کنفرانس مهندسی برق ایران

صفحه اصلی / سی و چهارمین کنفرانس بین المللی مهندسی برق

An explainable multimodal diagnostic tool for Alzheimer’s disease diagnosis

نویسندگان :

Samaneh Rezaei¹ Amirhossein Ebrahimi² Jamshid Bagherzadeh Mohasefi³ Uffe Kock Wiil⁴ Reza Rawassizadeh⁵ Hamed Ghanipour Keshtiban⁶

1- دانشگاه ارومیه 2- دانشگاه ارومیه 3- دانشگاه ارومیه 4- University of Southern Denmark 5- Boston University 6- دانشگاه علوم پزشکی ارومیه

کلمات کلیدی :

Alzheimer’s diagnosis،Multimodal،super-resolution،explainable AI،Deep learning

چکیده :

Alzheimer’s disease (AD) is an irreversible neurodegenerative disease most common in the aged. The complex nature of AD makes physicians use multimodal data for the diagnosis process. Interrelated regions of MRIs that are essential for the AD diagnosis process could be analyzed with CNN models. As some details are obscured in standard-resolution MRIs, a super-resolution technique is used to reveal more anatomical structures. Additionally, clinical features can be analyzed using shallow machine learning models. As the interpretability of the deep models improves, the decision-making process becomes more reliable. Consequently, Shapley Additive Explanations (SHAP) and Score-Weighted Visual Explanations (Score-Cam) are utilized for shallow and CNN models, respectively. Thus, we proposed an explainable multimodal diagnostic tool named MX-Net to fuse the extracted features from MRIs along with clinical features. Moreover, the explainable techniques provide insight into the diagnostic process. MX-Net trained and tested on the ADNI dataset with 97.73% accuracy on the test set.