46/#972 Prediction of platinum-based chemotherapy resistance in epithelial ovarian cancer using apparent diffusion coefficient of magnetic resonance image and machine learning algorithm. (4th December 2022)

• Record Type: Journal Article
• Authors: Lee, Sung-Jong; Woo, Jae Yeon; Song, Heekyoung; Cha, Jimin
• Is Part Of: International journal of gynecological cancer. Volume 32(2022)Supplement 3
• Page Start: A47
• Page End: A47
• Publication Date: 2022-12-04
• Subjects: Generative organs, Female -- Cancer -- Periodicals; 616.99465
• DOI: 10.1136/ijgc-2022-igcs.90
• Languages: English
• ISSNs: 1048-891X

Abstract

Abstract : Objectives: To investigate the role of a preoperative apparent diffusion coefficient (ADC) of magnetic resonance imaging (MRI) and machine learning for the prediction of platinum-based chemotherapy resistance in patients with epithelial ovarian cancer (EOC). Methods: The ADC of MRI was preoperatively evaluated on the largest solid portion of the ovarian mass on the axial MRI maps. All patients underwent platinum-based chemotherapy after cytoreductive surgery. Logistic regression and machine learning applications were used to investigate the role of the ADC and clinical factors for the prediction of platinum-based chemotherapy resistance in ovarian cancer. Results: Of the 168 patients, 97 had high-grade serous ovarian cancer (HGSOC) and 71 had non-HGSOC patients; 33 clear cell carcinoma, 18 mucinous carcinoma, 15 endometrioid carcinoma, 5 low-grade serous carcinoma. The patients were divided into the platinum-sensitive group (n=146) and the platinum-resistance group (n=22). The gradient boosting machine algorithm showed the highest accuracy in differentiating histologic types of ovarian cancers (accuracy: 0.91, AUC: 0.93). In the ROC curve, CA 125 and the ratio of solid to the total area were significantly associated with platinum-based chemotherapy resistance (AUC: 0.758, AUC: 0.687, respectively). The deep learning algorithm demonstrated increased accuracy (AUC: 0.814). In cox regression analysis, the area of the solid portion was significantly related to the resistance to chemotherapy (hazard ratio: 1.033, p=0.014). Conclusions: The ADC and area of the solid portion on MRI using machine learning can be helpful to predict histologic types and resistance of platinum-based chemotherapy in EOC.