Deep Learning Perspective for Preliminary Detection and Classification of Ovarian Cancer

Histopathological images are analyzed by a deep convolutional neural network in this study for the purpose of detecting and classifying the four subtypes of ovarian cancer (DCNN). Ovarian cancer has the lowest 5-year survival rate of any gynecologic cancer but is also the fifth most frequent and the most aggressive kind of this disease. Ovarian epithelial carcinoma may manifest in four primary forms: serous, mucinous, endometroid, and clear cell. Computers are increasingly being used to analyze medical images for clues to the presence of illnesses including cancer, brain tumors, seizures, and Alzheimer's. In this research, we present and implement an enhanced DCNN-based architecture for distinguishing between healthy and cancerous cells. If cancerous, it might be classified as a subtype. The authors created an extra 24,742 histopathological photos from The Cancer Genome Atlas (TCGA-OV) set of 500 images available to the public. When more images were used during training, the KK-Net classification model's accuracy rose from 75% to 91%. Receiver Operating Characteristics: The performance of this model was examined using the area under the curve analysis. It was shown that the AUC-ROC curve had an average accuracy of 95%. We also evaluate the proposed model's performance to the state-of-the-art using four additional networks (GoogleNet, VGG-19, VGG-16, and AlexNet). The newly formed distinctive design may serve as a benchmark for predicting and diagnosing the illness, allowing pathologists to discover ovarian cancer in its early stages.