Doctoral colloquium - Jafari Fatana (9.3.2026)
Monday 9.3.2026 at 13:10, Lecture room I 9
Jafari Fatana:
Hybrid Xception-Vision Transformer Model for Automatic Diagnosis and Classification of White Matter Lesions with Explainable AI Techniques
Abstract:
White matter hyperintensities (WMHs) are brain abnormalities associated with cerebral small-vessel disease and are readily detectable on magnetic resonance imaging (MRI), playing a critical role in the diagnosis of complex neurological disorders. Recent advances in artificial intelligence (AI), particularly deep learning, have enabled powerful automated approaches for medical image analysis, including detection, segmentation, and classification tasks, thereby improving diagnostic accuracy and efficiency in neuroimaging. However, despite their strong performance, many AI models remain difficult to interpret due to their “black-box” nature, which limits clinical acceptance and trust among healthcare professionals. Explainable artificial intelligence (XAI) addresses this challenge by providing insights into model decision-making processes and enhancing transparency.
In this study, we propose a framework for the detection and classification of white matter hyperintensities from brain MRI using deep learning models integrated with explainable AI techniques. A hybrid architecture combining Xception and Vision Transformer (ViT) models will be employed for WMH detection and classification. In addition, several deep learning architectures, including Capsule Network (CapsNet), VGG, and ResNet, are implemented and evaluated for comparative analysis. To enhance interpretability, XAI methods, such as Gradient-weighted Class Activation Mapping (Grad-CAM), Local Interpretable Model-agnostic Explanations (LIME), and Integrated Gradients are applied to provide visual and quantitative explanations of model predictions. The proposed approach aims to improve both predictive performance and model transparency, supporting reliable and clinically interpretable AI-assisted neuroimaging analysis.