Monday, Aug 3: 10:30 AM - 12:20 PM
1262
Invited Paper Session
Thomas M. Menino Convention & Exhibition Center
Room: CC-253A
Applied
Yes
Main Sponsor
Section on Statistics in Imaging
Co Sponsors
ENAR
Section on Nonparametric Statistics
Presentations
Multiple sclerosis (MS), an inflammatory and degenerative disease of the central nervous system, affects approximately a million Americans. While MS lesions detectable on magnetic resonance imaging, their radiological presentation is like that of many other neurological conditions. The overinterpretation of imaging findings has given rise to an epidemic of misdiagnosis of MS. As novel imaging modalities that allow for better lesion characterization promising increased specificity for MS have increased in popularity, the challenges of quantifying these signatures of disease are mounting. Furthermore, other features, including morphology and morphometry of normal-appearing brain structures that are not detectable by the human eye, have recently been demonstrated to have diagnostic value and are now included in the diagnostic criteria. Leveraging multi-modal imaging approaches that focus on knowledge about etiology is critical for developing the next generation of robust and generalizable diagnostic imaging biomarkers.
Keywords
imaging
multiple sclerosis
The problem of time-varying graphical modeling arises in the analysis of multivariate functional data, where repeated measurements of multiple correlated variables are collected over a sequence of time points. We propose a novel functional approach for estimating time-varying graphical models to characterize dynamic brain network patterns from single-subject fMRI data. The proposed method employs kernel-weighted penalties based on the Kullback–Leibler (KL) divergence to capture local temporal information while enforcing structural smoothness across graphs. In addition, the method incorporates task-design information in fMRI studies, enhancing its ability to detect rapidly evolving network structures. Simulation studies and analyses of task-based fMRI data demonstrate that the proposed approach effectively integrates information from both local and global temporal neighborhoods, resulting in improved temporal resolution and more accurate inference of dynamic brain networks.
Keywords
time-varying graphical model
functional data
functional connectivity
Generative AI has rapidly transformed the biomedical imaging field by enabling image synthesis, helping address challenges of limited data availability, privacy, and diversity in biomedical research. Yet, the adoption of AI-generated images in biomedical studies requires rigorous methods to ensure their reliability for downstream analysis. In this talk, I will introduce novel and rigorous nonparametric approaches that strengthen the trustworthiness and statistical validity of synthetic biomedical imaging data. We develop simultaneous confidence regions to rigorously quantify uncertainty and detect meaningful differences between synthetic and original imaging data. To further enhance fidelity and utility, we propose a transformation that aligns the mean and covariance structures of synthetic images with those of the originals. I will also discuss methods for imputing missing imaging phenotypes using generative models and demonstrate how joint analysis of observed and imputed traits enhances inference while accounting for imputation error. Extensive simulations and applications to brain imaging data validate the proposed framework, demonstrating how these methods empower rigorous statistical inference and promote trustworthy advances in biomedical imaging.
Keywords
Trustworthy AI
Statistical Inference
Synthetic Data
Brain Imaging Studies
Electrophysiological brain signals are typically acquired through indirect and noisy measurements, providing transformed representations of the underlying neural activity. Source reconstruction --- the inverse problem of resolving underlying neural signals from these measurements --- is essential for accurate brain function mapping. However, this task remains extremely challenging due to its mathematical ill-posedness and resulting sensitivity to noise. Deep learning methods have shown promise across a range of inverse problems, but they often disregard the underlying physical principles governing the data generation process, leading to inefficient learning. In this talk, I will introduce a novel physics-informed geometric deep learning framework that embeds potentially ill-posed physics constraints into the model via a custom layer, resulting in more efficient learning and improved reconstruction performance. This custom layer enables the neural network to adapt to subject-specific variations in the physics of signal generation and, as a byproduct, to seamlessly handle missing data within the sensor measurements. We demonstrate the proposed method on magnetoencephalography source reconstruction from a cohort of adolescents, which is characterized by particularly low signal-to-noise ratios.
Keywords
brain function
Deep learning
neural network
Functional data consist of trajectories observed over a continuous domain, such as time, space, or wavelength. Here we consider curves observed on different groups of subjects and propose a Bayesian multi-group functional factor analysis framework that jointly models the data via an explicit decomposition into group-specific mean functions and latent components that capture both common and distinct latent structures across the groups. We impose a parameter-expanded cumulative shrinkage process prior on the factor loadings, which induces increasing shrinkage and enables data-driven basis selection. For real data analysis, we apply the model to EEG data on alcoholic and healthy subjects and identify shared latent factors that capture canonical characteristic components of the EEG curves which are common to both groups of subjects, along with group-specific factors that reveal group-specific neural activity patterns.
Keywords
Functional data
Bayesian analysis
data-driven basis
EEG data