Authors:
Nicoletta Balletti
1
;
2
;
Emanuela Guglielmi
2
;
Gennaro Laudato
2
;
Rocco Oliveto
3
;
2
;
Jonathan Simeone
3
and
Roberto Zinni
4
Affiliations:
1
Center for Biotechnology, Institute of Biomedical Sciences of the Ministry of Defense, Rome, Italy
;
2
University of Molise, Pesche (IS), Italy
;
3
Datasound srl, Pesche (IS), Italy
;
4
WordPower, San Salvo (CH), Italy
Keyword(s):
Gait Analysis, Clinical Biomarkers, Explainable Artificial Intelligence, Parkinson’s Disease Detection.
Abstract:
Several machine learning (ML) approaches have been introduced for gait and posture analysis, recognized as crucial for early diagnosing neurological disorders, particularly Parkinson’s disease. However, these existing methods are often limited by their lack of integration with other clinical biomarkers and their inability to provide transparent, explainable predictions. To overcome these limitations, we introduce EDAM (Explainable Diagnosis Recommender), a system that leverages Explainable Artificial Intelligence (XAI) techniques to deliver both accurate predictions and clear, interpretable explanations of its diagnostic decisions. We evaluate the capabilities of EDAM in two main areas: distinguishing between healthy individuals and those with Parkinson’s disease, and classifying abnormal gait patterns that may indicate early-stage Parkinson’s disease. To ensure a comprehensive evaluation, we constructed one of the largest known dataset by merging and standardizing several existing d
atasets. This dataset includes 557 features and 7,303 labelled instances, covering a wide range of gait patterns and clinical features. Results show that EDAM achieves high accuracy in both tasks, demonstrating its potential for early detection of neurological disorders.
(More)