[Paper Review] Bridging the Compression-Precision Paradox: A Hybrid Architecture for Clinical EEG Report Generation with Guaranteed Measurement Accuracy

Wuyang Zhang, Zhen Luo+5 more|arXiv (Cornell University)|Feb 11, 2026

EEG and Brain-Computer Interfaces0 citations

TL;DR

The paper introduces a hybrid EEG-reporting system that guarantees clinical measurement accuracy by extracting exact values via signal processing before compression and using a cross-modal bridge to generate narratives around immutable measurement slots, achieving lower false alarms and faster detection.

ABSTRACT

Automated EEG monitoring requires clinician-level precision for seizure detection and reporting. Clinical EEG recordings exceed LLM context windows, requiring extreme compression (400:1+ ratios) that destroys fine-grained temporal precision. A 0.5 Hz error distinguishes absence epilepsy from Lennox-Gastaut syndrome. LLMs lack inherent time-series comprehension and rely on statistical associations from compressed representations. This dual limitation causes systems to hallucinate clinically incorrect measurement values. We separate measurement extraction from text generation. Our hybrid architecture computes exact clinical values via signal processing before compression, employs a cross-modal bridge for EEG-to-language translation, and uses parameter-efficient fine-tuning with constrained decoding around frozen slots. Multirate sampling maintains long-range context while preserving event-level precision. Evaluation on TUH and CHB-MIT datasets achieves 60% fewer false alarms, 50% faster detection, and sub-clinical measurement precision. This is the first system guaranteeing clinical measurement accuracy in automated EEG reports.

Motivation & Objective

• Formalize the compression-precision paradox in clinical EEG reporting under extreme compression constraints.
• Propose a hybrid architecture that separates measurement extraction from narrative generation to guarantee clinical value accuracy.
• Demonstrate clinical performance gains in seizure detection, value extraction precision, and report traceability on benchmark datasets.

Proposed method

• Hierarchical multirate sampling to balance long-range context with event-level precision.
• Measurement-first guardrails that compute exact values (frequency, duration, amplitude) with full provenance before neural processing.
• Graph-aware, dual-view neural architecture with group and channel-graph attention plus linear-time long-range modeling (SSM) for EEG signals.
• Cross-modal bridge mapping EEG features to language space with progressive semantic anchors and contrastive alignment.
• Constrained, LoRA-adapted language model generation that fills a structured JSON schema with frozen measurement slots, ensuring no numeric hallucinations.

Figure 1. Hybrid architecture: hierarchical sampling balances context and precision; signal processing extracts measurements before compression; cross-modal bridge translates to language space; constrained decoder generates reports around frozen slots.

Experimental results

Research questions

• RQ1 Can end-to-end neural models preserve clinically meaningful EEG measurements when operating under extreme compression?
• RQ2 Does separating measurement extraction from text generation improve clinical accuracy and reduce false alarms in automated EEG reporting?
• RQ3 How does hierarchical sampling and graph-aware processing affect detection latency, localization accuracy, and measurement MAE compared to baselines?
• RQ4 Can a cross-modal bridge reliably map EEG signals to language while preserving measurement fidelity and provenance?

Key findings

• The proposed system achieves 0.51 false alarms per 24 hours with 10.5 s latency on TUH (vs. 1.16 FA/24h and 24.2 s latency for EEGNet; 0.81 FA/24h and 20.3 s for BENDR).
• Value extraction MAE is 0.18 Hz (frequency), 1.16 s (duration), and 3.83 μV (amplitude) for the proposed method, outperforming baselines (EEGNet: 0.48 Hz, 2.32 s, 7.59 μV; BENDR: 0.41 Hz, 2.05 s, 6.83 μV).
• Localization accuracy improves to 85%+ with Jaccard overlap > 0.7 on multi-focal patterns due to graph-aware modeling.
• Ablation studies show measurement guardrails and hierarchical sampling contribute substantially to precision and latency, with calibration and artifact robustness enhancing coverage and reliability.
• The approach yields sub-minute end-to-end latency and maintains clinical provenance for every value, supporting FDA-grade traceability.

(a) Ablation impacts