Multi-class Classification of Ovarian Cancer from Histopathological Images using Deep Learning - VGG-16
Kokila. R. Kasture, Bharati B. Sayankar, Pravin N. Matte
Abstract
Ovarian cancer (OC) is the primary gynecological malignancy. Precise prediction and subtype classification of OC (Serous, Mucinous, Endometrioid, Clear cell) are vital within the diverse diagnosis. Pathologists currently rely on Computer Aided Diagnosis (CAD) to correctly work out the diagnosis. Our study demonstrates a pre-trained Deep Convolutional Neural Networks (DCNN) architecture Visual Geometry Group – 16 (VGG16) to automatically detect, predict & classify the subtypes of ovarian cancers from histopathological images.VGG-16 comprises 16 layers (13 convolution layers, three fully-connected layers, five max-pooling layers, and one softmax layer). Initially, we trained the model with 500 images, 100 of each class, and achieved an accuracy of 50%. We then augmented the dataset of 500 images by image rotation, flipping, enhancing, and zooming to generate 24742 images, multiplying our input dataset. After training the model with this augmented dataset, the classification accuracy increased from 50% to 84.64%. The proposed work is the first attempt to use VGG16 to detect, predict and classify subtypes of ovarian cancer from histopathological images.