The study examined whether embeddings from four foundation models—Mammo-CLIP, HOPPR, MedImageInsight, and BiomedCLIP—could detect pre‑diagnostic changes in screening mammograms. Using 1,773 biopsied women and matched controls, the researchers measured the speed of movement along a data‑derived “cancer direction” in embedding space over successive screening intervals. They found that embeddings from clinically grounded models (Mammo‑CLIP, HOPPR, MedImageInsight) showed faster drift in malignant cases compared to controls, while the general biomedical model BiomedCLIP did not, indicating that foundation model embeddings can encode early tissue changes without task‑specific fine‑tuning.
Foundation model embeddings of screening mammograms may encode pre-diagnostic tissue change without task-specific adaptation. We tested whether embeddings move faster along a data-derived "cancer dire...
arXiv:2609.26443v1 Announce Type: new Abstract: Recent advances in Artificial Intelligence (AI)-powered Computer-Aided Diagnosis (CAD) systems have substantially improved breast cancer screening, dia...
arXiv:2608.24688v1 Announce Type: new Abstract: Precision oncology necessitates a longitudinal model of patient state that captures cancer evolution and treatment over time, integrating multimodal ob...
The paper introduces SEM‑HD, a framework that leverages longitudinal mammography history as privileged information during training to improve risk prediction while requiring only a single current exam at inference. By having a student model predict latent representations of past visits and using teacher supervision from actual longitudinal data, SEM‑HD preserves temporal modeling benefits without needing prior exams at deployment. Experiments on three cohorts and two backbone architectures show consistent gains in long‑horizon AUC and pAUC, especially in low false‑positive‑rate regions, and recover much of the performance gap to full‑history models.
arXiv:2607. 10358v1 Announce Type: cross Abstract: Foundation models are increasingly used as image feature extractors for mammography, but their robustness under external domain shift remains unclear.
The study evaluates whether mammography foundation models, pretrained for breast cancer tasks, can predict 5‑year major adverse cardiovascular events (MACE) in women using only screening mammograms. In a cohort of 22,497 women (500 MACE events), the models achieved AUROCs of 0.823 and 0.822, outperforming an age‑only baseline. The models also identified higher risk in patients with radiologist‑documented breast arterial calcifications, despite never being trained on that label.
arXiv:2607. 11343v1 Announce Type: cross Abstract: Accurate breast cancer risk prediction from screening mammography is critical for enabling personalized screening intervals and early detection.
arXiv:2607. 02768v1 Announce Type: cross Abstract: Pathologic complete response and tumor shrinkage measure whether breast cancer responds to neoadjuvant therapy, but not whether that response was structurally favorable, persistent, or hidden beneath volume loss.
arXiv:2606. 07590v1 Announce Type: cross Abstract: Pathology foundation models are pretrained on large streams of WSI-derived patches, while supervision during data construction is often slide-level, sparse, or heterogeneous.
arXiv:2607. 25497v1 Announce Type: cross Abstract: Pathology foundation models are approaching clinical deployment, yet remain vulnerable to systematic non-biological variation across centres.
The paper introduces TopKSigLIP, a vision‑language model tailored for mammography that tackles two key challenges: high‑resolution imaging and homogeneous radiology reports. It replaces standard CLIP training with a TopK‑Patch module that selects sparse high‑resolution patches likely to contain lesions, and a Sup‑sigmoid loss that uses soft labels from structured data instead of contrastive loss. TopKSigLIP outperforms existing open‑source mammography and general medical VLMs on zero‑shot tasks such as density assessment, BI‑RADS classification, finding subtyping, and cancer prediction, while also providing better lesion localization than Grad‑CAM.
arXiv:2609.14676v1 Announce Type: new Abstract: Purpose: To compare four multimodal large language models (MLLMs) with radiologists of varying expertise in breast density assessment, BI-RADS assessme...