Colonoscopy Polyp Detection Using Bi-Directional Conv-LSTM U-Net with Densely Connected Convolution

Several researchers have focused in recent years on improving the efficiency of abdominal diagnostics by segmenting colonoscopy images with machine learning techniques. Previously, colonoscopy images were manually segmented by experts in this field. This eventually became time-consuming work that was prone to human error. Advances in technology, such as increased computer power and the availability of libraries for manipulating colonoscopy images, enabled automated segmentation. In recent year, deep learning networks are using in medical segmentation field due to its versatility, high performance, high generalization capacity. Recently, new heights of effectiveness have been achieved in the process of medical image segmentation carried out by deep learning model. The process of medical image segmentation has been effectively improved by the application of deep learning models such as U-NET, RS-NET, and RS-NET++. In this study, we apply the benefits of U-Net, Bi-directional Conv-LSTM, and method of dense convolution. We applied these to the Kvasir-SEG and CVC-Clinic DB datasets and achieved the 0.92 and 0.93 dice coefficient respectively.