INS NYC 2024 Program

Poster	Poster Session 04 Program Schedule 02/15/2024 12:00 pm - 01:15 pm Room: Majestic Complex (Posters 61-120) Poster Session 04: Neuroimaging \| Neurostimulation/Neuromodulation \| Teleneuropsychology/Technology Final Abstract #81 Convergent Patterns of Cognitive Heterogeneity in Prodromal Stages of Dementia with Data-Driven Machine Learning Truc Nguyen, Taiwan International Graduate Program in Interdisciplinary Neuroscience, National Taiwan University and Academia Sinica, Taipei, Taiwan Ming-Shan Tsai, Department of Psychology, College of Science, National Taiwan University, Taipei, Taiwan Jing-Rong Wang, Department of Psychology, College of Science, National Taiwan University, Taipei, Taiwan Yue-Ling Chiu, Department of Psychology, College of Science, National Taiwan University, Taipei, Taiwan Cheng-Yun Lee, Department of Psychology, College of Science, National Taiwan University, Taipei, Taiwan Yu-Ling Chang, Department of Psychology, College of Science, National Taiwan University, Taipei, Taiwan Category: Cognitive Neuroscience Keyword 1: neuropsychological assessment Keyword 2: mild cognitive impairment Keyword 3: cognitive neuroscience Objective: Growing evidence reveals that mild cognitive impairment (MCI) involves diverse individuals displaying distinct patterns in cognitive profiles, brain imaging, and biomarkers. While previous studies have achieved some success in identifying clinical groups using clustering algorithms, few have explored the performance of multiple strategies when applied to a single dataset. We posit that if meaningful subgroups are consistently discovered, then distinct data-driven cluster analyses must have captured the inherent structure or pattern within the data. In this study, we employ data-driven methods to analyze neuropsychological performance in individuals with normal cognition (NC) and MCI. Our objective is to validate the clustering solutions, thereby ensuring their reproducibility and ability to genuinely reflect clinically meaningful subtypes. Participants and Methods: A total of 311 Taiwanese adults without dementia (age 70.9 ± 7.6 years, female 61.7%, education 13.1 ± 3.4 years) were enrolled in the study. Participants completed a comprehensive battery of 21 neuropsychological tests that spanned five cognitive domains (memory, attention, executive function, language, and visuospatial function). Brain structural MRI data underwent processing using FreeSurfer to derive measurements of cortical gray matter volume and thickness within predefined regions of interest (ROI). The analysis employed two clustering algorithms: nonnegative matrix factorization (NMF) and model-based clustering (MCLUST). Results: Both NMF and MCLUST uncovered two- and three-cluster solutions that exhibited satisfactory fit of the data. The NMF-derived two-cluster profiles encompassed (1) a cognitively intact (CI) group and (2) a cognitively suboptimal group, which were well-distinguished by their cognitive performance, aligning with participants’ baseline diagnosis (NC vs. MCI). The cognitively intact group, compared to the cognitively suboptimal one, was significantly younger, more educated, exhibited better MMSE scores, fewer subjective memory complaints, lower clinical dementia ratings during the baseline visit, less severe hippocampal atrophy, and greater volume of cerebral cortex. Upon setting the number of clusters to three, NMF identified (1) CI–memory, (2) CI–non-memory, and (3) cognitively suboptimal groups. The CI–memory group performed worse in tests of episodic verbal and visual memory and better in other measures of attention, executive, and visuospatial abilities compared to the CI–non-memory profile, but both groups shared comparable demographic characteristics and MRI ROI measurements. Remarkably, similar pattern of cognitive heterogeneity and its association with demographic and neuroimaging variables also surfaced from MCLUST analysis. Conclusions: The results indicate that two distinct data-driven algorithms, using different heuristics, have arrived at a consensus regarding the observed pattern of cognitive heterogeneity within NC and MCI. Furthermore, with each algorithm, the two- and three-cluster solutions consistently revealed one group, the cognitively suboptimal profile, that performed worse in all cognitive domains, who was also older, had less education, and smaller hippocampal volume. This study extended previous research by effectively connecting machine learning algorithms to extract valuable insights from data pertaining to individuals in the prodromal stages of dementia, ensuring a satisfactory level of reliability. Future efforts will focus on integrating longitudinal analysis to identify cognitive phenotypes that exhibit heightened vulnerability to disease progression.

Poster

02/15/2024
12:00 pm - 01:15 pm
Room: Majestic Complex (Posters 61-120)

Poster Session 04: Neuroimaging | Neurostimulation/Neuromodulation | Teleneuropsychology/Technology

Final Abstract #81

Convergent Patterns of Cognitive Heterogeneity in Prodromal Stages of Dementia with Data-Driven Machine Learning

Truc Nguyen, Taiwan International Graduate Program in Interdisciplinary Neuroscience, National Taiwan University and Academia Sinica, Taipei, Taiwan
Ming-Shan Tsai, Department of Psychology, College of Science, National Taiwan University, Taipei, Taiwan
Jing-Rong Wang, Department of Psychology, College of Science, National Taiwan University, Taipei, Taiwan
Yue-Ling Chiu, Department of Psychology, College of Science, National Taiwan University, Taipei, Taiwan
Cheng-Yun Lee, Department of Psychology, College of Science, National Taiwan University, Taipei, Taiwan
Yu-Ling Chang, Department of Psychology, College of Science, National Taiwan University, Taipei, Taiwan

Category: Cognitive Neuroscience

Keyword 1: neuropsychological assessment
Keyword 2: mild cognitive impairment
Keyword 3: cognitive neuroscience

Objective:

Growing evidence reveals that mild cognitive impairment (MCI) involves diverse individuals displaying distinct patterns in cognitive profiles, brain imaging, and biomarkers. While previous studies have achieved some success in identifying clinical groups using clustering algorithms, few have explored the performance of multiple strategies when applied to a single dataset. We posit that if meaningful subgroups are consistently discovered, then distinct data-driven cluster analyses must have captured the inherent structure or pattern within the data. In this study, we employ data-driven methods to analyze neuropsychological performance in individuals with normal cognition (NC) and MCI. Our objective is to validate the clustering solutions, thereby ensuring their reproducibility and ability to genuinely reflect clinically meaningful subtypes.

Participants and Methods:

A total of 311 Taiwanese adults without dementia (age 70.9 ± 7.6 years, female 61.7%, education 13.1 ± 3.4 years) were enrolled in the study. Participants completed a comprehensive battery of 21 neuropsychological tests that spanned five cognitive domains (memory, attention, executive function, language, and visuospatial function). Brain structural MRI data underwent processing using FreeSurfer to derive measurements of cortical gray matter volume and thickness within predefined regions of interest (ROI). The analysis employed two clustering algorithms: nonnegative matrix factorization (NMF) and model-based clustering (MCLUST).

Results:

Both NMF and MCLUST uncovered two- and three-cluster solutions that exhibited satisfactory fit of the data. The NMF-derived two-cluster profiles encompassed (1) a cognitively intact (CI) group and (2) a cognitively suboptimal group, which were well-distinguished by their cognitive performance, aligning with participants’ baseline diagnosis (NC vs. MCI). The cognitively intact group, compared to the cognitively suboptimal one, was significantly younger, more educated, exhibited better MMSE scores, fewer subjective memory complaints, lower clinical dementia ratings during the baseline visit, less severe hippocampal atrophy, and greater volume of cerebral cortex. Upon setting the number of clusters to three, NMF identified (1) CI–memory, (2) CI–non-memory, and (3) cognitively suboptimal groups. The CI–memory group performed worse in tests of episodic verbal and visual memory and better in other measures of attention, executive, and visuospatial abilities compared to the CI–non-memory profile, but both groups shared comparable demographic characteristics and MRI ROI measurements. Remarkably, similar pattern of cognitive heterogeneity and its association with demographic and neuroimaging variables also surfaced from MCLUST analysis.

Conclusions:

The results indicate that two distinct data-driven algorithms, using different heuristics, have arrived at a consensus regarding the observed pattern of cognitive heterogeneity within NC and MCI. Furthermore, with each algorithm, the two- and three-cluster solutions consistently revealed one group, the cognitively suboptimal profile, that performed worse in all cognitive domains, who was also older, had less education, and smaller hippocampal volume. This study extended previous research by effectively connecting machine learning algorithms to extract valuable insights from data pertaining to individuals in the prodromal stages of dementia, ensuring a satisfactory level of reliability. Future efforts will focus on integrating longitudinal analysis to identify cognitive phenotypes that exhibit heightened vulnerability to disease progression.