2.5D multi-instance learning on MRI distinguishes SCLC from NSCLC brain metastases

A 2.5D multi-instance learning MRI model differentiated small-cell from non-small-cell lung cancer brain metastases with an external validation AUC of 0.826, comparable to a clinical model (AUC 0.826, p=1.000).

• Retrospective study of 366 patients from four centers; 2.5D multi-instance learning model trained on biparametric MRI (contrast-enhanced T1 and FLAIR).
• On pooled external test set (n=82), model AUC 0.826 vs clinical model AUC 0.826 (p=1.000), radiomics AUC 0.696, and 3D-DL AUC 0.664.
• The model did not significantly outperform the clinical model; larger prospective studies are needed (Evidence Level 3).

Evidence grade: 3

Automated summary

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