Poster No:
256
Submission Type:
Abstract Submission
Authors:
Christina Young1, Srikanth Ryali1, Vinod Menon1, Kaustubh Supekar1
Institutions:
1Stanford University, Stanford, CA
First Author:
Co-Author(s):
Introduction:
Neuropsychiatric symptoms (NPS) are almost ubiquitous in individuals with Alzheimer's Disease (AD). Among these, psychotic symptoms are particularly clinically relevant, associated with rapid cognitive decline, functional impairment, increased institutionalization, and heightened mortality. Psychosis is present in nearly 41% of individuals with AD, constituting a significant public health concern. However, despite growing evidence linking psychopathology to aberrations in the functional interactions of brain circuit regions over time, the specific brain circuit signatures of AD-related psychosis remain largely unexplored. This gap in knowledge is primarily due to inconsistent findings from small-scale studies lacking the power to detect robust effects, compounded by inadequate analytical methods not well-suited for examining brain circuits. In this study, we aim to bridge the knowledge gap by using open-source data and recent artificial intelligence (AI) advances to identify brain circuit signatures unique to AD psychosis and explore their overlap with non-AD psychosis.
Methods:
We examined multi-cohort clinical and task-free fMRI from 629 participants (214 with AD dementia; 40% of them had psychosis, 120 with non-AD schizophrenia, 120 with non-AD early psychosis, and 175 healthy controls), using a novel explainable AI (XAI) based framework. The field of XAI has been revolutionized in recent years by deep neural networks (DNNs); however, no study to date has employed DNNs to identify brain circuit signatures unique to AD psychosis using functional brain imaging data. This gap is due to the many challenges associated with applying DNNs to functional brain imaging data. We addressed these challenges by developing a novel spatiotemporal DNN (stDNN) model, which takes as its input fMRI time series data from brain regions of interest and models the underlying dynamic spatiotemporal characteristics of brain activity to distinguish between groups. We trained an stDNN to distinguish AD individuals with psychosis from those without and evaluated its performance using cross-validation analysis. To determine overlap with non-AD psychosis, we trained two additional stDNNs de novo – one to distinguish non-AD individuals with schizophrenia from those without, and another to distinguish non-AD individuals with early psychosis from those without. To identify brain circuit signatures associated with AD psychosis, non-AD schizophrenia, and non-AD early psychosis, we applied an XAI method, integrated gradients, to the three trained models respectively.
Results:
stDNN achieved a high cross-validation accuracy of 80.0±1.23% in distinguishing between AD individuals with psychosis and those without. Additionally, stDNN achieved high accuracies of 82.0±1.58% and 86.0±2.41% in distinguishing non-AD individuals with schizophrenia from those without, and in distinguishing between non-AD individuals with early psychosis from those without, respectively. Notably, the stDNN model trained for distinguishing non-AD individuals with early psychosis could also distinguish between AD individuals with and without psychosis. However, the model trained for distinguishing schizophrenia in non-AD individuals did not show this capability. XAI analysis revealed that brain features in the insula node of the salience network, PCC and MTL nodes of the default mode network, and DLPFC node of the frontoparietal network significantly contributed to predicting psychosis in AD as well as early psychosis in non-AD individuals.
Conclusions:
Our findings reveal distinct brain circuit signatures associated with psychosis in AD, showing evidence of their overlap with early, non-AD psychosis rather than established non-AD schizophrenia within the triple-network, providing substantial empirical support for the theoretical aberrant salience-based model of psychosis. These insights advance our neurobiological understanding of psychosis in AD and inform the development of more targeted therapeutic approaches.
Disorders of the Nervous System:
Neurodegenerative/ Late Life (eg. Parkinson’s, Alzheimer’s) 1
Psychiatric (eg. Depression, Anxiety, Schizophrenia)
Modeling and Analysis Methods:
Classification and Predictive Modeling 2
fMRI Connectivity and Network Modeling
Novel Imaging Acquisition Methods:
BOLD fMRI
Keywords:
FUNCTIONAL MRI
Machine Learning
Neurological
Open Data
Schizophrenia
1|2Indicates the priority used for review
Provide references using author date format
Ismail, Z., Creese, B., Aarsland, D. et al (2022), 'Psychosis in Alzheimer disease — mechanisms, genetics and therapeutic opportunities'. Nat Rev Neurol 18, 131–144.