Explainable Deep Radiomics Framework for MRI-Based Early Detection and Staging of Dementia

Abstract

Dementia is a term used to indicate a series of neurodegenerative diseases that may lead to decline the thinking ability, memory loss etc. It is the result of damage of the nerve cells and the connections between the cells. This may create various challenges to the patients. Dementia is a broad term which is classified into Alzheimer's Disease, Vascular Dementia, Lewy Body Dementia (LBD), Frontotemporal Dementia (FTD) etc. Effective treatment plan is essential for identifying the early stages of dementia progress. In this study a Convolutional Neural Network (CNN) designed for classify the brain images into different categorizes. To ensure the interpretability of CNN predictions, Gradient-weighted Class Activation Mapping (Grad-CAM) was introduced which produced heatmaps that highlighted regions of interest (ROIs) in MRI images that contributed most to the CNN's decision. For this study a publically available Kaggle data set is used which consist of different categories of MRI images i.e No Impairment, Very Mild Impairment, Mild Impairment, and Moderate Impairment. Here the Custom CNN presents the multiple stages of dementia progression.

Ähnliche Arbeiten