A prospective multicenter validation study of a machine learning algorithm classifier on quantitative electroencephalogram for differentiating between dementia with Lewy bodies and Alzheimer's dementia

• Published in: PloS one Vol. 17; no. 3; p. e0265484
• Main Authors: Suzuki, Yukiko, Suzuki, Maki, Shigenobu, Kazue, Shinosaki, Kazuhiro, Aoki, Yasunori, Kikuchi, Hirokazu, Baba, Toru, Hashimoto, Mamoru, Araki, Toshihiko, Johnsen, Kristinn, Ikeda, Manabu, Mori, Etsuro
• Format: Journal Article
• Language: English
• Published: United States Public Library of Science 31.03.2022; Public Library of Science (PLoS)
• Subjects: Algorithms; Alzheimer Disease - diagnosis; Alzheimer's disease; Alzheimers disease; Analysis; Biology and Life Sciences; Biomarkers; Child development; Computer and Information Sciences; Dementia; Dementia disorders; Diagnosis; Diagnosis, Differential; Donepezil; EEG; Electroencephalography; Ethnicity; Evaluation; Hospitals; Humans; Learning algorithms; Lewy bodies; Lewy Body Disease - diagnosis; Machine Learning; Medical diagnosis; Medical technology; Medicine; Medicine and Health Sciences; Neurodegenerative diseases; Neurology; Parkinson's disease; Parkinsons disease; Patients; Physical Sciences; Prospective Studies; Psychiatry; Research and Analysis Methods; Scintigraphy; Sensitivity; Systematic review; Japan; Iceland
• ISSN: 1932-6203; 1932-6203
• DOI: 10.1371/journal.pone.0265484

An early and accurate diagnosis of Dementia with Lewy bodies (DLB) is critical because treatments and prognosis of DLB are different from Alzheimer's disease (AD). This study was carried out in Japan to validate an Electroencephalography (EEG)-derived machine learning algorithm for discriminating DLB from AD which developed based on a database of EEG records from two different European countries. In a prospective multicenter study, patients with probable DLB or with probable AD were enrolled in a 1:1 ratio. A continuous EEG segment of 150 seconds was recorded, and the EEG data was processed using MC-004, the EEG-based machine learning algorithm, with all clinical information blinded except for age and gender. Eighteen patients with probable DLB and 21 patients with probable AD were the included for the analysis. The performance of MC-004 differentiating probable DLB from probable AD was 72.2% (95% CI 46.5-90.3%) for sensitivity, 85.7% (63.7-97.0%) for specificity, and 79.5% (63.5-90.7%) for accuracy. When limiting to subjects taking ≤5 mg donepezil, the sensitivity was 83.3% (95% CI 51.6-97.9), the specificity 89.5% (66.9-98.7), and the accuracy 87.1% (70.2-96.4). MC-004, the EEG-based machine learning algorithm, was able to discriminate between DLB and AD with fairly high accuracy. MC-004 is a promising biomarker for DLB, and has the potential to improve the detection of DLB in a diagnostic process.