Classification of Colon Cancer through analysis of histopathology images using Transfer Learning
Mallela Siva Naga Raju, Battula Srinivasa Rao · 2022
The importance of histopathological image analysis in the early detection of colorectal cancer cannot be overstated. Because of the subjective nature of the evaluation, visual evaluation is time consuming and unreliable. Furthermore, the varying architectural and color characteristics of histological images make automated analysis extremely difficult. In the proposed work, we present a deep learning technique for distinguishing adenocarcinomas from healthy tissues and benign lesions by using Convolutional Neural Networks (CNNs). Although, a fully trained CNN on a large set of annotated Colorectal Cancer (CRC) samples yield high classification accuracy, it suffers from process tediousness. In our research, we used transfer learning techniques based on MobileNetV2 and InceptionResnetV2 models that have been pretrained on the LC25000 dataset. Transfer learning outperformed a fully pretrained CNN on CRC samples with 99.98% accuracy on the test dataset.