Robust Graph-based Learning in Digital Pathology | Dr. Nasim Yahya Soltani | IEEE CIS

Streamed as part of the IEEE Schenectady Section Computational Intelligence Society (CIS) Chapter webinar series. In this technical talk, Dr. Nasim Yahya Soltani (Northwestern Mutual Assistant Professor at Marquette University) presents a unified graph-based framework for robust and uncertainty-aware Whole Slide Image (WSI) analysis in digital pathology. Gigapixel-scale WSIs provide rich spatial and contextual data for cancer diagnosis but are frequently vulnerable to acquisition artifacts, staining variability, adversarial noise, and distribution shifts. Dr. Soltani details an innovative approach that represents WSIs as graphs to capture complex spatial dependencies, leveraging Graph Neural Networks (GNNs)—including GCNs, GATs, and GraphSAGE—for advanced feature learning. Key topics covered include: Overcoming gigapixel-scale data constraints and distribution shifts in deep learning. Input-specific denoising modules to mitigate motion and staining variability artifacts. Defending against image- and graph-level adversarial perturbations. Transformer-based global contextual modeling for refined cancer classification. Multi-head GNN frameworks for predictive uncertainty estimation and out-of-distribution (OOD) detection. 🔗 Event Details: https://events.vtools.ieee.org/m/557649 📌 Subscribe to our channel for more technical talks:    / @ieeeschenectadysection   ABOUT THE SPEAKER Dr. Nasim Yahya Soltani received her Ph.D. in Electrical Engineering from the University of Minnesota. She has served as a Research Associate at the Digital Technology Center (UMN) and a Senior Data Scientist at Harley-Davidson Motor Company. Since 2019, she has been an Assistant Professor in Computer Science at Marquette University. Her research spans statistical signal processing, machine learning, optimization theory, and network science with key applications in healthcare systems. #IEEE #DigitalPathology #GraphNeuralNetworks #MachineLearning #GNN #CancerDetection #Bioinformatics #IEEEComputationalIntelligence