AI Science Tracker
EN

Learned ultrasound analysis identifies heel-texture differences between clinical groups

University of Shanghai for Science and Technology; Shanghai University of Medicine and Health Sciences; Fudan University; Shanghai University of Traditional Chinese Medicine; Shanghai Institute of Technology · 5 Nov 2026 · medicine

Discovery

Learned ultrasound analysis identifies heel-texture differences between clinical groups

Source check: Source checked. Not independent scientific confirmation.

Finding: Learned segmentation and texture classifiers identified heel-pad ultrasound differences between 76 participants with diabetes and 47 controls. Recurrent wavelet-correlation features contributed to group separation, but the groups also differed in age and sex without adjustment. These are image-level group associations, not diabetes-specific tissue mechanisms or validated patient-level diagnosis.

Technology: H-SAM segmentation with LoRA; XGBoost and other classifiers; SHAP interpretation

Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.

AI contribution
H-SAM learned segmentation produced layer masks; learned classifiers and feature-attribution analyses identified texture descriptors that consistently separated the sampled clinical groups.
Test and result
Single-center study of 123 participants and 289 heel-pad ultrasound images. H-SAM segmented two tissue layers; wavelet/texture features underwent training-only statistical screening and grouped feature selection. Seven classifiers shared participant-disjoint partitions. The final XGBoost model evaluated 45 image-derived vectors from 18 held-out participants, with AUC 0.9407. Six features recurred across classifiers; macrochamber wavelet-correlation features showed consistent model contribution patterns.
Limitation / still unknown
Patients were older than controls (median 61 versus 49 years), and sex distributions differed; no adjusted analyses isolated diabetes effects. Multiple image vectors per participant define the reported AUC, not independent participant-level diagnosis. Single device/center, segmentation uncertainty and incomplete diabetes/PAD characterization limit interpretation. No histological, biomechanical or causal tissue mechanism, clinical benefit or independent cohort replication was established.
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/PMC13736012/

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: Read actual primary H-SAM training, participant-level partition, feature-selection/tuning Methods, final image-vector Results, recurrent-feature/SHAP findings and demographic limitations. Original canonical primary online date November 5 is distinguished from index-filled November 1. Writing-assistance disclosures are not counted as the enabling AI role. Exact primary identity/date and fresh primary PubMed correction/retraction relationships checked..

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.

Open in the full catalog for filters, charts and related discoveries. Browse medicine discoveries.

Related catalog entries

Links use documented relationships where available, plus shared catalog context. A shared topic or model family does not establish a scientific dependency.