Alzheimer's disease is a common neurodegenerative disease that affects millions of people worldwide, with the number expected to rise in the future. As a consequence, early diagnosis and treatments are crucial aspects. Current diagnostic methods, such as clinical, imaging, biochemical, and genetic data, have limitations in terms of cost, invasiveness, and availability. Hand-writing analysis has been proposed as a non-invasive and cost-effective alternative for tracking disease progression and machine learning methods have been used to discriminate between healthy and diseased subjects based on handwriting dynamics. However, collecting patient data is still a challenging task. Nevertheless, these issues can be tackled using one-class classifiers, which need only samples produced by healthy subjects to recognize samples drawn by patients. The paper proposes two variants of the Negative Selection Algorithm for the diagnosis of Alzheimer's disease, i.e., RNSA-VR and RNSA-DE, which make use of detectors endowed with variable radii. To assess the effectiveness of the proposed approaches, the largest dataset present in literature, i.e., the DARWIN dataset, has been used, which includes handwriting tasks executed by 174 subjects and the data were split in such a way to reproduce a data scarcity scenario, quite different from the one usually adopted for performance evaluation in classification experiments.
Introducing the RSNA-VR and the RSNA-DE Algorithms for Diagnosing Alzheimer's Disease
Della Cioppa, Antonio
;Marcelli, Angelo
;
2024-01-01
Abstract
Alzheimer's disease is a common neurodegenerative disease that affects millions of people worldwide, with the number expected to rise in the future. As a consequence, early diagnosis and treatments are crucial aspects. Current diagnostic methods, such as clinical, imaging, biochemical, and genetic data, have limitations in terms of cost, invasiveness, and availability. Hand-writing analysis has been proposed as a non-invasive and cost-effective alternative for tracking disease progression and machine learning methods have been used to discriminate between healthy and diseased subjects based on handwriting dynamics. However, collecting patient data is still a challenging task. Nevertheless, these issues can be tackled using one-class classifiers, which need only samples produced by healthy subjects to recognize samples drawn by patients. The paper proposes two variants of the Negative Selection Algorithm for the diagnosis of Alzheimer's disease, i.e., RNSA-VR and RNSA-DE, which make use of detectors endowed with variable radii. To assess the effectiveness of the proposed approaches, the largest dataset present in literature, i.e., the DARWIN dataset, has been used, which includes handwriting tasks executed by 174 subjects and the data were split in such a way to reproduce a data scarcity scenario, quite different from the one usually adopted for performance evaluation in classification experiments.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
