ABSTRACT
Chest radiography is commonly used in annual health screenings to determine the health of the lungs. As a result, developing a smart system to assist clinicians or radiologists in automatically detecting possible malformations in radiographs would be preferable. This develops a new anomaly identifier method which relies on an “autoencoder” that delivers the reestablished version of the input besides the uncertainty prediction using pixel. Higher uncertainty is frequently seen in regular areas or boundaries, where reconstruction errors are relatively larger, but not at happening regions where abnormalities are present. As a result, a natural measurement for detecting abnormalities such as Pleural effusion, Consolidation, Pneumothorax, Edema, Atelectasis and Cardiomegaly in images is the normalized reconstruction error multiplied by uncertainty.
