Subtyping Disease Progression Pattern in Dementia: BBB leakage and Cortical thinning Trajectories

1389 

Abstract Submission 

Yanghee Im¹, Won-Jin Moon², Hong Jun Jeon², Hyun Woo Chung³, Kyoung Ja Kwon⁴, Min-Young Noh⁵, Hee-Jin Kim⁵, Seol-Heui Han², Yeonsil Moon², Hosung Kim¹

¹University of Southern California, Los Angeles, CA, ²Research Institute of Medical Science, Konkuk University of Medicine, Seoul, Korea, Republic of, ³Departments of Nuclear Medicine, Konkuk University Medical Center, Seoul, Korea, Republic of, ⁴Center for Geriatric Neuroscience Research, Institute of Biomedical Science and Technology, Seoul, Korea, Republic of, ⁵Department of Neurology, Hanyang University Medical Center, Hanyang University College of Medicine, Seoul, Korea, Republic of

Yanghee Im  
University of Southern California
Los Angeles, CA

Won-Jin Moon, MD, PhD  
Research Institute of Medical Science, Konkuk University of Medicine
Seoul, Korea, Republic of

Hong Jun Jeon  
Research Institute of Medical Science, Konkuk University of Medicine
Seoul, Korea, Republic of

Hyun Woo Chung  
Departments of Nuclear Medicine, Konkuk University Medical Center
Seoul, Korea, Republic of

Kyoung Ja Kwon  
Center for Geriatric Neuroscience Research, Institute of Biomedical Science and Technology
Seoul, Korea, Republic of

Min-Young Noh  
Department of Neurology, Hanyang University Medical Center, Hanyang University College of Medicine
Seoul, Korea, Republic of

Hee-Jin Kim  
Department of Neurology, Hanyang University Medical Center, Hanyang University College of Medicine
Seoul, Korea, Republic of

Seol-Heui Han  
Research Institute of Medical Science, Konkuk University of Medicine
Seoul, Korea, Republic of

Yeonsil Moon  
Research Institute of Medical Science, Konkuk University of Medicine
Seoul, Korea, Republic of

Hosung Kim  
University of Southern California
Los Angeles, CA

Dementia, a cognitive decline syndrome, often stems from neurodegenerative diseases like Alzheimer's disease. The blood-brain barrier (BBB) guards the brain against toxins but can break down, allowing harmful substances to enter and cause neuronal injury. Although BBB's impact on neuronal injury is acknowledged, our understanding of the temporal and spatial aspects remains limited. The Subtype and Stage Inference (SuStaIn), a machine-learning approach, reveals the heterogeneity and temporal complexity in diseases like Alzheimer's and Parkinson's disease. SuStaIn identifies subtypes with unique progression trajectories using cross-sectional patient data (Young et al., 2018). This study employs multimodal imaging markers, including BBB permeability and cortical thickness explaining neuronal loss, to stratify dementia into subgroups. The goal is to associate clinical symptoms with each identified subtype.

A total of 137 participants composed of 36 patients of normal cognition (NC), 52 patients of mild cognitive impairment (MCI) and 49 patients of dementia were studied. The diagnoses of MCI and dementia were based on previous criteria (Petersen et al., 1999). Patients with MCI performed normal daily living activities; nonetheless, they exhibited an objective memory impairment, implying <1.5 standard deviation from the norm in at least one memory test. Dynamic Contrast-Enhanced (DCE) using Gadobutrol (1.0 mmol/kg of body weight) and T1-weighted MRI acquisitions were conducted at the Konkuk University Medical Center using a Magnetom Skyra 3.0 Tesla unit (Siemens Medical Systems, Erlangen, Germany). The specific parameters and protocols can be found in (Moon et al., 2023). Automated brain region segmentation was conducted using the InBrain (https://www.inbrain.co.kr). The cortical thickness and BBB K-trans metrics extracted from DCE-MRI were mapped and averaged in each cortical ROI based on the Desikan Killiany atlas (Desikan et al., 2006). The SuStaIn model was applied to 12 features: two metrics, K-trans (measure of BBB-permeability) and cortical thickness (measure of neuronal loss), in six different ROIs. Each feature was z-scored transformed with respect to mean and standard deviation of NC subjects. A sequential transition of each feature from z-score of 1 to 2 and 3 represented disease progression pattern for each subtype, resulting in a series of stages. Subjects were assigned to the most probable subtype and stage by SuStaIn based on their maximum likelihood. The optimal number of subtypes was determined by the lower Cross-Validation Information Criterion (CVIC) calculated through 10-fold cross validation. Demographic comparisons between NC, MCI, and dementia groups were conducted using two-sample t-tests and chi-squared tests (Table 1). Pearson correlation analysis was performed to examine the correlation between the SuStaIn Stage and cognitive scores.

The model with two subtypes was the most stable showing the lowest variation in the log likelihood (Fig. 1A). Among 137 subjects, 77 were classified as subtype 1 and 60 as subtype 2 (Fig. 1B). All patients were assigned to each subtype with high probability (>50%, Fig. 1C). Each subtype was characterized by distinct patterns of sequential increases in BBB permeability and cortical thinning (Fig. 1D and E). For subtype 1, both biomarkers become abnormal simultaneously, whereas dysfunction of BBB precedes cortical thinning for subtype 2. In subtype 1, significant correlations were found between SuStaIn stage and MMSE and CDR-SOB, but not in subtype 2 (Fig. 2).

In the present study, we revealed two different spatiotemporal trajectories of dementia using two crucial imaging biomarkers. Each subtype displayed a different relationship with clinical features. Further studies analyzing crucial factors, including various cytokines, NfL, and other biomarkers related to BBB permeability and neuronal injury, are needed.

Neurodegenerative/ Late Life (eg. Parkinson’s, Alzheimer’s) ²

Classification and Predictive Modeling ¹

Aging

Computational Neuroscience

Machine Learning

Other - disease progression;Alzheimer's disease;blood-brain barrier;classification