System-Level Design Trade-Offs and Their Impact on Artificial Intelligence in Functional Near-Infrared Spectroscopy: A Review

Abstract

Functional near-infrared spectroscopy (fNIRS) has become a widely adopted optical neuroimaging modality due to its portability, safety, and suitability for operation outside conventional laboratory environments. While significant advances have been made in signal processing and data analysis, the achievable performance and reliability of fNIRS measurements remain fundamentally governed by system-level hardware design choices. The increasing diversity of custom and application-specific fNIRS platforms highlights the need for an engineering-focused analysis of instrumentation architectures and their associated trade-offs. This paper presents a system-level analysis of contemporary fNIRS instrumentation, with emphasis on optical and electronic design choices including light-source and detector technologies, probe geometry, amplification strategies, power management, and subsystem coupling. Representative system implementations are examined to illustrate how architectural decisions influence signal-to-noise ratio (SNR), depth sensitivity (DS), scalability, and wearability. Particular attention is given to probe-related parameters, such as channel counts, source-detector separation (SDS), and detector integration, and their impact on sensitivity to cortical versus superficial signals. The analysis further demonstrates that several challenges commonly addressed through artificial intelligence (AI)-based post-processing, including limited DS, motion susceptibility, and variability in signal quality, originate from hardware-level constraints. As fNIRS systems continue to expand toward mobile, real-world, and multimodal applications, these constraints become increasingly critical. By clarifying key design trade-offs and system-level limitations, this work aims to support more informed evaluation and development of future optical fNIRS instrumentation.