Description
Ovarian cancer remains one of the most aggressive and fatal gynecological cancers due to late-stage diagnosis and subtle early symptoms. This research paper introduces a new deep learning–based framework for the automated prediction and subtype classification of ovarian cancer using histopathological images.
The authors design and implement a custom Deep Convolutional Neural Network architecture inspired by AlexNet but significantly modified to improve medical image classification performance. The model incorporates eight convolutional layers, four max-pooling layers, and four fully connected layers. Exponential Linear Unit activation functions and Mean Squared Error loss are applied to enhance learning stability and accuracy.
A comprehensive dataset sourced from the National Cancer Institute’s Genomic Data Commons is used. Image augmentation techniques including rotation, zooming, flipping, and brightness enhancement expand the dataset from 500 to 24,742 images. Experimental results confirm that augmentation plays a critical role in improving performance, raising classification accuracy from 72 percent to 83.93 percent.
This paper is highly relevant for researchers, data scientists, medical imaging specialists, and healthcare professionals working in cancer diagnostics, deep learning, and computer-aided diagnosis systems. It also serves as a strong reference for implementing custom CNN architectures in biomedical applications.