Tuesday, Aug 6: 10:30 AM - 12:20 PM
1144
Invited Paper Session
Oregon Convention Center
Room: CC-F151
As we navigate an increasingly data-centric world, the fusion of Data Sciences with diverse imaging technologies is opening novel corridors for research and application. This interdisciplinary session brings together an exceptional panel of experts to spotlight this fusion.
Our first presentation (Dr. Daiwei Zhang) introduces iStar, a groundbreaking method that combines spatial transcriptomics and high-resolution histology. This technique paves the way for Data Sciences to revolutionize gene expression mapping at unprecedented scales.
Our second speaker (Dr. Laurent Younes) continues to focus on spatial transcriptomics, tackling the complex issue of atlas to data alignment. By employing advanced algorithms like large deformation diffeomorphic mapping, the talk will explain how Data Sciences can solve non-linear spatial transformations for more accurate cell-type probability distributions.
The third presentation (Dr. Simon Vandekar) explores the neuroscience domain through the lens of Brain-wide Association Studies (BWAS). With an emphasis on data analytics, the talk will highlight how Data Sciences can offer nuanced approaches to improve replicability and power in BWAS.
Our fourth talk (Dr. Yang Chen) centers on space weather forecasting and introduces VISTA, a Data Science-based method for the accurate imputation and prediction of Total Electron Content (TEC) maps essential for GPS operations globally.
Highlight on Diversity:
Our panel is notably diverse, featuring one female speaker and a range of career stages—from senior and mid-career faculty to postdoc researchers. This breadth adds multiple perspectives to the conversation, enriching our exploration of Data Sciences and diverse imaging technologies.
Special Discussant:
We are privileged to feature a world-leading expert (Dr. Hongtu Zhu) in imaging statistics and machine learning as our discussant. His role will be pivotal in synthesizing the technical facets and diverse views presented, offering a holistic understanding of the current landscape of Data Sciences in imaging technology.
Through this rich and multifaceted session, attendees will not only gain a deeper understanding of the role Data Sciences play in diverse imaging disciplines but also engage in meaningful dialogue that could lead to future interdisciplinary collaborations.
Applied
Yes
Main Sponsor
Section on Statistics in Imaging
Co Sponsors
Biometrics Section
International Chinese Statistical Association
Presentations
Spatial transcriptomics (STs) has demonstrated enormous potential for generating intricate molecular maps of cells within tissues. Despite the availability of many ST platforms, none of them provides a comprehensive solution. An ideal ST platform should achieve single-cell resolution, cover the whole transcriptome, and be cost-effective. While such ST data are difficult to collect physically using existing platforms, they can be constructed in silico using innovative machine learning algorithms. Here we present iStar, a generative computer vision model that integrates low-resolution ST measurements with high-resolution histology images to construct spatial gene expression data at near-single-cell-resolution. The resulting model not only enhances gene expression resolution but also enables gene expression prediction in tissue sections where only histology images are available. The application of iStar to healthy and diseased samples across 12 diverse datasets demonstrates our method's efficacy in facilitating scientific inquiries and performing clinical tasks, including tissue segmentation, cell type inference, cancer detection, and tumor microenvironment analysis, all with state-o
The total electron content (TEC) maps can be used to estimate the signal delay of GPS due to the ionospheric electron content between a receiver and a satellite. This delay can result in a GPS positioning error. Thus, it is crucial to monitor & forecast the TEC maps. However, the observed TEC maps have big patches of missingness in the ocean and scattered small areas on the land. In this talk, I first present extensions of existing matrix completion algorithms to achieve TEC map reconstruction, accounting for spatial smoothness and temporal consistency while preserving essential structures of the TEC maps. We show that our proposed method achieves better reconstructed TEC maps as compared to existing methods in the literature. Then, I present a new model for forecasting time series data distributed on a matrix-shaped spatial grid, using the historical spatiotemporal data and auxiliary vector-valued time series data. We model the matrix time series as an auto-regressive process, where a future matrix is jointly predicted by the historical values of the matrix time series and an auxiliary vector time series. Asymptotic results and numerical illustrations with TEC data will be given.
We will address the problem of registering anatomical atlases to observed image data by simultaneously estimating a non-linear spatial transformation of the atlas and a probabilistic correspondence between the atlas labels and the observed image values. This approach can be applied to spatially resolved transcriptomics data, where the image modality can be transformed into probability distributions over cell types in each segmented region. The non-linear spatial transformation uses the large deformation diffeomorphic mapping algorithm. We will describe the relevant algorithm and provide experimental results.
Several recent studies showed that thousands of study participants are required to improve the replicability of brain-phenotype associations because actual effect sizes are much smaller than those reported in smaller studies. Here, we perform a meta-analysis of brain volume associations with age using 63 longitudinal and cross-sectional magnetic resonance imaging studies (77,695 total scans) to show that studies with larger covariate variance have larger effect size estimates and that the longitudinal studies we examined have systematically larger standardized effect sizes than cross-sectional studies. Analyzing age effects on global and regional brain measures in the LBCC, we show that modifying longitudinal study design to increase between-subject variability and adding a single additional longitudinal measurement per subject improves effect sizes. However, evaluating these longitudinal sampling schemes on other phenotype-brain associations in the ABCD dataset shows that longitudinal studies can be detrimental to effect sizes. We integrate these empirical results with statistical theory to elucidate their meaning and establish when longitudinal designs improve replicability.
Co-Author(s)
Kaidi Kang, Vanderbilt University
Jakob Seidlitz, University of Pennsylvania
Richard Bethlehem, University of Cambridge
Jiangmei Xiong
Megan Jones
Kahini Mehta, University of Pennsylvania
Arielle Keller, University of Pennsylvania
Jonathan Schildcrout, Vanderbilt University
Ran Tao, Vanderbilt University Medical Center
Anita Randolph, University of Minnesota Medical School
Bart Larsen, University of Minnesota Medical School
Brenden Tervo-Clemmens, University of Minnesota Medical School
Eric Feczko, University of Minnesota Medical School
Oscar Miranda Dominguez, University of Minnesota Medical School
Steve Nelson, University of Minnesota Medical School
Damien Fair, University of Minnesota Medical School
Theodore Satterthwaite, Univ of Pennsylvania
Aaron Alexander-Bloch, University of Pennsylvania
Speaker
Simon Vandekar, Vanderbilt University