Learned ultrasound features classify breast sentinel-node involvement
Source check: Source checked. Not independent scientific confirmation.
Finding: Image-pretrained convolutional models were fine-tuned on breast ultrasound images and combined with axillary ultrasound findings. In a retrospective second-hospital test of 395 women, the combined model classified pathology-referenced sentinel-node involvement with an AUC of 0.848. Avoided surgery was hypothetical, not an observed clinical benefit.
Technology: ImageNet-pretrained convolutional neural networks with ultrasound fine-tuning
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
- AI contribution
- Convolutional networks pre-trained on 1.28 million ImageNet images learned ultrasound features through fine-tuning; their predicted probabilities enabled the node-involvement evaluation.
- Test and result
- 937 patients from two hospitals: 542 for development and 395 for testing; 3,049 images. The reference was surgical lymph-node assessment. Radial and anti-radial image-model probabilities were fused.
- Limitation / still unknown
- Retrospective exclusions and ultrasound-visible lesion selection limit transfer. Images were processed individually without explicitly modeling their correlation. The separate sequential non-sentinel-node analysis has conditional denominators and is not represented by the quoted AUC. Supplementary architecture and threshold-selection details were not independently inspected; no reduced surgery or outcome benefit was measured.
Evidence sources and annotation context
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
Evidence sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC7519251/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review date: 21 Nov 2026. Review scope: Indexed primary clinical design, convolutional-model operation, reference evaluation and limitations; exact primary DOI/title/PMCID and fresh PubMed electronic date/notice metadata..
No follow-up recorded; this is not a claim that no follow-up exists. Follow-up review scope and date are not recorded. No source-supported relationships recorded.
Editorial assessment: Impact 2/5 · Validation 2/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.