Stergios CHristodoulidis

Joseph Boyd, Irène Villa, Marie-Christine Mathieu et al.

In whole slide imaging, commonly used staining techniques based on hematoxylin and eosin (H&E) and immunohistochemistry (IHC) stains accentuate different aspects of the tissue landscape. In the case of detecting metastases, IHC provides a distinct readout that is readily interpretable by pathologists. IHC, however, is a more expensive approach and not available at all medical centers. Virtually generating IHC images from H&E using deep neural networks thus becomes an attractive alternative. Deep generative models such as CycleGANs learn a semantically-consistent mapping between two image domains, while emulating the textural properties of each domain. They are therefore a suitable choice for stain transfer applications. However, they remain fully unsupervised, and possess no mechanism for enforcing biological consistency in stain transfer. In this paper, we propose an extension to CycleGANs in the form of a region of interest discriminator. This allows the CycleGAN to learn from unpaired datasets where, in addition, there is a partial annotation of objects for which one wishes to enforce consistency. We present a use case on whole slide images, where an IHC stain provides an experimentally generated signal for metastatic cells. We demonstrate the superiority of our approach over prior art in stain transfer on histopathology tiles over two datasets. Our code and model are available at https://github.com/jcboyd/miccai2022-roigan.

Leo Fillioux, Joseph Boyd, Maria Vakalopoulou et al.

Multiple instance learning is an ideal mode of analysis for histopathology data, where vast whole slide images are typically annotated with a single global label. In such cases, a whole slide image is modelled as a collection of tissue patches to be aggregated and classified. Common models for performing this classification include recurrent neural networks and transformers. Although powerful compression algorithms, such as deep pre-trained neural networks, are used to reduce the dimensionality of each patch, the sequences arising from whole slide images remain excessively long, routinely containing tens of thousands of patches. Structured state space models are an emerging alternative for sequence modelling, specifically designed for the efficient modelling of long sequences. These models invoke an optimal projection of an input sequence into memory units that compress the entire sequence. In this paper, we propose the use of state space models as a multiple instance learner to a variety of problems in digital pathology. Across experiments in metastasis detection, cancer subtyping, mutation classification, and multitask learning, we demonstrate the competitiveness of this new class of models with existing state of the art approaches. Our code is available at https://github.com/MICS-Lab/s4_digital_pathology.

Loïc Le Bescond, Maria Vakalopoulou, Stergios Christodoulidis et al.

The deployment of machine learning solutions in real-world scenarios often involves addressing the challenge of out-of-distribution (OOD) detection. While significant efforts have been devoted to OOD detection in classical supervised settings, the context of weakly supervised learning, particularly the Multiple Instance Learning (MIL) framework, remains under-explored. In this study, we tackle this challenge by adapting post-hoc OOD detection methods to the MIL setting while introducing a novel benchmark specifically designed to assess OOD detection performance in weakly supervised scenarios. Across extensive experiments based on diverse public datasets, KNN emerges as the best-performing method overall. However, it exhibits significant shortcomings on some datasets, emphasizing the complexity of this under-explored and challenging topic. Our findings shed light on the complex nature of OOD detection under the MIL framework, emphasizing the importance of developing novel, robust, and reliable methods that can generalize effectively in a weakly supervised context. The code for the paper is available here: https://github.com/loic-lb/OOD_MIL.

Leo Fillioux, Enzo Ferrante, Paul-Henry Cournède et al.

Large foundation models have emerged in the last years and are pushing performance boundaries for a variety of tasks. Training or even finetuning such models demands vast datasets and computational resources, which are often scarce and costly. Adaptation methods provide a computationally efficient solution to address these limitations by allowing such models to be finetuned on small amounts of data and computing power. This is achieved by appending new trainable modules to frozen backbones with only a fraction of the trainable parameters and fitting only these modules on novel tasks. Recently, the VeRA adapter was shown to excel in parameter-efficient adaptations by utilizing a pair of frozen random low-rank matrices shared across all layers. In this paper, we propose PVeRA, a probabilistic version of the VeRA adapter, which modifies the low-rank matrices of VeRA in a probabilistic manner. This modification naturally allows handling inherent ambiguities in the input and allows for different sampling configurations during training and testing. A comprehensive evaluation was performed on the VTAB-1k benchmark and seven adapters, with PVeRA outperforming VeRA and other adapters. Our code for training models with PVeRA and benchmarking all adapters is available https://github.com/leofillioux/pvera.

Andreas Lolos, Theofilos Christodoulou, Aris L. Moustakas et al.

In computational pathology, weak supervision has become the standard for deep learning due to the gigapixel scale of WSIs and the scarcity of pixel-level annotations, with Multiple Instance Learning (MIL) established as the principal framework for slide-level model training. In this paper, we introduce a novel setting for MIL methods, inspired by proceedings in Neural Partial Differential Equation (PDE) Solvers. Instead of relying on complex attention-based aggregation, we propose an efficient, aggregator-agnostic framework that removes the complexity of correlation learning from the MIL aggregator. CAPRMIL produces rich context-aware patch embeddings that promote effective correlation learning on downstream tasks. By projecting patch features -- extracted using a frozen patch encoder -- into a small set of global context/morphology-aware tokens and utilizing multi-head self-attention, CAPRMIL injects global context with linear computational complexity with respect to the bag size. Paired with a simple Mean MIL aggregator, CAPRMIL matches state-of-the-art slide-level performance across multiple public pathology benchmarks, while reducing the total number of trainable parameters by 48%-92.8% versus SOTA MILs, lowering FLOPs during inference by 52%-99%, and ranking among the best models on GPU memory efficiency and training time. Our results indicate that learning rich, context-aware instance representations before aggregation is an effective and scalable alternative to complex pooling for whole-slide analysis. Our code is available at https://github.com/mandlos/CAPRMIL

Leo Fillioux, Emilie Gontran, Jérôme Cartry et al.

Over the last ten years, Patient-Derived Organoids (PDOs) emerged as the most reliable technology to generate ex-vivo tumor avatars. PDOs retain the main characteristics of their original tumor, making them a system of choice for pre-clinical and clinical studies. In particular, PDOs are attracting interest in the field of Functional Precision Medicine (FPM), which is based upon an ex-vivo drug test in which living tumor cells (such as PDOs) from a specific patient are exposed to a panel of anti-cancer drugs. Currently, the Adenosine Triphosphate (ATP) based cell viability assay is the gold standard test to assess the sensitivity of PDOs to drugs. The readout is measured at the end of the assay from a global PDO population and therefore does not capture single PDO responses and does not provide time resolution of drug effect. To this end, in this study, we explore for the first time the use of powerful large foundation models for the automatic processing of PDO data. In particular, we propose a novel imaging-based high-throughput screening method to assess real-time drug efficacy from a time-lapse microscopy video of PDOs. The recently proposed SAM algorithm for segmentation and DINOv2 model are adapted in a comprehensive pipeline for processing PDO microscopy frames. Moreover, an attention mechanism is proposed for fusing temporal and spatial features in a multiple instance learning setting to predict ATP. We report better results than other non-time-resolved methods, indicating that the temporality of data is an important factor for the prediction of ATP. Extensive ablations shed light on optimizing the experimental setting and automating the prediction both in real-time and for forecasting.

Badr-Eddine Marani, Julio Silva-Rodriguez, Ismail Ben Ayed et al.

Deep neural networks have achieved remarkable success across a variety of tasks, yet they often suffer from unreliable probability estimates. As a result, they can be overconfident in their predictions. Conformal Prediction (CP) offers a principled framework for uncertainty quantification, yielding prediction sets with rigorous coverage guarantees. Existing conformal training methods optimize for overall set size, but shaping the prediction sets in a class-conditional manner is not straightforward and typically requires prior knowledge of the data distribution. In this work, we introduce Class Adaptive Conformal Training (CaCT), which formulates conformal training as an augmented Lagrangian optimization problem that adaptively learns to shape prediction sets class-conditionally without making any distributional assumptions. Experiments on multiple benchmark datasets, including standard and long-tailed image recognition as well as text classification, demonstrate that CaCT consistently outperforms prior conformal training methods, producing significantly smaller and more informative prediction sets while maintaining the desired coverage guarantees.

Leo Fillioux, Omprakash Chakraborty, Ismail Ben Ayed et al.

With the increasing adoption of vision-language models (VLMs) in critical decision-making systems such as healthcare or autonomous driving, the calibration of their uncertainty estimates becomes paramount. Yet, this dimension has been largely underexplored in the VLM test-time prompt-tuning (TPT) literature, which has predominantly focused on improving their discriminative performance. Recent state-of-the-art advocates for enforcing full orthogonality over pairs of text prompt embeddings to enhance separability, and therefore calibration. Nevertheless, as we theoretically show in this work, the inherent gradients from fully orthogonal constraints will strongly push semantically related classes away, ultimately making the model overconfident. Based on our findings, we propose Semantic Orthogonal Calibration (SoC), a Huber-based regularizer that enforces smooth prototype separation while preserving semantic proximity, thereby improving calibration compared to prior orthogonality-based approaches. Across a comprehensive empirical validation, we demonstrate that SoC consistently improves calibration performance, while also maintaining competitive discriminative capabilities.

Badr-Eddine Marani, Mohamed Hanini, Nihitha Malayarukil et al.

The proliferation of machine learning models in critical decision making processes has underscored the need for bias discovery and mitigation strategies. Identifying the reasons behind a biased system is not straightforward, since in many occasions they are associated with hidden spurious correlations which are not easy to spot. Standard approaches rely on bias audits performed by analyzing model performance in pre-defined subgroups of data samples, usually characterized by common attributes like gender or ethnicity when it comes to people, or other specific attributes defining semantically coherent groups of images. However, it is not always possible to know a-priori the specific attributes defining the failure modes of visual recognition systems. Recent approaches propose to discover these groups by leveraging large vision language models, which enable the extraction of cross-modal embeddings and the generation of textual descriptions to characterize the subgroups where a certain model is underperforming. In this work, we argue that incorporating visual explanations (e.g. heatmaps generated via GradCAM or other approaches) can boost the performance of such bias discovery and mitigation frameworks. To this end, we introduce Visually Grounded Bias Discovery and Mitigation (ViG-Bias), a simple yet effective technique which can be integrated to a variety of existing frameworks to improve both, discovery and mitigation performance. Our comprehensive evaluation shows that incorporating visual explanations enhances existing techniques like DOMINO, FACTS and Bias-to-Text, across several challenging datasets, including CelebA, Waterbirds, and NICO++.

Cameron Dahan, Stergios Christodoulidis, Maria Vakalopoulou et al.

In the clinical setting of histopathology, whole-slide image (WSI) artifacts frequently arise, distorting regions of interest, and having a pernicious impact on WSI analysis. Image-to-image translation networks such as CycleGANs are in principle capable of learning an artifact removal function from unpaired data. However, we identify a surjection problem with artifact removal, and propose an weakly-supervised extension to CycleGAN to address this. We assemble a pan-cancer dataset comprising artifact and clean tiles from the TCGA database. Promising results highlight the soundness of our method.

Nicolás Gaggion, Maria J. Ledesma-Carbayo, Stergios Christodoulidis et al.

Graph-based medical image segmentation represents anatomical structures using boundary graphs, providing fixed-topology landmarks and inherent population-level correspondences. However, their clinical adoption has been hindered by a major requirement: training datasets with manually annotated landmarks that maintain point-to-point correspondences across patients rarely exist in practice. We introduce Mask-HybridGNet, a framework that trains graph-based models directly using standard pixel-wise masks, eliminating the need for manual landmark annotations. Our approach aligns variable-length ground truth boundaries with fixed-length landmark predictions by combining Chamfer distance supervision and edge-based regularization to ensure local smoothness and regular landmark distribution, further refined via differentiable rasterization. A significant emergent property of this framework is that predicted landmark positions become consistently associated with specific anatomical locations across patients without explicit correspondence supervision. This implicit atlas learning enables temporal tracking, cross-slice reconstruction, and morphological population analyses. Beyond direct segmentation, Mask-HybridGNet can extract correspondences from existing segmentation masks, allowing it to generate stable anatomical atlases from any high-quality pixel-based model. Experiments across chest radiography, cardiac ultrasound, cardiac MRI, and fetal imaging demonstrate that our model achieves competitive results against state-of-the-art pixel-based methods, while ensuring anatomical plausibility by enforcing boundary connectivity through a fixed graph adjacency matrix. This framework leverages the vast availability of standard segmentation masks to build structured models that maintain topological integrity and provide implicit correspondences.

David Restrepo, Ira Ktena, Maria Vakalopoulou et al.

Vision-language models (VLMs) are increasingly proposed for clinical decision support, yet their reliability in real-world scenarios that require integrating both visual and textual context from medical records remains poorly characterized. This paper identifies three failure modes: (1) modality over-reliance on text over images, (2) spurious reliance on irrelevant clinical history, and (3) prompt sensitivity across semantically equivalent inputs. We evaluate a diverse set of general-domain and medically-tuned open and closed VLMs on chest x-ray tasks using MIMIC-CXR. By systematically manipulating image-text alignment, clinical history, and prompt formulations, we found that VLM decisions are dominated by the text modality, even when visual evidence is available. Moreover, we observed that VLMs are heavily influenced by irrelevant reports, while minor prompt changes can reverse correct image-based predictions. Our findings underscore the need for explicit safeguards and stress-testing before considering the use of these models in clinical practice.

Varun Belagali, Saarthak Kapse, Pierre Marza et al.

The interpretation of small tiles in large whole slide images (WSI) often needs a larger image context. We introduce TICON, a transformer-based tile representation contextualizer that produces rich, contextualized embeddings for ''any'' application in computational pathology. Standard tile encoder-based pipelines, which extract embeddings of tiles stripped from their context, fail to model the rich slide-level information essential for both local and global tasks. Furthermore, different tile-encoders excel at different downstream tasks. Therefore, a unified model is needed to contextualize embeddings derived from ''any'' tile-level foundation model. TICON addresses this need with a single, shared encoder, pretrained using a masked modeling objective to simultaneously unify and contextualize representations from diverse tile-level pathology foundation models. Our experiments demonstrate that TICON-contextualized embeddings significantly improve performance across many different tasks, establishing new state-of-the-art results on tile-level benchmarks (i.e., HEST-Bench, THUNDER, CATCH) and slide-level benchmarks (i.e., Patho-Bench). Finally, we pretrain an aggregator on TICON to form a slide-level foundation model, using only 11K WSIs, outperforming SoTA slide-level foundation models pretrained with up to 350K WSIs.

David Restrepo, Miguel L Martins, Chenwei Wu et al.

Vision-Language Models (VLMs) exhibit a characteristic "cone effect" in which nonlinear encoders map embeddings into highly concentrated regions of the representation space, contributing to cross-modal separation known as the modality gap. While this phenomenon has been widely observed, its practical impact on supervised multimodal learning -particularly in medical domains- remains unclear. In this work, we introduce a lightweight post-hoc mechanism that keeps pretrained VLM encoders frozen while continuously controlling cross-modal separation through a single hyperparameter {λ}. This enables systematic analysis of how the modality gap affects downstream multimodal performance without expensive retraining. We evaluate generalist (CLIP, SigLIP) and medically specialized (BioMedCLIP, MedSigLIP) models across diverse medical and natural datasets in a supervised multimodal settings. Results consistently show that reducing excessive modality gap improves downstream performance, with medical datasets exhibiting stronger sensitivity to gap modulation; however, fully collapsing the gap is not always optimal, and intermediate, task-dependent separation yields the best results. These findings position the modality gap as a tunable property of multimodal representations rather than a quantity that should be universally minimized.

Pablo Morales-Álvarez, Stergios Christodoulidis, Maria Vakalopoulou et al.

The emergence of large pre-trained vision-language models (VLMs) represents a paradigm shift in machine learning, with unprecedented results in a broad span of visual recognition tasks. CLIP, one of the most popular VLMs, has exhibited remarkable zero-shot and transfer learning capabilities in classification. To transfer CLIP to downstream tasks, adapters constitute a parameter-efficient approach that avoids backpropagation through the large model (unlike related prompt learning methods). However, CLIP adapters have been developed to target discriminative performance, and the quality of their uncertainty estimates has been overlooked. In this work we show that the discriminative performance of state-of-the-art CLIP adapters does not always correlate with their uncertainty estimation capabilities, which are essential for a safe deployment in real-world scenarios. We also demonstrate that one of such adapters is obtained through MAP inference from a more general probabilistic framework. Based on this observation we introduce BayesAdapter, which leverages Bayesian inference to estimate a full probability distribution instead of a single point, better capturing the variability inherent in the parameter space. In a comprehensive empirical evaluation we show that our approach obtains high quality uncertainty estimates in the predictions, standing out in calibration and selective classification. Our code will be publicly available upon acceptance of the paper.

Pierre Marza, Leo Fillioux, Sofiène Boutaj et al.

Progress in a research field can be hard to assess, in particular when many concurrent methods are proposed in a short period of time. This is the case in digital pathology, where many foundation models have been released recently to serve as feature extractors for tile-level images, being used in a variety of downstream tasks, both for tile- and slide-level problems. Benchmarking available methods then becomes paramount to get a clearer view of the research landscape. In particular, in critical domains such as healthcare, a benchmark should not only focus on evaluating downstream performance, but also provide insights about the main differences between methods, and importantly, further consider uncertainty and robustness to ensure a reliable usage of proposed models. For these reasons, we introduce THUNDER, a tile-level benchmark for digital pathology foundation models, allowing for efficient comparison of many models on diverse datasets with a series of downstream tasks, studying their feature spaces and assessing the robustness and uncertainty of predictions informed by their embeddings. THUNDER is a fast, easy-to-use, dynamic benchmark that can already support a large variety of state-of-the-art foundation, as well as local user-defined models for direct tile-based comparison. In this paper, we provide a comprehensive comparison of 23 foundation models on 16 different datasets covering diverse tasks, feature analysis, and robustness. The code for THUNDER is publicly available at https://github.com/MICS-Lab/thunder.

Daphne Tsolissou, Theofanis Ganitidis, Konstantinos Mitsis et al.

Reliable risk assessment for carotid atheromatous disease remains a major clinical challenge, as it requires integrating diverse clinical and imaging information in a manner that is transparent and interpretable to clinicians. This study investigates the potential of state-of-the-art and recent large vision-language models (LVLMs) for multimodal carotid plaque assessment by integrating ultrasound imaging (USI) with structured clinical, demographic, laboratory, and protein biomarker data. A framework that simulates realistic diagnostic scenarios through interview-style question sequences is proposed, comparing a range of open-source LVLMs, including both general-purpose and medically tuned models. Zero-shot experiments reveal that even if they are very powerful, not all LVLMs can accurately identify imaging modality and anatomy, while all of them perform poorly in accurate risk classification. To address this limitation, LLaVa-NeXT-Vicuna is adapted to the ultrasound domain using low-rank adaptation (LoRA), resulting in substantial improvements in stroke risk stratification. The integration of multimodal tabular data in the form of text further enhances specificity and balanced accuracy, yielding competitive performance compared to prior convolutional neural network (CNN) baselines trained on the same dataset. Our findings highlight both the promise and limitations of LVLMs in ultrasound-based cardiovascular risk prediction, underscoring the importance of multimodal integration, model calibration, and domain adaptation for clinical translation.

Sofiène Boutaj, Marin Scalbert, Pierre Marza et al.

Whole slide image (WSI) analysis in digital pathology presents unique challenges due to the gigapixel resolution of WSIs and the scarcity of dense supervision signals. While Multiple Instance Learning (MIL) is a natural fit for slide-level tasks, training robust models requires large and diverse datasets. Even though image augmentation techniques could be utilized to increase data variability and reduce overfitting, implementing them effectively is not a trivial task. Traditional patch-level augmentation is prohibitively expensive due to the large number of patches extracted from each WSI, and existing feature-level augmentation methods lack control over transformation semantics. We introduce HistAug, a fast and efficient generative model for controllable augmentations in the latent space for digital pathology. By conditioning on explicit patch-level transformations (e.g., hue, erosion), HistAug generates realistic augmented embeddings while preserving initial semantic information. Our method allows the processing of a large number of patches in a single forward pass efficiently, while at the same time consistently improving MIL model performance. Experiments across multiple slide-level tasks and diverse organs show that HistAug outperforms existing methods, particularly in low-data regimes. Ablation studies confirm the benefits of learned transformations over noise-based perturbations and highlight the importance of uniform WSI-wise augmentation. Code is available at https://github.com/MICS-Lab/HistAug.

David Restrepo, Ira Ktena, Maria Vakalopoulou et al.

Clinical decision-making relies on the integrated analysis of medical images and the associated clinical reports. While Vision-Language Models (VLMs) can offer a unified framework for such tasks, they can exhibit strong biases toward one modality, frequently overlooking critical visual cues in favor of textual information. In this work, we introduce Selective Modality Shifting (SMS), a perturbation-based approach to quantify a model's reliance on each modality in binary classification tasks. By systematically swapping images or text between samples with opposing labels, we expose modality-specific biases. We assess six open-source VLMs-four generalist models and two fine-tuned for medical data-on two medical imaging datasets with distinct modalities: MIMIC-CXR (chest X-ray) and FairVLMed (scanning laser ophthalmoscopy). By assessing model performance and the calibration of every model in both unperturbed and perturbed settings, we reveal a marked dependency on text input, which persists despite the presence of complementary visual information. We also perform a qualitative attention-based analysis which further confirms that image content is often overshadowed by text details. Our findings highlight the importance of designing and evaluating multimodal medical models that genuinely integrate visual and textual cues, rather than relying on single-modality signals.

Théo Sourget, David Restrepo, Céline Hudelot et al.

Motivated by the strong performance of CLIP-based models in natural image-text domains, recent efforts have adapted these architectures to medical tasks, particularly in radiology, where large paired datasets of images and reports, such as chest X-rays, are available. While these models have shown encouraging results in terms of accuracy and discriminative performance, their fairness and robustness in the different clinical tasks remain largely underexplored. In this study, we extensively evaluate six widely used CLIP-based models on chest X-ray classification using three publicly available datasets: MIMIC-CXR, NIH-CXR14, and NEATX. We assess the models fairness across six conditions and patient subgroups based on age, sex, and race. Additionally, we assess the robustness to shortcut learning by evaluating performance on pneumothorax cases with and without chest drains. Our results indicate performance gaps between patients of different ages, but more equitable results for the other attributes. Moreover, all models exhibit lower performance on images without chest drains, suggesting reliance on spurious correlations. We further complement the performance analysis with a study of the embeddings generated by the models. While the sensitive attributes could be classified from the embeddings, we do not see such patterns using PCA, showing the limitations of these visualisation techniques when assessing models. Our code is available at https://github.com/TheoSourget/clip_cxr_fairness

Andreas Lolos, Stergios Christodoulidis, Maria Vakalopoulou et al.

Multiple Instance Learning (MIL) offers a natural solution for settings where only coarse, bag-level labels are available, without having access to instance-level annotations. This is usually the case in digital pathology, which consists of gigapixel sized images. While deterministic attention-based MIL approaches achieve strong bag-level performance, they often overlook the uncertainty inherent in instance relevance. In this paper, we address the lack of uncertainty quantification in instance-level attention scores by introducing \textbf{SGPMIL}, a new probabilistic attention-based MIL framework grounded in Sparse Gaussian Processes (SGP). By learning a posterior distribution over attention scores, SGPMIL enables principled uncertainty estimation, resulting in more reliable and calibrated instance relevance maps. Our approach not only preserves competitive bag-level performance but also significantly improves the quality and interpretability of instance-level predictions under uncertainty. SGPMIL extends prior work by introducing feature scaling in the SGP predictive mean function, leading to faster training, improved efficiency, and enhanced instance-level performance. Extensive experiments on multiple well-established digital pathology datasets highlight the effectiveness of our approach across both bag- and instance-level evaluations. Our code will be made publicly available.

Joseph Boyd, Mykola Liashuha, Eric Deutsch et al.

The examination of histopathology images is considered to be the gold standard for the diagnosis and stratification of cancer patients. A key challenge in the analysis of such images is their size, which can run into the gigapixels and can require tedious screening by clinicians. With the recent advances in computational medicine, automatic tools have been proposed to assist clinicians in their everyday practice. Such tools typically process these large images by slicing them into tiles that can then be encoded and utilized for different clinical models. In this study, we propose a novel generative framework that can learn powerful representations for such tiles by learning to plausibly expand their visual field. In particular, we developed a progressively grown generative model with the objective of visual field expansion. Thus trained, our model learns to generate different tissue types with fine details, while simultaneously learning powerful representations that can be used for different clinical endpoints, all in a self-supervised way. To evaluate the performance of our model, we conducted classification experiments on CAMELYON17 and CRC benchmark datasets, comparing favorably to other self-supervised and pre-trained strategies that are commonly used in digital pathology. Our code is available at https://github.com/jcboyd/cdpath21-gan.

Leo Fillioux, Julio Silva-Rodríguez, Ismail Ben Ayed et al.

Recent advances in self-supervision and contrastive learning have brought the performance of foundation models to unprecedented levels in a variety of tasks. Fueled by this progress, these models are becoming the prevailing approach for a wide array of real-world vision problems, including risk-sensitive and high-stakes applications. However, ensuring safe deployment in these scenarios requires a more comprehensive understanding of their uncertainty modeling capabilities, which has been barely explored. In this work, we delve into the behaviour of vision and vision-language foundation models under Conformal Prediction (CP), a statistical framework that provides theoretical guarantees of marginal coverage of the true class. Across extensive experiments including popular vision classification benchmarks, well-known foundation vision models, and three CP methods, our findings reveal that foundation models are well-suited for conformalization procedures, particularly those integrating Vision Transformers. We also show that calibrating the confidence predictions of these models, a popular strategy to improve their uncertainty quantification, actually leads to efficiency degradation of the conformal set on adaptive CP methods. Furthermore, few-shot adaptation of Vision-Language Models (VLMs) to downstream tasks, whose popularity is surging, enhances conformal scores compared to zero-shot predictions. Last, our empirical study exposes APS as particularly promising in the context of vision foundation models, as it does not violate the marginal coverage guarantees across multiple challenging, yet realistic scenarios.

Varun Belagali, Pierre Marza, Srikar Yellapragada et al.

Learning dense correspondences, critical for application such as video label propagation, is hindered by tedious and unscalable manual annotation. Self-supervised methods address this by using a cross-view pretext task, often modeled with a masked autoencoder, where a masked target view is reconstructed from an anchor view. However, acquiring effective training data remains a challenge - collecting diverse video datasets is difficult and costly, while simple image crops lack necessary pose variations. This paper introduces CDG-MAE, a novel MAE-based self-supervised method that uses diverse synthetic views generated from static images via an image-conditioned diffusion model. These generated views exhibit substantial changes in pose and perspective, providing a rich training signal that overcomes the limitations of video and crop-based anchors. We present a quantitative method to evaluate local and global consistency of generated images, discussing their use for cross-view self-supervised pretraining. Furthermore, we enhance the standard single-anchor MAE setting to a multi-anchor strategy to effectively modulate the difficulty of pretext task. CDG-MAE significantly outperforms state-of-the-art MAE methods reliant only on images and substantially narrows the performance gap to video-based approaches.

Bhakti Baheti, Satrajit Chakrabarty, Hamed Akbari et al.

Registration of longitudinal brain MRI scans containing pathologies is challenging due to dramatic changes in tissue appearance. Although there has been progress in developing general-purpose medical image registration techniques, they have not yet attained the requisite precision and reliability for this task, highlighting its inherent complexity. Here we describe the Brain Tumor Sequence Registration (BraTS-Reg) challenge, as the first public benchmark environment for deformable registration algorithms focusing on estimating correspondences between pre-operative and follow-up scans of the same patient diagnosed with a diffuse brain glioma. The BraTS-Reg data comprise de-identified multi-institutional multi-parametric MRI (mpMRI) scans, curated for size and resolution according to a canonical anatomical template, and divided into training, validation, and testing sets. Clinical experts annotated ground truth (GT) landmark points of anatomical locations distinct across the temporal domain. Quantitative evaluation and ranking were based on the Median Euclidean Error (MEE), Robustness, and the determinant of the Jacobian of the displacement field. The top-ranked methodologies yielded similar performance across all evaluation metrics and shared several methodological commonalities, including pre-alignment, deep neural networks, inverse consistency analysis, and test-time instance optimization per-case basis as a post-processing step. The top-ranked method attained the MEE at or below that of the inter-rater variability for approximately 60% of the evaluated landmarks, underscoring the scope for further accuracy and robustness improvements, especially relative to human experts. The aim of BraTS-Reg is to continue to serve as an active resource for research, with the data and online evaluation tools accessible at https://bratsreg.github.io/.

Théo Estienne, Maria Vakalopoulou, Stergios Christodoulidis et al.

Explainability of deep neural networks is one of the most challenging and interesting problems in the field. In this study, we investigate the topic focusing on the interpretability of deep learning-based registration methods. In particular, with the appropriate model architecture and using a simple linear projection, we decompose the encoding space, generating a new basis, and we empirically show that this basis captures various decomposed anatomically aware geometrical transformations. We perform experiments using two different datasets focusing on lungs and hippocampus MRI. We show that such an approach can decompose the highly convoluted latent spaces of registration pipelines in an orthogonal space with several interesting properties. We hope that this work could shed some light on a better understanding of deep learning-based registration methods.

Mihir Sahasrabudhe, Stergios Christodoulidis, Roberto Salgado et al.

Segmentation and accurate localization of nuclei in histopathological images is a very challenging problem, with most existing approaches adopting a supervised strategy. These methods usually rely on manual annotations that require a lot of time and effort from medical experts. In this study, we present a self-supervised approach for segmentation of nuclei for whole slide histopathology images. Our method works on the assumption that the size and texture of nuclei can determine the magnification at which a patch is extracted. We show that the identification of the magnification level for tiles can generate a preliminary self-supervision signal to locate nuclei. We further show that by appropriately constraining our model it is possible to retrieve meaningful segmentation maps as an auxiliary output to the primary magnification identification task. Our experiments show that with standard post-processing, our method can outperform other unsupervised nuclei segmentation approaches and report similar performance with supervised ones on the publicly available MoNuSeg dataset. Our code and models are available online to facilitate further research.

Guillaume Chassagnon, Maria Vakalopoulou, Enzo Battistella et al.

Chest computed tomography (CT) is widely used for the management of Coronavirus disease 2019 (COVID-19) pneumonia because of its availability and rapidity. The standard of reference for confirming COVID-19 relies on microbiological tests but these tests might not be available in an emergency setting and their results are not immediately available, contrary to CT. In addition to its role for early diagnosis, CT has a prognostic role by allowing visually evaluating the extent of COVID-19 lung abnormalities. The objective of this study is to address prediction of short-term outcomes, especially need for mechanical ventilation. In this multi-centric study, we propose an end-to-end artificial intelligence solution for automatic quantification and prognosis assessment by combining automatic CT delineation of lung disease meeting performance of experts and data-driven identification of biomarkers for its prognosis. AI-driven combination of variables with CT-based biomarkers offers perspectives for optimal patient management given the shortage of intensive care beds and ventilators.

Ya Lu, Thomai Stathopoulou, Maria F. Vasiloglou et al.

Regular monitoring of nutrient intake in hospitalised patients plays a critical role in reducing the risk of disease-related malnutrition. Although several methods to estimate nutrient intake have been developed, there is still a clear demand for a more reliable and fully automated technique, as this could improve data accuracy and reduce both the burden on participants and health costs. In this paper, we propose a novel system based on artificial intelligence (AI) to accurately estimate nutrient intake, by simply processing RGB Depth (RGB-D) image pairs captured before and after meal consumption. The system includes a novel multi-task contextual network for food segmentation, a few-shot learning-based classifier built by limited training samples for food recognition, and an algorithm for 3D surface construction. This allows sequential food segmentation, recognition, and estimation of the consumed food volume, permitting fully automatic estimation of the nutrient intake for each meal. For the development and evaluation of the system, a dedicated new database containing images and nutrient recipes of 322 meals is assembled, coupled to data annotation using innovative strategies. Experimental results demonstrate that the estimated nutrient intake is highly correlated (> 0.91) to the ground truth and shows very small mean relative errors (< 20%), outperforming existing techniques proposed for nutrient intake assessment.

Matthias Fontanellaz, Stergios Christodoulidis, Stavroula Mougiakakou

Direct computer vision based-nutrient content estimation is a demanding task, due to deformation and occlusions of ingredients, as well as high intra-class and low inter-class variability between meal classes. In order to tackle these issues, we propose a system for recipe retrieval from images. The recipe information can subsequently be used to estimate the nutrient content of the meal. In this study, we utilize the multi-modal Recipe1M dataset, which contains over 1 million recipes accompanied by over 13 million images. The proposed model can operate as a first step in an automatic pipeline for the estimation of nutrition content by supporting hints related to ingredient and instruction. Through self-attention, our model can directly process raw recipe text, making the upstream instruction sentence embedding process redundant and thus reducing training time, while providing desirable retrieval results. Furthermore, we propose the use of an ingredient attention mechanism, in order to gain insight into which instructions, parts of instructions or single instruction words are of importance for processing a single ingredient within a certain recipe. Attention-based recipe text encoding contributes to solving the issue of high intra-class/low inter-class variability by focusing on preparation steps specific to the meal. The experimental results demonstrate the potential of such a system for recipe retrieval from images. A comparison with respect to two baseline methods is also presented.

Ya Lu, Thomai Stathopoulou, Maria F. Vasiloglou et al.

Regular nutrient intake monitoring in hospitalised patients plays a critical role in reducing the risk of disease-related malnutrition (DRM). Although several methods to estimate nutrient intake have been developed, there is still a clear demand for a more reliable and fully automated technique, as this could improve the data accuracy and reduce both the participant burden and the health costs. In this paper, we propose a novel system based on artificial intelligence to accurately estimate nutrient intake, by simply processing RGB depth image pairs captured before and after a meal consumption. For the development and evaluation of the system, a dedicated and new database of images and recipes of 322 meals was assembled, coupled to data annotation using innovative strategies. With this database, a system was developed that employed a novel multi-task neural network and an algorithm for 3D surface construction. This allowed sequential semantic food segmentation and estimation of the volume of the consumed food, and permitted fully automatic estimation of nutrient intake for each food type with a 15% estimation error.

Stergios Christodoulidis, Mihir Sahasrabudhe, Maria Vakalopoulou et al.

Image registration and in particular deformable registration methods are pillars of medical imaging. Inspired by the recent advances in deep learning, we propose in this paper, a novel convolutional neural network architecture that couples linear and deformable registration within a unified architecture endowed with near real-time performance. Our framework is modular with respect to the global transformation component, as well as with respect to the similarity function while it guarantees smooth displacement fields. We evaluate the performance of our network on the challenging problem of MRI lung registration, and demonstrate superior performance with respect to state of the art elastic registration methods. The proposed deformation (between inspiration & expiration) was considered within a clinically relevant task of interstitial lung disease (ILD) classification and showed promising results.

Marios Anthimopoulos, Stergios Christodoulidis, Lukas Ebner et al.

Early and accurate diagnosis of interstitial lung diseases (ILDs) is crucial for making treatment decisions, but can be challenging even for experienced radiologists. The diagnostic procedure is based on the detection and recognition of the different ILD pathologies in thoracic CT scans, yet their manifestation often appears similar. In this study, we propose the use of a deep purely convolutional neural network for the semantic segmentation of ILD patterns, as the basic component of a computer aided diagnosis (CAD) system for ILDs. The proposed CNN, which consists of convolutional layers with dilated filters, takes as input a lung CT image of arbitrary size and outputs the corresponding label map. We trained and tested the network on a dataset of 172 sparsely annotated CT scans, within a cross-validation scheme. The training was performed in an end-to-end and semi-supervised fashion, utilizing both labeled and non-labeled image regions. The experimental results show significant performance improvement with respect to the state of the art.

Stergios Christodoulidis, Marios Anthimopoulos, Lukas Ebner et al.

Early diagnosis of interstitial lung diseases is crucial for their treatment, but even experienced physicians find it difficult, as their clinical manifestations are similar. In order to assist with the diagnosis, computer-aided diagnosis (CAD) systems have been developed. These commonly rely on a fixed scale classifier that scans CT images, recognizes textural lung patterns and generates a map of pathologies. In a previous study, we proposed a method for classifying lung tissue patterns using a deep convolutional neural network (CNN), with an architecture designed for the specific problem. In this study, we present an improved method for training the proposed network by transferring knowledge from the similar domain of general texture classification. Six publicly available texture databases are used to pretrain networks with the proposed architecture, which are then fine-tuned on the lung tissue data. The resulting CNNs are combined in an ensemble and their fused knowledge is compressed back to a network with the original architecture. The proposed approach resulted in an absolute increase of about 2% in the performance of the proposed CNN. The results demonstrate the potential of transfer learning in the field of medical image analysis, indicate the textural nature of the problem and show that the method used for training a network can be as important as designing its architecture.