🤖 AI Summary
This study addresses the joint optimization of exam-level malignancy prediction and lesion localization in mammography. The authors propose a unified multi-task DETR-based framework that leverages shared features to simultaneously perform image-level classification and region-level localization. For the first time, they systematically evaluate the performance of various modern vision backbones—including ConvNeXtV2, DINOv3, and MambaVision—on this dual objective. Experimental results demonstrate that advanced backbones substantially outperform conventional ResNet architectures: on the OPTIMAM dataset, ConvNeXtV2 achieves 97.96% AUC and 25.08% mAP@.5, while on SGM1k, DINOv3 attains 90.97% AUC and 27.04% mAP@.5, underscoring the critical influence of backbone architecture on multi-task performance in mammographic analysis.
📝 Abstract
Joint exam-level prediction and candidate-region localization may improve the usefulness of AI support in mammography. We study this setting using a multi-task DETR framework, where shared representations support both image-level malignancy prediction and lesion localization, and evaluate its performance on OPTIMAM and a biopsy-confirmed SGM1k cohort. Across both datasets, modern backbones consistently outperformed older ResNet-style features, with ConvNeXtV2 and DINOv3 giving the strongest overall results, whereas MambaVision was less competitive. On OPTIMAM, ConvNeXtV2 achieved the best overall performance, reaching 97.96% AUC, 99.89% sensitivity, 25.08% mAP@.5, and 74.38% recall@.25. On SGM1k, DINOv3 gave the strongest overall results, with 90.97% AUC, 86.28% sensitivity, 82.00% specificity, 27.04% mAP@.5, and 77.32% recall@.25. These findings suggest that backbone quality is a critical factor in effective multi-task mammography, with ConvNeXtV2 emerging as a particularly strong and well-matched CNN backbone for mammography in this framework.