2025 American Academy of Neurology Abstract Website

Rehab Diab¹, Ibrahim Serag², Amr Hassan³, Ahmed Azzam⁴, Mohamed Diab⁵, Mohammed Ahmed Ali⁶, Mahmoud Tarek Hefnawy⁷, Ahmed Negida⁸
¹Faculty of Medicine, Al-Azhar University, Cairo, Egypt, ²Faculty of Medicine, Mansoura University, ³University of California, Irvine, CA, USA, ⁴Faculty of Medicine, October 6 University, Giza, Egypt, ⁵Faculty of Medicine Alexandria University, Alexandria, Egypt, ⁶Qena Faculty of Medicine, South Valley University, Qena, Egypt, ⁷Faculty of Medicine, Zagazig University, Mansoura, Egypt, ⁸Virginia Commonwealth University

Objective:

This review aims to investigate the current evidence regarding the detection of LBD using XAI.

Background:

Lewy Body Dementia (LBD) is a combination of cognitive impairment, visual hallucinations, parkinsonism, and fluctuations in attention. These heterogeneities of symptoms often complicate the diagnostic process, making early detection and accurate classification critical for effective management. Explainable AI (XAI) emerges as a promising tool that enhances AI models' interpretability while preserving their predictive performance

Design/Methods:

A comprehensive search of four databases PubMed, Web of Science, Scopus, and Cochrane Library to get relevant studies previously published till September 2024. All studies have been included to investigate the role of XAI in the detection of LBD including different parameters regarding preprocessing, model algorithms, model inputs, and performance.

Results:

a review of three studies including over 700 patients showed that the MUQUBIA SVM model and multimodal MRI data showed the highest accuracy, 87.5%, regarding the differentiation of dementia types. AI model based on Random Forest and RIPPER algorithm over clinical data distinguishes LBD from Alzheimer's disease (AD) with 86% accuracy. Multi-class SVM with FDG PET data, yielding 78% and 86% accuracy for pattern-based and ROI-based classifiers, respectively. While the models' objectives and datasets differ, each model has advantages and drawbacks. The model chosen is determined by clinical aims such as multi-class categorization, LBD-AD differentiation, or discriminating between several dementias

Conclusions:

Using XAI approaches like SHAP improved the models' interpretability and accuracy in differentiating LBD from other types of dementia. Despite the potential benefits, integrating XAI into clinical practice for LBD involves several barriers, including the necessity for validation and ethical concerns.