Seyed Yahya Shirazi

How does the brain communicate with muscles during unexpected perturbations, and how does this change with age? We investigated corticomuscular connectivity during perturbed recumbent stepping in young and older adults using high-density EEG and EMG. Young adults demonstrated selective connectivity between error-processing brain regions and specific muscles, with strong involvement of the anterior cingulate cortex. In contrast, older adults showed elevated baseline connectivity and relied on diffuse patterns dominated by motor and posterior parietal cortices, connecting to multiple muscles simultaneously regardless of their biomechanical role. This reveals a strategic reorganization: young adults use dynamic, error-driven control, while older adults employ a stability-focused approach that maintains comparable performance through constitutive hyperconnectivity. These distinct connectivity signatures establish perturbed recumbent stepping as a valuable tool for assessing age-related sensorimotor changes and developing targeted rehabilitation interventions.

The Lab Streaming Layer (LSL) presents a software-based solution for synchronizing data streams across multiple instruments in neurophysiological research. Utilizing per-sample time stamps and LAN-based time synchronization, LSL ensures accurate, continuous recording despite varying device clocks. It automatically corrects for network delays and jitters, maintaining data integrity through disruptions. Supporting over 150 device classes and compatible with numerous programming languages, LSL has become a vital tool for integrating diverse data acquisition systems. Its robustness and adaptability have extended its application beyond research, into art, performance, and commercial realms, making it a cornerstone for multimodal data collection and synchronization.

Explore the theoretical foundations and practical implementation of signal processing techniques for ring-based wearable biosensors. This guide covers digital filter theory, z-transforms, real-time processing constraints, and multi-sensor fusion approaches with both mathematical foundations and practical examples.

Learn how to evaluate AI systems in healthcare using specialized metrics and frameworks that address clinical validity, FDA regulatory requirements, bias detection, safety assessment, and practical implementation strategies. This comprehensive guide provides insights into designing robust evaluation pipelines for health AI applications.

Learn how to apply statistical methods for robust evaluation of models, including power analysis, mixed-effects models, bootstrap confidence intervals, multiple comparison corrections, and effect size calculations. This guide provides practical algorithms and Python code snippets to help researchers ensure their evaluations are statistically sound and meaningful.

Learn about various methods for evaluating large language models (LLMs), including automatic metrics like BLEU and ROUGE, the LLM-as-judge paradigm, human-in-the-loop strategies, and specialized approaches for health-related applications. This comprehensive guide also covers best practices for benchmark design, red teaming, and scaling evaluations.

This post provides a detailed overview of human evaluation and psychometrics in the context of AI systems, covering key concepts, reliability metrics, scale design, and practical implementation strategies. It includes algorithms and code snippets to help practitioners design robust evaluation frameworks.

This dashboard provides comprehensive analysis of dataset citations within the NEMAR ecosystem, revealing collaboration patterns, research trends, and the impact of open neuroscience data sharing on the research community.

Learn how to transform your Hugo static site into an interactive dashboard powerhouse using Chart.js, structured JSON data, and modern web development practices.

This is a data visualization dashboard for exploring the Healthy Brain Network EEG dataset. Features age and sex distributions, task availability metrics, mental health correlations, and per-release analysis across 11 dataset releases with over 3,600 participants.