Machine Learning Based Common Radiologist-Level Pneumonia Detection on Chest X-rays

Abstract

For digital pathology, automatic recognition of different tissue types in histological images is important for diagnostic assistance and healthcare. Since histological images generally contain more than one tissue type, multi-class texture analysis plays a critical role to solve this problem. The chest X-ray is one of the most commonly accessible radiological examinations for screening and diagnosis of many lung diseases. This study examines the important statistical features including use the Convolutional Layer, ReLU Layer and Pooling Layer for detect pneumonia from Chest X-rays identification by using Convolutional Neural Network (CNN) and decision fusion of feature selection. Our algorithm, is a 144 layers Convolutional Neural Network (CNN) trained on Chest X-rays 14 diseases, currently the largest publicly available Chest X-rays dataset, containing over 100,000 frontal view Chest X-rays classification of histological images with 14 diseases. We detect all 14 diseases in Chest X-rays and achieve state of the art results on all 14 diseases. The average experimental results achieves high identification rate which is significantly superior to the existing known methods. In summary, the proposed method based on machine learning outperforms the techniques described in the literatures and achieve high classification accuracy rate at 80.90% for 144 layers of Convolutional Neural Network (CNN) which demonstrates promising applications for Chest X-rays classification of histological images.