NeuroMM 2026: Interictal Epileptiform Discharge Detection and Localization in Multimodal Neuro-Signals

Qi Tian, Fei Ma, Zitong YU, Haibo He, Laizhong Cui, Nan Lin, Zheng Lian, Philippe Fournier-Viger, Zebang Cheng

Keywords: multimodal neurophysiological intelligence, epilepsy IED detection, cross-modal EEG reasoning, unified neuro-signal benchmark, NeuroMM challenge

TL;DR: We introduce NeuroMM 2026, a unified multimodal benchmark and challenge for epilepsy IED detection and localization that pushes AI from unimodal EEG analysis toward clinically realistic cross-modal neuro-signal reasoning.

Abstract: Understanding the brain through multimodal neurophysiological signals is a critical frontier for trustworthy artificial intelligence. Signals such as EEG, ECG, EMG, and synchronized behavioral video encode complex and partially observable neural dynamics, yet current epilepsy analysis methods remain largely unimodal and fail to capture cross-modal reasoning essential for clinical robustness. In this proposal, we introduce the **NeuroMM 2026 Challenge**, a unified benchmark for multimodal neuro-signal intelligence centered on Interictal Epileptiform Discharge (IED) detection and localization. Built upon the clinically grounded vEpiSet dataset, NeuroMM integrates heterogeneous physiological signals with synchronized visual context under standardized acquisition, annotation, and evaluation protocols. The challenge comprises three tracks targeting multimodal IED detection, vision-enhanced artifact robustness, and epileptogenic zone localization, reframing epilepsy analysis as a multimodal reasoning problem. Evaluation emphasizes clinically meaningful precision–recall trade-offs under extreme class imbalance, where precision and sensitivity are defined as $Precision=\frac{TP}{TP+FP}$ and $Sensitivity=\frac{TP}{TP+FN}$. Binary detection tracks adopt AUPRC as the primary metric, while localization employs weighted F1 aggregation $$Weighted\text{-}F1=\sum_{c=1}^{C} w_c F1_c,$$ ensuring balanced spatial performance. By providing a rigorous and ethically curated benchmark, NeuroMM establishes a standardized evaluation paradigm for cross-modal neuro-intelligence, aiming to advance robust AI systems that bridge machine perception and clinical decision-making in real-world brain health diagnostics.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 7