Objective: Recent evidence shows that indicators testing conventional olfactory function have a high degree of similarity to cognitive function tests and the potential to diagnose early-stage Alzheimer's disease (AD). In this study, the efficacy of functional near-infrared spectroscopy time-series data obtained through olfactory stimulation was investigated as an early diagnostic tool for mild cognitive impairment in AD using random forest, a machine learning algorithm.
Patients and methods: We conducted a patient-level, single-group, diagnostic interventional trial using near-infrared signals measured during olfactory stimulation in the prefrontal cortex of 178 older adults ranging from normal to participants with AD as markers to discriminate AD stages. We first divided the participants into normal older adults, AD mild cognitive impairment, and AD groups using dementia diagnostic criteria such as the Mini-Mental State Examination and Seoul Neuropsychological Screening Battery. We compared the left and right oxygenation difference by calculating the relative oxygenation difference from the change in relative oxygen concentration.
Results: A total of 168 participants met the eligibility criteria: 70 (41.6%) had normal cognitive function; 42 (25%) mild cognitive impairment; 21 (12.5%) mild AD; and 35 (20.8%) moderate AD. A random forest machine learning model was developed to predict the AD stage, with an area under the receiver operating characteristic curve of 90.7% for mild cognitive impairment and AD, 90.99% for mild cognitive impairment, and 93.34% for AD only.
Conclusions: Based on the classification of the oxygenation difference index of the left and right prefrontal cortices during olfactory stimulation through machine learning, we found that it was possible to detect early-stage mild cognitive impairment in AD. Our results highlight the potential for early AD diagnosis using near-infrared signals from the prefrontal cortex obtained upon olfactory stimulation. Moreover, the results showed high similarity to the existing cognitive function tests and high accuracy in AD stage classification.