Xavier Lladó

Diabetic foot ulcers (DFUs) are a severe complication of diabetes mellitus, associated with high rates of amputation and mortality. Automated detection and segmentation from clinical images could support timely intervention and remote monitoring, yet existing approaches lack systematic benchmarks and rarely exploit the synergy between lesion localization and delineation. In this work, we present a comprehensive and insight-driven study of detection and segmentation strategies for DFU analysis using datasets from the Diabetic Foot Ulcer Challenge (DFUC). We benchmark four state-of-the-art object detection frameworks alongside an in-depth optimization study, and evaluate four global segmentation models. Furthermore, we propose a hybrid detection-guided segmentation pipeline leveraging a domain-adapted foundation model. Our analysis reveals that, in detection-guided pipelines, lesion coverage (recall) is more critical than standard detection metrics such as mAP, and that the quality of spatial prompts plays a key role in downstream segmentation performance. Ensemble strategies are further explored to improve robustness and generalization. The optimized detection ensemble ranks first on the DFUC2020 challenge leaderboard, and the proposed segmentation pipeline achieves Dice scores of up to 0.721 and 0.786 on two independent test sets with distinct acquisition conditions, demonstrating consistent generalization beyond the training distribution. These results provide design guidelines and a comprehensive benchmark for automated DFU analysis.

Machine learning models for medical image analysis are vulnerable to hidden confounders, which can compromise generalization and clinical reliability. Existing detection strategies typically require explicit knowledge or labels of the confounder, which are often unavailable. In this work, we propose an ensemble-based framework to detect potential confounder-driven learning without explicitly defining the confounders, but only which samples might be affected. Our approach leverages the variability of model performance across ensembles to identify signatures of shortcut learning. Shortcut learning occurs when a model uses non-robust features or correlations rather than learning the true underlying task, and it is often observed when confounders are present. We generate controlled dataset variants with increasing confounding levels and analyze distributions of AUC (area under the ROC curve) scores across training, validation, and test splits, revealing converging performance and reduced variance as confounding intensifies. We validate our method on two clinically relevant tasks, diabetic retinopathy detection from retinal fundus images and tumor detection from brain MRI slices. Then, we further demonstrate its practical utility on another dataset and image modality with a stroke reperfusion prediction task with suspected hidden confounders. This work provides a practical, data-driven diagnostic tool to flag potential confounding and support the reliability assessment of machine learning models in medical imaging.

The human brain exhibits heterogeneous, region-specific structural changes across the lifespan, modulated by neurodegenerative diseases. Structural MRI enables early detection, but robust modelling of brain evolution is hindered by scarce longitudinal data. We propose BrainST, a diffusion-based framework for controlled synthesis, anatomical transformation, and longitudinal prediction of T1-weighted brain MRI, trained solely on cross-sectional data. Image generation is conditioned on volumetric measurements of 18 brain regions of interest (ROIs), with localised anatomical control enforced via a conditioning alignment penalty. ROI volumes can be manually specified or predicted from demographic and cognitive features. Evaluation on an external longitudinal dataset demonstrates realistic cross-sectional and longitudinal MRI generation, precise region-specific transformations, and plausible simulations of aging and Alzheimer’s disease. BrainST enables counterfactual image generation with localised control, providing a practical tool for studying brain structural patterns when longitudinal data are limited.

Accurate fetal brain tissue segmentation and biometric analysis are essential for studying brain development in utero. The FeTA Challenge 2024 advanced automated fetal brain MRI analysis by introducing biometry prediction as a new task alongside tissue segmentation. For the first time, our diverse multi-centric test set included data from a new low-field (0.55T) MRI dataset. Evaluation metrics were also expanded to include the topology-specific Euler characteristic difference (ED). Sixteen teams submitted segmentation methods, most of which performed consistently across both high- and low-field scans. However, longitudinal trends indicate that segmentation accuracy may be reaching a plateau, with results now approaching inter-rater variability. The ED metric uncovered topological differences that were missed by conventional metrics, while the low-field dataset achieved the highest segmentation scores, highlighting the potential of affordable imaging systems when paired with high-quality reconstruction. Seven teams participated in the biometry task, but most methods failed to outperform a simple baseline that predicted measurements based solely on gestational age, underscoring the challenge of extracting reliable biometric estimates from image data alone. Domain shift analysis identified image quality as the most significant factor affecting model generalization, with super-resolution pipelines also playing a substantial role. Other factors, such as gestational age, pathology, and acquisition site, had smaller, though still measurable, effects. Overall, FeTA 2024 offers a comprehensive benchmark for multi-class segmentation and biometry estimation in fetal brain MRI, underscoring the need for data-centric approaches, improved topological evaluation, and greater dataset diversity to enable clinically robust and generalizable AI tools.