Deep learning on ascites cytology predicts platinum resistance in advanced ovarian cancer

A deep learning model analyzing pretreatment ascites cytology predicted platinum resistance risk with AUCs of 0.894 (internal), 0.863, and 0.828 (two external validation cohorts), outperforming KELIM score (AUC 0.619). External validation supports potential clinical accessibilit…

• Retrospective diagnostic-accuracy study in 438 patients with FIGO stage IIIB-IV epithelial ovarian cancer using a multi-scale deep-learning framework (OVCAP) on whole-slide images.
• Attention-guided review and single-cell RNA sequencing linked high-risk morphologic patterns (cytoplasmic vacuolization, interaction-rich aggregates) to lipid reprogramming and hypoxia signaling.
• Primary limitation: retrospective design; requires prospective validation to confirm clinical utility before first-line therapy.

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