A Review Based On Deep Learning Techniques Of Ovarian Cancer Detection

Abstract

Ovarian cancer remains one of the most lethal gynecological malignancies, primarily due to delayed diagnosis
and the disease’s histological heterogeneity. Recent advancements in artificial intelligence (AI) and deep learning (DL) have demonstrated significant promise in improving early detection and accurate classification of ovarian tumors through non-invasive
imaging modalities. This study synthesizes findings from four contemporary AI-based research approaches utilizing ultrasound
and multi-parametric magnetic resonance imaging (mpMRI) for early ovarian cancer diagnosis. A systematic review and meta analysis revealed that AI-enhanced ultrasound achieved pooled sensitivity and specificity rates of 81% and 92%, respectively. Another approach developed a DL model leveraging multi-sequence mpMRI data, which effectively distinguished high-grade serous carcinoma from clear cell carcinoma with an AUC of 91.62%. A deep learning radiomics nomogram (DLR Nomogram) derived from ultrasound images outperformed the traditional O-RADS classification, achieving AUCs as high as 0.985. Additionally, a multiclassification framework incorporating multiple DL models and explainable AI (XAI) techniques, including InceptionV3, achieved up to 97.96% accuracy, while enhancing model interpretability. Collectively, these AI-driven strategies demonstrate powerful potential for improving diagnostic accuracy, enabling precise subtype identification, and advancing personalized treatment planning in the early detection of ovarian cancer.