Download PDFOpen PDF in browser

Chest Disease Detection from Human X-Ray Scans Using Deep Learning

EasyChair Preprint 2557

11 pagesDate: February 5, 2020

Abstract

Millions of chest X-rays produced world-wide are currently analyzed almost entirely visually on a scan-by-scan basis. This requires a relatively high degree of skill and concentration, and is time-consuming, expensive, prone to operator bias (data distortion or wrong interpretation), and unable to exploit the invaluable informatics contained in such large-scale data. Errors and delay in these diagnostic methods still contribute to a large number of patient deaths in hospitals, making these errors one of the largest causes of death along with heart disease and cancer. Moreover, due to the complexity of these scans, it is challenging even for radiologists to differentiate various diseases on them, resulting in the shortage of expert radiologists, particularly in rural areas who are competent to read chest radiographs. Therefore, it is of utmost significance to design and implement automated algorithms for computer-aided diagnosis of diseases on chest radiography. 

Deep learning has transformed healthcare. It’s being used extensively to diagnose cancer, pneumonia, hernia and other diseases. Deep learning is more accurate and faster at diagnosis than real doctors. Automation of X-ray analysis can prevent a lot of mishaps, speed up diagnosis and reveal new patterns thus aiding in medical research. Hence we develop a deep learning model using Deep Convolutional Neural Networks (DCNN) architecture which can predict various chest diseases like Pneumonia, Pneumothorax, Atelectasis, Effusion etc with significant accuracy (>80%) and provide other insights about the analysis performed by generating heat maps and other visualizations. The model will also be able to localize the pathology by generating Class Activation Maps(CAM). We will start first with Pneumothorax owing to the availability of good dataset.

Keyphrases: Class Activation Map, Deep Convolutional Neural Network, Receiver Operating Characteristics

BibTeX entry
BibTeX does not have the right entry for preprints. This is a hack for producing the correct reference:
@booklet{EasyChair:2557,
  author    = {Ravi Soni and Rohan Singh and Shadib Shah and Shivang Malik},
  title     = {Chest Disease Detection from Human X-Ray Scans Using Deep Learning},
  howpublished = {EasyChair Preprint 2557},
  year      = {EasyChair, 2020}}
Download PDFOpen PDF in browser