arXiv:2511.17828v1 Announce Type: new 
Abstract: Foundation models hold promise for specialized medical imaging tasks, though their effectiveness in breast imaging remains underexplored. This study leverages BiomedCLIP as a foundation model to address challenges in model generalization. BiomedCLIP was adapted for automated BI-RADS breast density classification using multi-modality mammographic data (synthesized 2D images, digital mammography, and digital breast tomosynthesis). Using 96,995 images, we compared single-modality (s2D only) and multi-modality training approaches, addressing class imbalance through weighted contrastive learning. Both approaches achieved similar accuracy (multi-modality: 0.74, single-modality: 0.73), with the multi-modality model offering broader applicability across different imaging modalities and higher AUC values consistently above 0.84 across BI-RADS categories. External validation on the RSNA and EMBED datasets showed strong generalization capabilities (AUC range: 0.80-0.93). GradCAM visualizations confirmed consistent and clinically relevant attention patterns, highlighting the models interpretability and robustness. This research underscores the potential of foundation models for breast imaging applications, paving the way for future extensions for diagnostic tasks.

أظهرت دراسة حديثة تكيف نموذج BiomedCLIP الأساسي لتصوير الثدي، مع التركيز على التصنيف الآلي لكثافة الثدي BI-RADS باستخدام مجموعة بيانات متنوعة تضم 96,995 صورة. قارن البحث بين أساليب التدريب ذات النمط الواحد والمتعدد، محققًا مستويات دقة مشابهة مع تسليط الضوء على قابلية تطبيق النموذج متعدد الأنماط وقدرته القوية على التعميم عبر أنماط التصوير المختلفة.

Un estudio reciente ha adaptado el modelo de fundación BiomedCLIP para la imagenología mamaria, centrándose en la clasificación automatizada de la densidad mamaria BI-RADS utilizando un conjunto de datos diverso de 96,995 imágenes. La investigación comparó enfoques de entrenamiento de modalidad única y multimodal, logrando niveles de precisión similares mientras destacaba la mayor aplicabilidad del modelo multimodal y sus fuertes capacidades de generalización a través de diferentes modalidades de imagenología.

Une étude récente a adapté le modèle de fondation BiomedCLIP pour l'imagerie mammaire, en se concentrant sur la classification automatisée de la densité mammaire BI-RADS à l'aide d'un ensemble de données diversifié de 96 995 images. La recherche a comparé les approches d'entraînement à modalité unique et multimodale, atteignant des niveaux de précision similaires tout en soulignant l'applicabilité plus large du modèle multimodal et ses fortes capacités de généralisation à travers différentes modalités d'imagerie.

A recent study has adapted the BiomedCLIP foundation model for breast imaging, focusing on automated BI-RADS breast density classification using a diverse dataset of 96,995 images. The research compared single-modality and multi-modality training approaches, achieving similar accuracy levels while highlighting the multi-modality model's broader applicability and strong generalization capabilities across different imaging modalities.

Toward explainable AI approaches for breast imaging: adapting foundation models to diverse populations

arXiv:2601.08193v1 Announce Type: new 
Abstract: Aggregating multi-site brain MRI data can enhance deep learning model training, but also introduces non-biological heterogeneity caused by site-specific variations (e.g., differences in scanner vendors, acquisition parameters, and imaging protocols) that can undermine generalizability. Recent retrospective MRI harmonization seeks to reduce such site effects by standardizing image style (e.g., intensity, contrast, noise patterns) while preserving anatomical content. However, existing methods often rely on limited paired traveling-subject data or fail to effectively disentangle style from anatomy. Furthermore, most current approaches address only single-sequence harmonization, restricting their use in real-world settings where multi-sequence MRI is routinely acquired. To this end, we introduce MMH, a unified framework for multi-site multi-sequence brain MRI harmonization that leverages biomedical semantic priors for sequence-aware style alignment. MMH operates in two stages: (1) a diffusion-based global harmonizer that maps MR images to a sequence-specific unified domain using style-agnostic gradient conditioning, and (2) a target-specific fine-tuner that adapts globally aligned images to desired target domains. A tri-planar attention BiomedCLIP encoder aggregates multi-view embeddings to characterize volumetric style information, allowing explicit disentanglement of image styles from anatomy without requiring paired data. Evaluations on 4,163 T1- and T2-weighted MRIs demonstrate MMH's superior harmonization over state-of-the-art methods in image feature clustering, voxel-level comparison, tissue segmentation, and downstream age and site classification.

تم تقديم إطار عمل جديد، MMH، من أجل توحيد تنسيق تصوير الرنين المغناطيسي للدماغ عبر مواقع متعددة وتسلسلات متعددة، حيث يتناول التحديات المتعلقة بالتباين غير البيولوجي في بيانات التصوير المغناطيسي الناتجة عن اختلافات محددة بالموقع. تهدف هذه الطريقة إلى توحيد نمط الصورة مع الحفاظ على المحتوى التشريحي، مما يعزز تدريب نماذج التعلم العميق.

Se ha introducido un nuevo marco, MMH, para la armonización unificada de la resonancia magnética cerebral multi-sitio y multi-secuencia, abordando los desafíos de la heterogeneidad no biológica en los datos de RM causados por variaciones específicas del sitio. Este método busca estandarizar el estilo de imagen mientras se preserva el contenido anatómico, mejorando así el entrenamiento de modelos de aprendizaje profundo.

Un nouveau cadre, MMH, a été introduit pour l'harmonisation unifiée des IRM cérébrales multi-sites et multi-séquences, abordant les défis de l'hétérogénéité non biologique dans les données IRM causée par des variations spécifiques aux sites. Cette méthode vise à standardiser le style d'image tout en préservant le contenu anatomique, améliorant ainsi l'entraînement des modèles d'apprentissage profond.

A new framework, MMH, has been introduced for unified multi-site multi-sequence brain MRI harmonization, addressing the challenges of non-biological heterogeneity in MRI data caused by site-specific variations. This method aims to standardize image style while preserving anatomical content, enhancing the training of deep learning models.

Unified Multi-Site Multi-Sequence Brain MRI Harmonization Enriched by Biomedical Semantic Style

One More Thing in AI – Your Shortcut to AI Mastery

Toward explainable AI approaches for breast imaging: adapting foundation models to diverse populations

Was this article worth reading? Share it

One More Thing in AI

LucidQuery AI

AIPortalX

Twofold Health

Attentive AI

LexiStock AI

Ready to build your own newsroom?