Vis-BUS AI-Assisted Breast Ultrasound: Diagnostic Performance and Reading Time

258 breast ultrasound examinations (129 malignant, 129 benign) were interpreted by six radiologists with and without AI assistance. Hyuksool Kwon published the study on figshare in March 2026. Vis-BUS AI assistance modestly increased the pooled AUROC from 0.921 to 0.953 and halved the median interpretation time from 6.0 to 3.0 seconds.

Use Cases

• Benchmarking AI-assisted diagnostic performance based on AUROC and AUPRC metrics
• Analyzing the impact of AI on radiologist interpretation efficiency based on recorded reading times
• Investigating AI performance for specific subgroups like dense breasts or small tumors mentioned in the results

Strengths

• Includes 258 examinations with a balanced set of 129 malignant and 129 benign lesions
• Compares diagnostic metrics (AUROC, accuracy, sensitivity, specificity) and interpretation times with statistical significance testing
• Provides results from a multi-reader study involving six radiologists

Limitations

• Row count and column-level documentation are absent; field semantics must be inferred after download
• Data is contained in a 126.5 KB DOCX file, suggesting a limited scope focused on summary results rather than raw data
• The retrospective study design may limit generalizability, as noted in the conclusion

Provenance

Source

figshare

Collection Method

Retrospective multi-reader study

Freshness

Last updated 2026-03-18 07:11:44