Early Detection and Sub-type Classification of Ovarian Cancer with XGBoost and Explainable

Abstract

Ovarian cancer remains one of the most life-threatening gynecological malignancies due to its silent progression and lack of reliable early-stage symptoms. Conventional diagnostic tools such as CA-125 biomarker testing and transvaginal ultrasound (TVUS) often suffer from low sensitivity and specificity, leading to delayed detection and treatment inefficiencies. To address these limitations, this work proposes an Explainable Multimodal XGBoost Framework that integrates clinical biomarkers, radiomics, histopathology features, and gene-expression data for early prediction and subtype classification of ovarian cancer. The model performs two-stage analysis: binary classification (benign vs malignant) and multiclass classification of major histopathological subtypes such as serous, mucinous, endometrioid, clear cell, and others. Explainability is achieved using SHAP (Shapley Additive Explanations), enabling transparent and clinically trustworthy decision support. The system is implemented as a web-based Clinical Decision Support System (CDSS) with a FastAPI backend and React-based interface, offering real-time inference with low latency. Experimental evaluation demonstrates a ROC–AUC of 0.93 for binary classification and a macro F1-score of 0.88 for subtype analysis. The approach highlights significant potential to improve early diagnosis, support oncologists, and enhance precision oncology workflows.