Classification of Gastrointestinal Images Based on Transfer Learning and Denoising Convolutional Neural Networks

Abstract

AbstractComputer-aided diagnosis systems have become a common approach in helping doctors to diagnose and recognise various diseases more quickly. Medical image classification which represents a major part of the computer-aided diagnosis process is a challenging field of research. Recently, pre-trained convolution neural networks (CNNs) along with image noise filters as a pre-processing tool have been successfully and widely used in endoscopic image classification for gastrointestinal disease detection. Herein, one of most well-known pre-trained CNNs, AlexNet, with no data augmentation and fine-tuned settings preceded by the denoising convolutional neural networks (DnCNNs) method as a pre-processing tool is proposed for endoscopic image classification to help with gastrointestinal disease detection. The denoising pre-processing stage is then combined with classification methods to carry out the eight classes of Kvasir dataset for endoscopic images. The experiment was implemented using one of the most well-known endoscopic medical images; Kvasir medical images shows that the performance evaluation of the proposed method has 90.17% classification accuracy and outperformed some of the similar state-of-the-art methods.KeywordsTransfer learningPre-trained neural networksMedical image classification