A hybridized channel selection approach with deep convolutional neural network for effective ovarian cancer prediction in periodic acid‐Schiff‐stained images

Abstract

Summary In today's world, cancers are becoming a crucial warning in current medical applications where they show a significant part in the prognosis and appraisal of ovarian malignancies in histopathological imaging. Automated approaches to systematize the formation and categorization of cancers in periodic acid‐Schiff (PAS)‐stained images have recently been quickly growing in the area of digital pathology. However, the existing literature lacks computerized approaches to systematize the localization and categorization of cancers in PAS‐stained images. In this work, a new deep fully connected convolutional neural network (DFCNN) along with hybridized channel selection (HCS) strategy has been proposed to diagnose ovarian cancers accurately in PAS‐stained images. Primarily, the HCS strategy selects the optimal channel for managing the input dataset. Afterward, the DFCNN is employed to extract the influential features from the PAS‐stained images. The autoencoder modeling can be carried out to build the proposed DFCNN layers. The appropriate layer selection in the proposed model facilitates to obtaining accurate cancer classification with minimal misclassification errors. The performance results manifest that the proposed model achieves a superior accuracy of 99.22% when compared with existing CNN+softmax, CNN+decision tree, and CNN+radial basis function models.