Endoscopic Artefact Detection with Deep Learning

Abstract

Object detection refers to distinguishing the objects in the images or videos from the part that is not interested in and judging whether there is a target. If there is a target, it determines the position of the target and determine the category which it belongs to. The difficulties of the task are the extraction and identification of candidates for the region to be tested. Therefore, there are three main steps in the whole procedure. Firstly, select some parts of the graph as candidate regions. Secondly, extract the features from the candidate regions. Finally, train the classifiers. It is a very important research topic in the computer vision field and plays an important role in our real life such as medical image analysis, network data mining, UAV navigation, remote sensing image analysis, national defense system and so on. Apart from computer scientists’ strong interests, object detection also attracts attentions from researchers and scientists from medical field. At present, most of the medical image processing and analyzing works are done by doctors manually. However, the analyzing result and performance of doctors are highly related to the doctors’ individual physical and mental condition, which are very destabilizing, labor-consuming and time-costing. Object detection methods based on computer vision are effective tools to help doctors to optimize their work through locating and identifying objects on the medical images and also increase the accuracy of diagnosis results. This report will introduce and summarize the work I have done in my final year project: applying three deep learning models on endoscopic artefact detection datasets to do object detection and analyzing and comparing their results. 1. Faster-RCNN model with ResNet101 for Endoscopic Artefact Detection dataset. 2. Libra-RCNN model with ResNet101 for Endoscopic Artefact Detection dataset. 3. Guided-Anchoring RCNN model with ResNet101for Endoscopic Artefact Detection dataset.