Two-Stage Deep Learning Architecture for Pneumonia Detection and its Diagnosis in Chest Radiographs

摘要

Approximately two million pediatric deaths occur every year due to Pneumonia. Detection and diagnosis of Pneumoniaplays an important role in reducing these deaths. Chest radiography is one of the most commonly used modalities todetect pneumonia. In this paper, we propose a novel two-stage deep learning architecture to detect pneumonia andclassify its type in chest radiographs. This architecture contains one network to classify images as either normal orpneumonic, and another deep learning network to classify the type as either bacterial or viral. In this paper, we study andcompare the performance of various stage one networks such as AlexNet, ResNet, VGG16 and Inception-v3 fordetection of pneumonia. For these networks, we employ transfer learning to exploit the wealth of information availablefrom prior training. For the second stage, we find that transfer learning with these same networks tends to overfit thedata. For this reason we propose a simpler CNN architecture for classification of pneumonic chest radiographs and showthat it overcomes the overfitting problem. We further enhance the performance of our system in a novel way byincorporating lung segmentation using a U-Net architecture. We make use of a publicly available dataset comprising5856 images (1583 – Normal, 4273 – Pneumonic). Among the pneumonia patients, 2780 patients are identified asbacteria type and the rest belongs to virus category. We test our proposed algorithm(s) on a set of 624 images and weachieve an area under the receiver operating characteristic curve of 0.996 for pneumonia detection. We also achieve anaccuracy of 97.8% for classification of pneumonic chest radiographs thereby setting a new benchmark for both detectionand diagnosis. We believe the proposed two-stage classification of chest radiographs for pneumonia detection and itsdiagnosis would enhance the workflow of radiologists.