VALIANT Pulse

Can deep learning help in the fight against emerging infectious diseases? In this episode of VALIANT Pulse, we explore the collaborative work led by Vanderbilt University, the NIH, and the Institut National de Recherche Biomédicale to automate mpox lesion counting using AI. Building on field data from the Democratic Republic of the Congo, the team compares leading segmentation models—including UNet++, Mask R-CNN, YOLOv8, and E2EC—t...

In this episode of VALIANT Pulse, we explore a bold attempt to unify aspects unsupervised learning, network science, and imaging neuroscience through the lens of brain data. Mika Rubinov from Vanderbilt describes a series of analytical equivalences and unifications—connecting modularity and k-means, canonical covariance and co-clustering, diffusion maps and co-neighbor matrices—all with mathematical rigor and a nod to neuroscience'...

In this episode of VALIANT Pulse, we feature the dissertation work of Can Cui at Vanderbilt, who pushes the frontier of machine learning in healthcare by tackling the complex challenge of multimodal data. From deep brain stimulation planning to cancer diagnosis with missing data, Cui’s work spans nine projects and modalities—from MRI and histopathology to genomics and free text. Central to this dissertation is the drive to make AI ...

In this episode of VALIANT Pulse, we feature the dissertation work of Thomas Z. Li, who takes on one of medicine’s most pressing challenges: how to predict and diagnose lung cancer with the data we already have. By integrating longitudinal chest CTs and multimodal electronic health records, Li develops a suite of AI models—transformers, joint embeddings, contrastive learning—to predict lung cancer diagnosis and long-term risk. From...

In this episode of VALIANT Pulse, we explore Ruining Deng’s dissertation at Vanderbilt, a sweeping journey through knowledge-infused AI systems for computational pathology. From democratizing annotation with lay users to building large-scale 3D models of renal tissue and creating efficient segmentation strategies for whole slide images, this work tackles real-world pathology challenges with creativity and technical depth. Deng’s re...

Can a single CT slice hold the key to long-term health insights? In this episode of VALIANT Pulse, we spotlight the PhD work of Xin Yu at Vanderbilt, who explores how longitudinal health patterns can be revealed using just single-slice CT scans. From active learning and generative models to transformer-based segmentation, this dissertation introduces a suite of AI-powered techniques to reduce manual effort, enhance consistency, and...

What happens to the brain’s wiring from birth to old age—and how can we measure it? In this episode of VALIANT Pulse, we feature two landmark studies from the Vanderbilt Imaging Science community that build comprehensive white matter atlases spanning the human lifespan. Kurt Schilling and Bennett Landman lead efforts to characterize the micro- and macrostructure of white matter pathways and their associated cortical features using ...

In this episode of VALIANT Pulse, we explore how Yuankai Huo and collaborators at Vanderbilt are transforming cancer diagnostics with 3D spatial transcriptomics. The team introduces ASIGN, an innovative anatomy-aware graph network that combines a single 2D molecular slice with 3D pathology images to reconstruct high-resolution gene expression maps—without the cost of full 3D sequencing. This episode dives into how ASIGN balances af...

Can your brain’s alertness be read from a single moment in time? In this episode of VALIANT Pulse, we explore a groundbreaking study led by Shengchao Zhang and Catie Chang at Vanderbilt, revealing how natural fluctuations in vigilance—our moment-to-moment arousal and wakefulness—are imprinted in the low-dimensional structure of fMRI data. Using unsupervised learning and concurrent EEG, the team shows that even individual brain snap...

What if we could map the quiet, hidden changes in brain wiring that predict who’s at risk for memory loss? In this episode, we explore a groundbreaking study that analyzed over 9,000 cognitive sessions from 4,467 participants, uncovering how subtle changes in the brain’s white matter—especially in the limbic system—signal cognitive decline.

Using harmonized diffusion MRI from nine major cohorts, the team found that free water (FW) i...

This episode spotlights inventive use of food starches to build low-cost hydrogel electrodes for wearable biosensors. By transforming agricultural waste into stable, skin-friendly electrodes, they open the door to sustainable health monitoring in resource-limited settings—without sacrificing signal quality. This summary was AI-generated and reviewed by a contributing author. Views are individual interpretations and not official pee...

In this episode, we explore stDyer, a deep learning framework for spatially resolved transcriptomics that blends Gaussian Mixture VAEs with dynamic graph embeddings. Developed with scalability in mind, stDyer reveals hidden spatial domains in tissue by adapting to boundaries and leveraging GPU acceleration. This AI-generated summary was crafted through prompt engineering and reviewed by a contributing author to ensure accuracy. Vie...

In this episode, we dive into Nancy Newlin’s work on harmonizing diffusion MRI data to enable large-scale analysis of neurodegeneration. By disentangling technical variability across sites, her approach preserves meaningful brain connectivity signatures while making datasets more interoperable. This summary was AI-generated and reviewed by a contributing author. Views shared reflect individual interpretation and are not official pe...