Anomaly Detection on Chest X-Rays with Deep Learning

Pneumonia is a common, and sometimes life threatening, disease. Chest x-ray is one of the main tools for detecting lung disease. The problem is that detecting lung diseases from x-ray images is a complicated task, due to number of causes: Sometimes there are overlaps with other diagnoses, sometimes the symptoms of the disease do not appear in the image, and often the image is vague. In addition, the demand for x-ray diagnosis is huge, and radiologist availability cannot meet this demand. Therefore, in this project, we tried to detect the binary existence of lung diseases (i.e., whether or not there is an illness) from x-ray images, using deep learning methods, in particular using unsupervised learning methods. We used a common model in unsupervised learning named autoencoder (and a variety of variations of this model) and trained it on x-ray images of healthy people. Examination of the results shows that this model is not currently suitable for detecting lung diseases, but it can be used for anomaly detection in such images.

Anomaly Detection on Chest X-Rays with Deep Learning
Anomaly Detection on Chest X-Rays with Deep Learning