The classification of Alzheimer’s disease stage by frontal cortex hemodynamics measurement during olfactory stimulation using functional near-infrared spectroscopy
- Abstract
- Alzheimer's disease is the most common type of dementia. There is no suitable treatment for Alzheimer's disease, and early detection and palliative treatment are known to be the only solutions. Moreover, Alzheimer's is a type of progressive dementia, and only a few methods are known for early diagnosis.
It is known that the olfactory function is closely related to the cognive functionm, and numerous studies have shown a correlation between olfatory and cognitive functions in Alzheimer's dementia. The olfactory function is performed by asking the patients what kind of smell or smell it has after an olfactory stimulation, such as the Sniff stick test or brief smell identification test (B-SIT). Several studies have reported that the olfactory function is unilateralized in normal people. Therefore, quantifying olfactory function or lateralization of olfactory function in real-time and comparing it with cognitive function evaluation can help early detection of patients with Alzheimer’s.
This study developed a the functional near infrared spectroscopy (fNIRS) system and studied an algorithm for early detection of Alzheimer's dementia by measuring changes in prefrontal blood flow that change in response to olfactory stimuli.
Chapter 1 describes fNIRS and presents research related on olfactory function and Alzheimer's dementia. We also introduced a method to calculate the hemodynamic response to obtain signals from the prefrontal/frontal cortex using fNIRS. In addition, we introduced studies related to olfaction in the left and right frontal lobes and presented the process of developing a system for quantifying olfactory function.
In Chapter 2, we used the left-right (LR) Oxygenation difference to check the olfactory lateralization function in the fNIRS signal. These figures quantified the olfactory function by averaging the olfactory stimulus intervals. According to each stage, normal, mild cognitive impairment, and dementia stages were investigated as quantified indicators. The analysis of covariance (ANCOVA) technique identifies differences between groups. We also observed the effect of eliminating covariates known to be associated with Alzheimer's disease. In addition, we investigated whether there is a statistical correlation with cognitive function tests. such as Seoul neuropsychological screening battery (SNSB) and mini-mental state examination (MMSE). Based on the uncorrected fNIRS data, we confirmed the effect of of magnetic resonance imaging (MRI) and positron emission tomography-computed tomography (PET-CT) against dementia diagnosis ability. In particular, the right orbital frontal cortex (OFC) was slightly activated during olfactory stimulation in normal subjects, but the left orbital prefrontal cortex was relatively activated in patients with mild cognitive impairment and dementia. These findings show the possibility of early detection of Alzheimer's disease from abnormal olfactory function from mild cognitive impairment patients.
In Chapter 3, external verification was performed using the average value of the olfactory section of the LR oxygenation difference made in Chapter 2. A retrospective study was conducted based on the patients gathered in Chapter 2 of statistical models and machine learning models, and new patients were recruited to prospectively compare the accuracy and predictability of statistical models and machine learning models. Decision tree-based algorithms such as XGboost and gradient boosting (GB), were used for machine learning algorithms. The machine learning model showed high performance in retrospective studies and also showed stable performance in new participants gathered for external validation. Through this, we expect that machine learning models to be applied stably to real life rather than statistical models that require several prior assumptions.
In Chapter 4, we performed machine learning using the LR oxygenation difference whole data. Since a lot of information is lost in the process of reducing the dimension of the representative values used in Chapters 2 and 3, when the entire time series data is used, the machine learning model uses more information than the representative values. Therefore, overall performance improvement can be expected. We also increased the number of patients by recruiting additional patients with moderate Alzheimer's dementia. The algorithm used in this chapter used the random forest method, and its performance was superior to the results previously reported in Chapters 2 and 3.
Our study found that LR oxygenation difference values measured in the prefrontal cortex for lateralization of olfactory function could provide relevant information to study Alzheimer's dementia.
- Author(s)
- Jaewon Kim
- Issued Date
- 2023
- Type
- Thesis
- URI
- https://scholar.gist.ac.kr/handle/local/19801
- 공개 및 라이선스
-
- 파일 목록
-
Items in Repository are protected by copyright, with all rights reserved, unless otherwise indicated.