BAC-JEPA: Label-Efficient Breast Arterial Calcification Segmentation via Synthetic Mammography-Guided Supervision

2026-06-20

AI Summary

This study addresses the challenge of segmenting breast arterial calcifications (BAC)—a significant biomarker of cardiovascular risk—whose clinical deployment is hindered by reliance on costly pixel-level expert annotations. The authors propose a label-efficient segmentation framework that, for the first time, enables annotation-free transfer learning via procedurally generated synthetic mammograms, which combine real mammographic backgrounds with precisely rendered calcification masks. The approach integrates a self-supervised Vision Transformer encoder, a high-resolution convolutional decoder, hard negative example augmentation, and multi-view inference. Evaluated on the BacSeg dataset, the model achieves an image-level AUROC of 0.8719 and a Dice coefficient of 0.6357 on the synthetic validation set, with quadruple-view inference per image completed in under 214 milliseconds.

Abstract

Breast arterial calcification (BAC) on screening mammograms is an emerging cardiovascular risk biomarker, but quantitative use requires reproducible segmentation and expert pixel-level labels are costly. We present BAC-JEPA, a label-efficient segmentation framework trained on procedurally generated arterial calcification inserted into real mammographic backgrounds with exact masks. Candidate backgrounds were selected from model-screened mammograms with low predicted BAC response; the generator samples arterial structure, disease burden, radiographic appearance, and hard-negative distractors including nonarterial calcifications and metallic objects. Synthetic masks are paired with mammography self-supervised Vision Transformer encoders and a high-resolution convolutional decoder to produce full-resolution segmentation maps. The study used 75,472 mammography studies from 34,956 patients for background selection and representation learning, trained on synthetic images from 10,000 backgrounds, selected checkpoints with 1,000 development backgrounds, and evaluated transfer on all 1,000 human-labeled BacSeg synthetic 2D mammograms. On held-out synthetic validation data, the larger backbone achieved IoU 0.5325 and Dice 0.6357. On BacSeg, image-level classification from segmentation probability maps reached AUROC 0.8719, with 0.8547 for the smaller backbone. Four-view inference required 110.68--213.63 ms on an RTX 5090 GPU, and severe-preset synthetic image generation averaged 2.7071 s per image on a multicore workstation. These results indicate that BAC-specific synthetic supervision can produce useful image-level transfer without human pixel-level training masks, while expert-reviewed real-mammogram segmentation remains necessary for clinical validation and calibration.

Problem

• Breast arterial calcification
• Label-efficient segmentation
• Mammography
• Pixel-level annotation
• Cardiovascular risk biomarker

Innovation

• synthetic supervision
• label-efficient segmentation
• procedural generation
• self-supervised Vision Transformer
• breast arterial calcification