Performance-weighted ensemble learning for detecting patients with FTD and ALS from brief reading speech

Abstract We developed a novel machine learning model named performance-weighted ensemble learning (PWEL) to detect frontotemporal dementia (FTD) and amyotrophic lateral sclerosis (ALS) using speech from word and sentence reading tasks. Overall, 197 participants (30 with FTD, 74 with ALS, and 93 healthy controls) were enrolled. The brief reading tasks consisted of materials adapted from the Western Aphasia Battery (WAB), and 405 speech features (i.e., 384 acoustic, 17 linguistic, and 4 temporal) were extracted. PWEL, introduced under-bagging and adaptive task selection, achieved an area under the curve (AUC) of 0.852, with a sensitivity of 0.837. The sensitivity for FTD subtypes was 0.900–1.000, and that for ALS reached 0.798. The model demonstrated consistent accuracy across recording sites and performed comparably to the WAB (AUC = 0.861). In conclusion, PWEL is practical, scalable, and adaptable to multilingual settings, and is a promising tool for the detection of neurodegenerative disorders.