Deep learning using multi-phase CT achieves AUC 0.894 for gastric GIST risk stratification in external validation

In external validation for 2–5 cm gastric GISTs, a multi-phase CT deep learning model (MPFNet) achieved AUC 0.894 (95% CI 0.827–0.942) vs 0.831 for a clinical-radiomics model (P=0.091), with net reclassification improvement of 0.41. Internally, AUC was 0.935 vs 0.872 (P=0.016).

• MPFNet outperformed three radiologists (external AUC 0.894 vs 0.735–0.812).
• Decision curve analysis showed net benefit at threshold probabilities 0.30–0.50.
• Limitation: false negatives were predominantly intermediate-risk tumors; performance varied between scanners.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.