Convolution Neural Network in Polyp Detection and Classification

  • Nagesh B S, N P Kavya


The objective of endoscopic examination is to find unusual sores also, decide further treatment from the acquired data. The goal of endoscopy is to distinguish malignant polyps from benign polyps at the early stages. In any case, the technique produces an assortment of non-useful casings and sores can be missed because of helpless video quality. Particularly when breaking down whole endoscopic recordings made by a novice medical practitioner, computer aided diagnosis is essential to avoid said drawbacks. Computer assisted Endoscopy examination includes classification issues like polyp identification. Artificial intelligence (AI) based applications have changed a few businesses and are broadly utilized in different customer items and administrations. In medicine domain, AI is still in its beginnings and can be utilized for classification and identification. AI and machine learning mentions to computer based intelligence in which the calculation, in light of the input crude information, investigates highlights in a different dataset without specifically being modified and conveys an expected classification result. An innovation approach for constructing a classifier ready to recognize polyps in images of endoscopy video is proposed in this article. The ensemble approach utilizing the image processing strategies for pre-processing and profound learning is proposed considering the convolution neural network. The neural network was first presented during the early 1950s.   Nevertheless, in light of the low processing and lacking datasets accessible around then, neural networks are not that much recognized and utilized. Ongoing investigations in this innovation estimates well for clinical and medical services applications, particularly in endoscopic imaging. This article gives viewpoints on the set of experiences, improvement, applications, and difficulties of profound learning innovation. The proposed neural network model can is trained and tested on the gastrointestinal endoscopy dataset which includes 7894 images. In building the classifier, the proposed work utilizes outcomes of investigations over convolution neural organizations utilized clinical observation that permits to follow the projected approach to deal with planning the design of a convolutional neural organization adjusted to a given undertaking. By summing up the highlights of multiple fruitful prototypes, this method built up a methodology towards making an efficient effective convolutional neural organization. According to proposed research, the neural network can be distributed into multiple blocks that substitute to empower forming the utmost productive engineering. The proposed classifier is based on convolutional neural networks, where the capacities proposed in the traditional methodology are overridden. The outcomes  suggests planned utilization of  created method for constructing grouping models meant for medical diagnosis as well as intended for common issues in the field of computer vision depending on the sample input data.

Keywords- Video Mining , Endoscopy, Convolution Neural Network, Polyps, Artificial Intelligence