Preventing Data Leakage in EEG-Based Survival Prediction: A Two-Stage Embedding and Transformer Framework

2026-03-26

AI Summary

This study addresses the critical yet previously overlooked issue of implicit data leakage in EEG-based prognostic prediction for comatose patients, where repeated use of overlapping segments across evaluation phases artificially inflates validation performance and undermines generalizability. The work systematically identifies this patient-level leakage risk and proposes a rigorously leakage-proof two-stage framework: in the first stage, short EEG segments are encoded into discriminative embeddings using a CNN combined with ArcFace; in the second stage, a Transformer aggregates these embeddings to produce patient-level predictions. Crucially, training and validation sets are strictly segregated at the cohort level to eliminate leakage pathways. Evaluated on a large post-cardiac arrest EEG dataset, the method achieves stable and generalizable performance, maintaining high sensitivity even at high specificity thresholds—thereby balancing clinical utility with reliable evaluation.

Abstract

Deep learning models have shown promise in EEG-based outcome prediction for comatose patients after cardiac arrest, but their reliability is often compromised by subtle forms of data leakage. In particular, when long EEG recordings are segmented into short windows and reused across multiple training stages, models may implicitly encode and propagate label information, leading to overly optimistic validation performance and poor generalization. In this study, we identify a previously overlooked form of data leakage in multi-stage EEG modeling pipelines. We demonstrate that violating strict patient-level separation can significantly inflate validation metrics while causing substantial degradation on independent test data. To address this issue, we propose a leakage-aware two-stage framework. In the first stage, short EEG segments are transformed into embedding representations using a convolutional neural network with an ArcFace objective. In the second stage, a Transformer-based model aggregates these embeddings to produce patient-level predictions, with strict isolation between training cohorts to eliminate leakage pathways. Experiments on a large-scale EEG dataset of post-cardiac-arrest patients show that the proposed framework achieves stable and generalizable performance under clinically relevant constraints, particularly in maintaining high sensitivity at stringent specificity thresholds. These results highlight the importance of rigorous data partitioning and provide a practical solution for reliable EEG-based outcome prediction.

Problem

• data leakage
• EEG-based prediction
• patient-level separation
• outcome prediction
• generalization

Innovation

• data leakage prevention
• two-stage embedding
• Transformer-based aggregation
• patient-level isolation
• EEG-based survival prediction