Anatomic Landmarks Detection by Deep Learning in Colonoscopy
摘要
International medical recommendations are indicating the landmark positions that have to be reached for a valid colonoscopy examination. In this preliminary study we compared several pre-trained deep learning neural structures, given the task of retraining for identification in real-time of four out of nine such important positions. We have improved the training dataset of colonoscopy frames three times to finally get successful results. Our trials were made using a Jetson Xavier NX developing microsystem.