Graph-Based Approaches to Learning Epileptogenic Zone Localization Using Stereo-EEG Recordings
研究通过比较不同图拓扑结构，使用基于图的学习模型从静息状态sEEG记录中定位癫痫源区，发现Region-Bridge-c拓扑在减少边数的同时提高了定位精度。
0 citationsRead paper
Selected work
Representative Papers
Deep Learning CNN and Recurrence Analysis for Alpha Gamma EEG Biomarkers in Fragile X Syndrome
Individuals with Fragile X Syndrome (FXS) exhibit aberrant alpha/gamma-band electroencephalographic (EEG) oscillations closely linked to impairments in inhibitory control, sensory processing, and cognition; however, efficient automated methods for identifying reliable biomarkers remain lacking. This study proposes a multimodal deep learning framework that, for the first time, integrates convolutional neural networks (CNNs), long short-term memory networks (LSTMs), and joint modeling of time–frequency and nonlinear dynamical features—including recurrence plot analysis—specifically tailored to alpha and gamma EEG signals. In subject-independent evaluations, the proposed approach significantly outperforms unimodal baselines, with the combined alpha–gamma representation achieving the highest discriminative performance. These results underscore the framework’s potential for facilitating automated FXS identification, diagnosis, and treatment monitoring.
0 citationsRead paper
FMRP-LEAN: A HIPAA-Compliant AI-Augmented LIMS Architecture for End-to-End Clinical Assay Workflow Optimization
This work addresses the opacity, reporting delays, and compliance risks inherent in clinical biomarker workflows—particularly pronounced in multi-day FMRP assays—stemming from reliance on spreadsheets and manual quality control. The authors propose a HIPAA-compliant, AI-augmented Laboratory Information Management System (LIMS) that models the entire sample lifecycle using a finite state machine, ensuring explicit state representation, controlled transitions, and observable dwell times. A novel MRN-UUIDv7 unified identifier combined with QR-code tracking enables end-to-end traceability under PHI residency constraints. Governance-constrained AI operates exclusively on aggregated projections, complemented by a deterministic fallback mechanism. Built on a hospital-hosted Supabase/PostgreSQL stack with hybrid isolation architecture, the system supports bidirectional REDCap synchronization and secure linkage between clinical and research data. Deployment markedly enhances workflow observability, reduces QC latency, and improves cross-role collaboration transparency.
0 citationsRead paper
Computer Vision Based Neurology Brain Activity Rejection Architecture and Implementation
This study addresses the inefficiency and limited scalability of manual artifact component identification in traditional electroencephalography (EEG) research following independent component analysis (ICA). To overcome this bottleneck, the work introduces computer vision techniques into the automatic labeling of ICA components for the first time, developing an end-to-end automated system compatible with both EEGLAB and ICLabel. The proposed method enables efficient detection and removal of non-neural components, substantially reducing reliance on expert annotation. It achieves a classification accuracy of 89.45% while accelerating processing speed by a factor of 7,200 compared to manual approaches, thereby facilitating large-scale and near real-time EEG analysis.
0 citationsRead paper
Approaching human parity in the quality of automated organoid image segmentation
Existing methods struggle to achieve high-accuracy and consistent automatic segmentation of organoid images across varying experimental conditions. This work proposes a hybrid approach that integrates the general-purpose vision foundation model Segment Anything Model (SAM) with domain-specific segmentation tools, marking the first application of such a combined framework for automated measurement of size and morphology in pluripotent stem cell–derived spheroids. The method delivers stable and accurate segmentation across the majority of tested images, achieving performance on par with or approaching inter-human annotator agreement. This advancement significantly enhances the automation and reliability of organoid image analysis, offering a robust solution for quantitative phenotypic assessment in organoid-based research.
0 citationsRead paper
Recent publications
Latest Papers
Graph-Based Approaches to Learning Epileptogenic Zone Localization Using Stereo-EEG Recordings
研究通过比较不同图拓扑结构，使用基于图的学习模型从静息状态sEEG记录中定位癫痫源区，发现Region-Bridge-c拓扑在减少边数的同时提高了定位精度。
0 citationsRead paper
Deep Learning CNN and Recurrence Analysis for Alpha Gamma EEG Biomarkers in Fragile X Syndrome
Individuals with Fragile X Syndrome (FXS) exhibit aberrant alpha/gamma-band electroencephalographic (EEG) oscillations closely linked to impairments in inhibitory control, sensory processing, and cognition; however, efficient automated methods for identifying reliable biomarkers remain lacking. This study proposes a multimodal deep learning framework that, for the first time, integrates convolutional neural networks (CNNs), long short-term memory networks (LSTMs), and joint modeling of time–frequency and nonlinear dynamical features—including recurrence plot analysis—specifically tailored to alpha and gamma EEG signals. In subject-independent evaluations, the proposed approach significantly outperforms unimodal baselines, with the combined alpha–gamma representation achieving the highest discriminative performance. These results underscore the framework’s potential for facilitating automated FXS identification, diagnosis, and treatment monitoring.
0 citationsRead paper
FMRP-LEAN: A HIPAA-Compliant AI-Augmented LIMS Architecture for End-to-End Clinical Assay Workflow Optimization
This work addresses the opacity, reporting delays, and compliance risks inherent in clinical biomarker workflows—particularly pronounced in multi-day FMRP assays—stemming from reliance on spreadsheets and manual quality control. The authors propose a HIPAA-compliant, AI-augmented Laboratory Information Management System (LIMS) that models the entire sample lifecycle using a finite state machine, ensuring explicit state representation, controlled transitions, and observable dwell times. A novel MRN-UUIDv7 unified identifier combined with QR-code tracking enables end-to-end traceability under PHI residency constraints. Governance-constrained AI operates exclusively on aggregated projections, complemented by a deterministic fallback mechanism. Built on a hospital-hosted Supabase/PostgreSQL stack with hybrid isolation architecture, the system supports bidirectional REDCap synchronization and secure linkage between clinical and research data. Deployment markedly enhances workflow observability, reduces QC latency, and improves cross-role collaboration transparency.
0 citationsRead paper
Computer Vision Based Neurology Brain Activity Rejection Architecture and Implementation
This study addresses the inefficiency and limited scalability of manual artifact component identification in traditional electroencephalography (EEG) research following independent component analysis (ICA). To overcome this bottleneck, the work introduces computer vision techniques into the automatic labeling of ICA components for the first time, developing an end-to-end automated system compatible with both EEGLAB and ICLabel. The proposed method enables efficient detection and removal of non-neural components, substantially reducing reliance on expert annotation. It achieves a classification accuracy of 89.45% while accelerating processing speed by a factor of 7,200 compared to manual approaches, thereby facilitating large-scale and near real-time EEG analysis.
0 citationsRead paper
Approaching human parity in the quality of automated organoid image segmentation
Existing methods struggle to achieve high-accuracy and consistent automatic segmentation of organoid images across varying experimental conditions. This work proposes a hybrid approach that integrates the general-purpose vision foundation model Segment Anything Model (SAM) with domain-specific segmentation tools, marking the first application of such a combined framework for automated measurement of size and morphology in pluripotent stem cell–derived spheroids. The method delivers stable and accurate segmentation across the majority of tested images, achieving performance on par with or approaching inter-human annotator agreement. This advancement significantly enhances the automation and reliability of organoid image analysis, offering a robust solution for quantitative phenotypic assessment in organoid-based research.
0 citationsRead paper