Predictive Modeling of Ovarian Cancer Using Advanced Image Recognition and Machine Learning

Abstract

Ovarian cancer is a clinical dilemma due to its originally asymptomatic nature and then varied presentation. Early and accurate detection becomes an urgent goal in the course of the optimization of the patient's outcome. We present predictive modeling of the ovarian cancer subtypes by deep learning-based detection of the histopathology image analysis. The methodologically designed five-class dataset of Clear Cell Ovarian Carcinoma, Endometrioid, Serous Carcinoma, Mucinous Carcinoma and Non-Cancerous tissues was used. The recent convolution neural networks such as ResNet50, VGG16, VGG19, MobileNetv2 and InceptionV3 have been compared upon using the preprocessing steps such as image resizing, stain normalization, bilateral filtering and augmentation in order to improve the performance of the model. The performance was compared using the metrics such as accuracy, precision, recall, F1-score and area under the receiver operating characteristic curve (AUC). Visualization of the model's attention regions as well as interpretability has also been conducted using Gradient-weighted Class Activation Mapping (Grad-CAM) and Shapley Additive exPlanations (SHAP). The outcome has revealed the superiority of InceptionV3 in the classification task. The results show the evidence that deep models are capable of bringing in order to make the conventional diagnostic pipelines stronger in the scenario of the ovarian cancer and to achieve more accurate, scalable and explainable outcomes in computational pathology.