A Novel Deep Learning Approach for Liver MRI Classification and HCC Detection
Résumé
This work proposes a deep learning algorithm based on the Convolutional Neural Network (CNN) architecture to detect HepatoCellular Carcinoma (HCC) from liver DCE-MRI (Dynamic Contrast-Enhanced MRI) sequences. The Deep Learning technique is an artificial intelligence technique (AI) that tries to imitate the human brain work in the training data and creating models used for decision. Actually, it is widely used for various clinical issues. To diagnose HCC, radiologists consider three different phases during contrast injection (before injection; arterial phase; portal phase for instance). This paper presents an approach that offers a parallel preprocessing algorithm. It allows HCC detection and localization in MRI images via a CNN algorithm. The created CNN model reached an accuracy level of 90% in both arterial and portal phases using MRI patches of 64×64 pixels. We mention also its ability to decrease false detection comparing with our previous works. The obtained good accuracy is considered to be ameliorated in our future works.
Domaines
Informatique [cs]
Origine : Fichiers produits par l'(les) auteur(s)