A liver-ultrasound classifier is evaluated in a South Indian community sample
Source check: Source checked. Not independent scientific confirmation.
Finding: A study evaluated a neural ultrasound classifier using expert-agreed fatty-liver grades in a rural South Indian cohort. On a 66-person internal holdout, AUC was 0.90 for moderate-to-severe steatosis. Performance declined on an external dataset with a different reference and target; the result does not establish clinical outcomes or universal transfer.
Technology: ImageNet-pretrained InceptionResNetV2 convolutional neural network
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
- AI contribution
- A pretrained neural image model was fine-tuned on liver-ultrasound labels and generated participant-level probabilities from video-frame predictions.
- Test and result
- Of 261 expert-review candidates, 219 had usable scans and agreement by two blinded radiologists. Training, validation and test splits were at participant level, and image probabilities were averaged within each participant at test. The 66-person internal holdout contained 15 moderate-to-severe cases and yielded AUC .90 (CI .77–1.00). Without additional fine-tuning, an external 55-person biopsy-referenced dataset yielded AUC .76 (CI .64–.89) for any steatosis, a different target.
- Limitation / still unknown
- This record represents the original clinical analysis, not pooled performance of the accompanying systematic review. Selecting only radiologist-agreed cases removes difficult disagreements; initial inter-reader kappa was .44. Reference grades are visual ultrasound judgments, not biopsy confirmation for the community holdout. Different reported thresholds are optimized by Youden index, with independent threshold locking not established. External testing changes both outcome definition and reference, and subsequent fine-tuning uses that external population. No patient benefit, independent replication or broad population calibration is demonstrated.
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/PMC12337209/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 21 Nov 2026. Review scope: Original community-analysis computational Methods, Table4 and primary article dates; systematic-review pooled findings kept separate..
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 1/5 · 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.