| Sumario: | Purpose Alzheimer's disease (AD) is known as a disease caused by the accumulation of beta-amyloid or tau protein in neurons, blocking signal transmission and causing death. Since there is no cure for this disease, early detection is an important aspect of AD (Rasmussen J et al., 2019). It has been reported that the olfactory function decreases earlier than the cognitive dysfunction in AD patients. (Roberts RO et al., 2015). We have previously reported that the brain hemodynamic signals measured by functional near-infrared spectroscopy during olfactory stimulation could classify the stage of Alzheimer's disease based on a statistical model using 98 patients (55 normal, 26 mild cognitive impairments, 16 dementia) (Kim J et al., 2022). In this study, we recruited additional 34 subjects and applied machine learning to find its potential in AD screening. Method The existing data from 97 participants were used for internal verification, and 34 new participants' data were used as external verification data. All patients were recruited from the GARD cohort. We applied a statistical method and machine learning algorithm to both datasets. For a statistical model, a linear regression was employed, and a naive Bayes-based machine learning algorithm was chosen as a machine learning algorithm applied for AD screening. Results and Discussion A naive Bayes-based machine learning performed better than the statistical method on both internal and external validation datasets. The statistical model reported a classification accuracy of 87% in patients with mild cognitive impairment and Alzheimer's dementia on 97 internal validations and 63% on 34 external validations. On the other hand, naive Bayes-based machine learning algorithm achieved 95% accuracy on internal validation data and 85% accuracy on external validation data suggesting that machine learning has the potential to screen AD from various patient groups better than a statistical model.
|
|---|