DIAGENS releases CT model covering 509 chest and abdominal findings

DIAGENS, a Hong Kong-listed medical imaging AI company, has introduced a CT model that identifies 509 diseases and imaging findings in chest and abdominal scans, and published benchmark results for iMed-6.0, its latest imaging foundation model, according to a press release.

The CT model, built on the iMed-6.0 Aegis model, covers both non-contrast and contrast-enhanced CT of the chest and abdomen. In the company's internal validation on nearly 40,000 CT exam and report pairs from several hospitals, it reached a macro-averaged AUC of 0.855 in zero-shot classification across all 509 targets, and 0.930 on 249 key targets that appear in public benchmarks.

On MedBench, the medical benchmark from Shanghai AI Laboratory, iMed-6.0 Thoth and Aegis took first and second place in the multimodal API evaluation with scores of 64.4 and 61.7. Thoth also scored 66.2% on the MMMU test, above the previous best published result of 65.9%. DIAGENS says it has R&D partnerships with 99 hospitals and lets clinicians train their own models without code.