Introduction We investigated the impact of ascites and peritoneal metastatic (PM) lesion volumes, measured by deep learning-based algorithm, on survival outcomes in advanced epithelial ovarian cancer.

Methods Applying our newly developed deep learning-based auto-segmentation algorithm to pre-treatment computed tomography (CT) images obtained from 195 patients with advance-stage EOC, we measured volumes of ascites and PM lesions in the abdominal-pelvic cavity. Using the median values of ascites and PM lesion volumes as cut-off values, patients were divided into high- and low-volumetric ascites groups and high- and low-volumetric PM groups. Thereafter, survival outcomes were compared between the two groups.

Results Of the study population, 34.9% had FIGO stage IV disease and 78.5% had high-grade serous carcinoma. Complete cytoreduction was achieved in 56.4%. The median volumes of ascites and PM lesions were 714.5 cm³ and 341.1 cm³, respectively. The high-volumetric ascites group showed significantly worse OS than the low-volumetric ascites group (5-year PFS rate, 68.7% vs. 46.1%, P=0.08), but similar PFS. In multivariate analyses adjusting for clinicopathologic factors, high-volumetric ascites was identified as an independent poor prognostic factor for OS (aHR, 1.801; 95% CI, 1.147–2.828; P=0.011). Limited to a subgroup of patients who achieved complete cytoreduction (n=110), high-volumetric PM was associated with significantly worse OS (aHR, 2.231; 95% CI, 1.066–4.669; P=0.033).

Conclusion/Implications We successfully measured volume of ascites and PM lesions in the abdominal-pelvic cavity using the newly developed deep learning-based auto-segmentation algorithm. Our study results indicate that volumetric measurement of ascites and PM lesions might be novel prognostic factors for survival outcomes in patients with advanced-stage epithelial ovarian cancer.