#MondaYmaging: seminari d'imatge mèdica | Evaluation of deep learning (DL) algorithms using visual explanation techniques to classify brain Arteriovenous Malformations (AVM) in Arterial Spin Labeling sequences (ASL)
Júlia Romagosa Pérez
Using a database of ASL images of pediatric patients provided by Hospital Sant Joan de Déu in Barcelona, an explainable Deep Learning model is developed to detect the presence of AVM in ASL. First, various models are defined and trained on this data. Subsequently, the Gradient-weighted Class Activation Mapping (GradCAM) method is applied as an explainability technique, and the best model is chosen taking into account the results of different metrics, as well as the explanations they generate about their decisions. Finally, a trained model is achieved that is capable of detecting the presence of AVM with high precision, in addition to providing transparent and logical explanations validated by neuroradiologists.