The Histological Diagnosis of Colonic Adenocarcinoma by Applying Partial Self Supervised Learning

BackgroundThe cancer of colon is one of the important cause of morbidity and mortality in adults. For the management of colonic carcinoma, the definitive diagnosis depends on the histological examination of biopsy specimens. With the development of whole slide imaging, the convolutional neural networks are being applied to diagnose colonic carcinoma by digital image analysis. AimThe main aim of the current study is to assess the application of deep learning for the histopathological diagnosis of colonic adenocarcinoma by analysing the digitized pathology images. Materials & MethodsThe images of colonic adenocarcinoma and non neoplastic colonic tissue have been acquired from the two datasets. The first dataset contains ten thousand images which were used to train and validate the convolutional neural network (CNN) architecture. From the second dataset (Colorectal Adenocarcinoma Gland (CRAG) Dataset) 40% of the images were used as a train set while 60% of the images were used as test dataset. Two histopathologists also evaluated these images. In this study, three variants of CNN (ResNet-18, ResNet-34 and ResNet-50) have been employed to evaluate the images. ResultsIn the present study, three CNN architectures(ResNet-18, ResNet-30, and ResNet-50) were applied for the classification of digitized images of colonic tissue. The accuracy (93.91%) of ResNet-50 was the highest which is followed by ResNet-30 and ResNet-18 with the accuracy of 93.04% each. ConclusionBased on the findings of the present study and analysis of previously reported series, the development of computer aided technology to evaluate the surgical specimens for the diagnosis of malignant tumors could provide a significant assistance to pathologists.