CNN-LSTM framework spots Fragile X biomarkers in EEG brain waves

Researchers at Cincinnati Children's Hospital Medical Center (Zag ElSayed, Payton Siekierski, Jack Yanchen Liu, and Ernest Pedapati) have published a new deep learning framework that automatically detects Fragile X syndrome (FXS) from electroencephalography (EEG) data. The paper, submitted to arXiv on August 1, 2026, proposes a multi-representation architecture that combines convolutional neural networks (CNNs), long short-term memory networks (LSTMs), and recurrence plot (RP) analysis to capture both spectral and temporal abnormalities in brain activity.

FXS is a neurodevelopmental disorder caused by reduced expression of fragile X mental retardation protein (FMRP), leading to cortical hyperexcitability and disrupted network synchronization. The researchers focused on alpha (8-12 Hz) and gamma (30-100 Hz) oscillations, which are consistently altered in FXS and relate to inhibitory control, sensory processing, and cognition. They decomposed raw EEG into these bands and transformed them into three complementary representations: temporal feature sequences, time-frequency maps, and recurrence plots encoding nonlinear dynamics. CNN modules learned spatial-spectral textures from image-based inputs, while LSTM modules modeled temporal modulations; a hybrid architecture jointly captured spatial, temporal, and nonlinear dependencies.

In subject-independent evaluation, the hybrid model outperformed single-modality baselines. Gamma features alone provided strong discriminative power, but integrating alpha and gamma oscillations yielded the best overall classification performance. The results suggest that deep learning with nonlinear representations offers a scalable approach for developing EEG biomarkers in FXS, with potential utility for diagnosis, patient stratification, and treatment monitoring in translational settings. The paper is published as arXiv:2608.00835 and is scheduled for presentation at Springer ISBCom 2026.