Automated Detection of Wireless Capsule Endoscopy Polyp Abnormalities with Deep Transfer Learning and Support Vector Machines

Abstract

Polyp detection in wireless capsule endoscopy (WCE) is still an unsolved problem due to the large variation of polyps in terms of shape, color and size. There are two major problems hindering its improvement. First, traditional hand crafting approaches of WCE abnormalities’ detection has to be designed from scratch; it suffers from either a very time consuming process and/or a lack of exactitude. Second, WCE datasets acquisition still provides a challenge owing to the lack of large and publicly available annotated datasets. Recently, deep transfer learning has been widely used to transfer knowledge to medical images enabling the extraction of highly representative features. This paper investigates different architectures of pre-trained convolution neural networks (CNNs) from scratch (or network fine-tuning) for WCE polyp classification task. We compare the results with the state-of-art methods. The experiments consistently demonstrate that the use of a well-known DCNN architecture named Inception V3 with adequate fine-tuning outperform or, in the worst case, perform as a CNN trained from scratch. The last fully connected (fc) layer is connected to the support vector machine (SVM) classifier to obtain better accuracy. The methodology exceeds traditional handcrafting features extraction methods in terms of performance for WCE polyp abnormalities’ detection.