesis step, which is the most important component of our pipeline, for generating realistic multi-label disease samples of seen and unseen classes. Our method is benchmarked against multiple competing methods and we outperform all of them based on experiments conducted on the publicly available NIH and CheXpert chest X-ray datasets.

Presents corrections to the paper, (Corrections to "Multi-Label Generalized Zero Shot Chest X-Ray Classification By Combining Image-Text Information With Feature Disentanglement").

Informative sample selection in active learning (AL) helps a machine learning system attain optimum performance with minimum labeled samples, thus improving human-in-the-loop computer-aided diagnosis systems with limited labeled data. Data augmentation is highly effective for enlarging datasets with less labeled data. Combining informative sample selection and data augmentation should leverage their respective advantages and improve performance of AL systems. We propose a novel approach to combine informative sample selection and data augmentation for multi-label active learning. Conventional informative sample selection approaches have mostly focused on the single-label case which do not perform optimally in the multi-label setting. We improve upon state-of-the-art multi-label active learning techniques by representing disease labels as graph nodes, use graph attention transformers (GAT) to learn more effective inter-label relationships and identify most informative samples. We generate transformations of these informative samples which are also informative. Experiments on public chest xray datasets show improved results over state-of-the-art multi-label AL techniques in terms of classification performance, learning rates, and robustness. We also perform qualitative analysis to determine the realism of generated images.

While supervised learning techniques have demonstrated state-of-the-art performance in many medical image analysis tasks, the role of sample selection is important. Selecting the most informative samples contributes to the system attaining optimum performance with minimum labeled samples, which translates to fewer expert interventions and cost. Active Learning (AL) methods for informative sample selection are effective in boosting performance of computer aided diagnosis systems when limited labels are available. Conventional approaches to AL have mostly focused on the single label setting where a sample has only one disease label from the set of possible labels. These approaches do not perform optimally in the multi-label setting where a sample can have multiple disease labels (e.g. in chest X-ray images). In this paper we propose a novel sample selection approach based on graph analysis to identify informative samples in a multi-label setting. For every analyzed sample, each class label is denoted as a separate node of a graph. Building on findings from interpretability of deep learning models, edge interactions in this graph characterize similarity between corresponding interpretability saliency map model encodings. We explore different types of graph aggregation to identify informative samples for active learning. We apply our method to public chest X-ray and medical image datasets, and report improved results over state-of-the-art AL techniques in terms of model performance, learning rates, and robustness.

In many real world medical image classification settings, access to samples of all disease classes is not feasible, affecting the robustness of a system expected to have high performance in analyzing novel test data. This is a case of generalized zero shot learning (GZSL) aiming to recognize seen and unseen classes. We propose a GZSL method that uses self supervised learning (SSL) for: 1) selecting representative vectors of disease classes; and 2) synthesizing features of unseen classes. We also propose a novel approach to generate GradCAM saliency maps that highlight diseased regions with greater accuracy. We exploit information from the novel saliency maps to improve the clustering process by: 1) Enforcing the saliency maps of different classes to be different; and 2) Ensuring that clusters in the space of image and saliency features should yield class centroids having similar semantic information. This ensures the anchor vectors are representative of each class. Different from previous approaches, our proposed approach does not require class attribute vectors which are essential part of GZSL methods for natural images but are not available for medical images. Using a simple architecture the proposed method outperforms state of the art SSL based GZSL performance for natural images as well as multiple types of medical images. We also conduct many ablation studies to investigate the influence of different loss terms in our method.

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

In supervised learning for medical image analysis, sample selection methodologies are fundamental to attain optimum system performance promptly and with minimal expert interactions (e.g. label querying in an active learning setup). In this article we propose a novel sample selection methodology based on deep features leveraging information contained in interpretability saliency maps. In the absence of ground truth labels for informative samples, we use a novel self supervised learning based approach for training a classifier that learns to identify the most informative sample in a given batch of images. We demonstrate the benefits of the proposed approach, termed Interpretability-Driven Sample Selection (IDEAL), in an active learning setup aimed at lung disease classification and histopathology image segmentation. We analyze three different approaches to determine sample informativeness from interpretability saliency maps: (i) an observational model stemming from findings on previous uncertainty-based sample selection approaches, (ii) a radiomics-based model, and (iii) a novel data-driven self-supervised approach. We compare IDEAL to other baselines using the publicly available NIH chest X-ray dataset for lung disease classification, and a public histopathology segmentation dataset (GLaS), demonstrating the potential of using interpretability information for sample selection in active learning systems. Results show our proposed self supervised approach outperforms other approaches in selecting informative samples leading to state of the art performance with fewer samples.

The analysis of Magnetic Resonance Imaging (MRI) sequences enables clinical professionals to monitor the progression of a brain tumor. As the interest for automatizing brain volume MRI analysis increases, it becomes convenient to have each sequence well identified. However, the unstandardized naming of MRI sequences makes their identification difficult for automated systems, as well as makes it difficult for researches to generate or use datasets for machine learning research. In the face of that, we propose a system for identifying types of brain MRI sequences based on deep learning. By training a Convolutional Neural Network (CNN) based on 18-layer ResNet architecture, our system can classify a volumetric brain MRI as a FLAIR, Tl, T1c or T2 sequence, or whether it does not belong to any of these classes. The network was evaluated on publicly available datasets comprising both, pre-processed (BraTS dataset) and non-pre-processed (TCGA-GBM dataset), image types with diverse acquisition protocols, requiring only a few slices of the volume for training. Our system can classify among sequence types with an accuracy of 96.81 %.

Saliency maps have been used as one possibility to interpret deep neural networks. This method estimates the relevance of each pixel in the image classification, with higher values representing pixels which contribute positively to classification. The goal of this study is to understand how the complexity of the network affects the interpretabilty of the saliency maps in classification tasks. To achieve that, we investigate how changes in the regularization affects the saliency maps produced, and their fidelity to the overall classification process of the network.The experimental setup consists in the calculation of the fidelity of five saliency map methods that were compare, applying them to models trained on the CIFAR-10 dataset, using different levels of weight decay on some or all the layers. Achieved results show that models with lower regularization are statistically (significance of 5%) more interpretable than the other models. Also, regularization applied only to the higher convolutional layers or fully-connected layers produce saliency maps with more fidelity.

Magnetic resonance fingerprinting (MRF) provides a unique concept for simultaneous and fast acquisition of multiple quantitative MR parameters. Despite acquisition efficiency, adoption of MRF into the clinics is hindered by its dictionary matching-based reconstruction, which is computationally demanding and lacks scalability. Here, we propose a convolutional neural network-based reconstruction, which enables both accurate and fast reconstruction of parametric maps, and is adaptable based on the needs of spatial regularization and the capacity for the reconstruction. We evaluated the method using MRF T1-FF, an MRF sequence for T1 relaxation time of water (T1H2O) and fat fraction (FF) mapping. We demonstrate the method’s performance on a highly heterogeneous dataset consisting of 164 patients with various neuromuscular diseases imaged at thighs and legs. We empirically show the benefit of incorporating spatial regularization during the reconstruction and demonstrate that the method learns meaningful features from MR physics perspective. Further, we investigate the ability of the method to handle highly heterogeneous morphometric variations and its generalization to anatomical regions unseen during training. The obtained results outperform the state-of-the-art in deep learning-based MRF reconstruction. The method achieved normalized root mean squared errors of 0.048 ± 0.011 for T1H2O maps and 0.027 ± 0.004 for FF maps when compared to the dictionary matching in a test set of 50 patients. Coupled with fast MRF sequences, the proposed method has the potential of enabling multiparametric MR imaging in clinically feasible time.

In this paper we report the set-up and results of the Multimodal Brain Tumor Image Segmentation Benchmark (BRATS) organized in conjunction with the MICCAI 2012 and 2013 conferences. Twenty state-of-the-art tumor segmentation algorithms were applied to a set of 65 multi-contrast MR scans of low- and high-grade glioma patients-manually annotated by up to four raters-and to 65 comparable scans generated using tumor image simulation software. Quantitative evaluations revealed considerable disagreement between the human raters in segmenting various tumor sub-regions (Dice scores in the range 74%-85%), illustrating the difficulty of this task. We found that different algorithms worked best for different sub-regions (reaching performance comparable to human inter-rater variability), but that no single algorithm ranked in the top for all sub-regions simultaneously. Fusing several good algorithms using a hierarchical majority vote yielded segmentations that consistently ranked above all individual algorithms, indicating remaining opportunities for further methodological improvements. The BRATS image data and manual annotations continue to be publicly available through an online evaluation system as an ongoing benchmarking resource.