CT habitat analysis and deep learning for preoperative prediction of lymphovascular and perineural invasion in gastric cancer

A combined CT habitat and deep learning model predicted lymphovascular/perineural invasion in locally advanced gastric cancer with AUCs of 0.89, 0.87, and 0.82 in training, internal, and external validation cohorts in a dual-center retrospective study.

• Dual-center retrospective study including 611 patients (training n=374, internal n=161, external n=76).
• The deep learning signature and whole-tumor habitat signature showed the highest discrimination.
• Limited by retrospective design and small external validation cohort (n=76).

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