An Efficient and Lightweight Convolutional Neural Network for Carcinogenic Polyp Identification

The advent of Computer Vision and Deep Learning techniques has facilitated the disease identification process by retrieving helpful information from the associated data. Although Colorectal Carcinoma has been increasing tremendously, very few research works have been conducted so far to build an efficient and dedicated computational system for its early-stage detection. One of the ways to impede this infection is by recognising the polyps occurring in the colon walls. Previously, large-scale image recognition schemes, Conventional Data Mining algorithms and Deep Neural Networks were trained on RGB and Gray Scale images for this purpose. However, it is important to create an independent framework to obtain better accuracy and reduce the training cost of colon polyp spotting. In this research, we have designed a stacked Convolutional Neural Network (CNN) to classify two particular types of Colon Polyps responsible for Colorectal Cancer. The experiment of this network on a benchmark dataset has achieved the highest 97.6% accuracy in the classification of ‘Hyperplastic’ and ‘Adenomatous’ polyps. The comparison with the baseline models has perceived a significant reduction in the training time.