Diagnosis and precise localization of cardiomegaly disease using U-NET
Abdelilah Bouslama, Yassin Laaziz, Abdelhak Tali
Abstract
This study examines an end-to-end technique which uses a Deep Convolutional Neural Network U-Net based architecture to detect Cardiomegaly disease. The learning phase is achieved by using Chest X-ray images extracted from the “ChestX-ray8” open source medical dataset. The Adaptive Histogram Equalization (AHE) method is deployed to enhance the contrast and brightness of the original images. These latter are compressed before undergoing a training stage to optimize computation time. By this method, we obtained a diagnostic accuracy greater than 93%, which outperforms published results for recognizing Cardiomegaly disease. In addition, with U-Net, precise localization of Cardiomegaly is possible, which is not the case in previous works.