EEGlass: A Fully Interpretable EEG Dementia Screening Assistant
Authors/Creators
- 1. Institute of Visual Computing, Graz University of Technology, Graz, Austria
- 2. University of Padova, Padova, Italy
Description
We present EEGlass, an open-source visual analytics dashboard for interpretable EEG-based dementia screening. Unlike prior work that applies explainability as a post-hoc add-on, EEGlass builds on a classifier whose every intermediate quantity has a direct physical meaning, exposing the full inference pipeline from raw signals through spectral features and spatial topomaps to per-class output logits interactively. A cohort-level Overview supports exploration of classification performance and activation structure; the Patient View lets clinicians trace any prediction back to its signal-level origin, with all panels linked and updating synchronously as the user navigates windows, electrodes, and frequency bands. An integrated Visual Analytics Context Protocol (VACP) layer enables natural-language dashboard navigation and interaction via a Large Language Model (LLM). Following a human-in-the-loop philosophy, EEGlass is designed as decision support for the dementia diagnosis / screening task.