Detection of fibrosis in Cardiac MR Images

GERMAN MATO; PABLO JIMENEZ

Montevideo

Conferencia; Latin American Meeting on Artificial Intelligence, Khipu 2023; 2023

Artificial Intelligence (AI) and specially automatic learning has been used to detect patterns beyond human sight. Specifically. AI is considered a powerful tool in the Medical Imaging field, since it can collect information from different textures and use them to identify and quantify tissue damage. We apply these  techniques in the Cardiology field. More specifically, we are interested in the detection of fibrosis in myocardial tissue of the left ventricle (LV) through Cardiac Magnetic Resonance (CMR) images without the use of a contrast agent. These agents facilitate the detection of lesions but are contraindicated in some cases. In our approach we create deep features based on an encoder-decoder Convolutional Neural Network architectures, which are much more complex than radiomic features and are created with the unique goal of detecting cardiac fibrosis. We do not analyze an entire CMR  image but patches of the LV, including endocardium. In a second stage we use a fingerprinting technique to obtain a prediction for each patient. Our method results in a competitive patient classification accuracy (about 80%) and results to be applicable for fibrosis segmentation and quantification.