Presentation: Generative Models and Explainable AI in Neuroinformatics

Abstract: This presentation outlines applied deep learning solutions to neuroinformatics problems from both generative and explanatory perspectives. We introduce a deep generative model that reconstructs anatomically accurate brain scans, offering a potential pathway for the early, non-invasive detection of brain anomalies. A key feature of this research is its mixed-method analysis, where the quality of generated images is rigorously validated against state-of-the-art quantitative metrics and through qualitative assessments by expert neuroradiologists. In parallel, we address diagnostic prediction in high-dimensional, low-sample (n << p) multimodal datasets, which is a persistent challenge in clinical machine learning. Here, we utilise eXplainable AI (XAI) with Shapley analysis to reveal the predictive logic of our models. This synthesis of generative and explanatory approaches aims to create transparent, reliable decision-support tools, highlighting transdisciplinary collaboration as the essential bridge from algorithm to clinical utility.

Biography: Dr Mousumi Rizia is a Research Fellow in the Neuroinformatics research group at the ANU School of Computing. She works at the intersection of artificial intelligence, neuroscience, and structural diagnostics, with a focus on developing non-invasive tools for anomaly detection across domains. Her current projects bring together deep generative modeling, explainable AI (XAI), and intelligent inspection methods in brain health. Prior to joining ANU, she held research roles at the NASA MIRO Aerospace Centre, University of Texas at El Paso, in collaboration with the National Energy Technology Laboratory (NETL) and the U.S. Department of Energy.