Abstract
Recent advancements in neural decoding have enabled the reconstruction of visual imagery from high-resolution fMRI data. However, the bulky nature of fMRI limits its practical application. This study explores the feasibility of using portable Time-Domain functional Near-Infrared Spectroscopy (TD-fNIRS) as a wearable alternative. By analyzing the spatial resolution requirements for image reconstruction and simulating optical performance, the researchers identify a 1-cm resolution threshold necessary for high-fidelity decoding. They further demonstrate that TD-fNIRS can achieve this target, providing a roadmap for affordable, portable "mind-reading" technology.
Introduction
High-fidelity visual decoding—reconstructing what a person sees or imagines—has traditionally required the millimeter-scale resolution of functional Magnetic Resonance Imaging (fMRI). While effective, fMRI machines are massive, expensive, and non-portable. Alternatives like EEG offer portability but lack the spatial resolution (typically 2–3 cm) required for complex image reconstruction. Functional Near-Infrared Spectroscopy (fNIRS) offers a middle ground, measuring blood oxygenation through infrared light. This research investigates whether fNIRS can reach the spatial resolution necessary to bridge the gap between lab-based fMRI and wearable brain-computer interfaces (BCIs).
Determining the Minimum Viable Resolution
To establish an engineering target for portable systems, the authors performed computational downsampling experiments using the Natural Scenes Dataset (NSD). High-resolution 7T fMRI data was blurred to various scales (1 cm, 1.5 cm, and 2 cm) and processed through MindEye, a reconstruction model utilizing CLIP and Stable Diffusion.
The results revealed a significant "performance cliff." While retrieval accuracy remained high at 1-cm resolution (71% top-1 accuracy), it collapsed to just 20% at the 2-cm mark. This finding establishes 1 cm as the critical spatial resolution threshold for meaningful visual decoding.
Time-Domain vs. Continuous-Wave fNIRS
The study then addressed whether fNIRS could meet this 1-cm requirement. Most current wearable fNIRS devices use Continuous-Wave (CW) technology, which measures constant light intensity. The authors compared this against Time-Domain (TD) fNIRS, which measures the time-of-flight of individual photons.
Simulations using the Radiative Transfer Equation (RTE) demonstrated that TD-fNIRS significantly outperforms CW-fNIRS. While CW systems are generally limited to ~2-cm resolution, TD-fNIRS can resolve inclusions at the 1-cm scale. This is attributed to "late-arriving" photons, which travel deeper into the cortex and provide more localized spatial information. Furthermore, TD-fNIRS showed the ability to detect inclusions at depths of up to 4 cm, doubling the effective range of standard optical systems.
Hardware Prototyping and Economic Feasibility
A key goal of the research was to demonstrate that high-performance TD-fNIRS hardware can be built affordably. The team designed a prototype using a gain-switching laser, Avalanche Photodiode (APD) detectors, and a Time-to-Digital Converter (TDC) implemented on a low-cost FPGA board. They estimate that a 10-channel TD-fNIRS system could be constructed for approximately $2,500—a fraction of the cost of commercial optical systems or fMRI infrastructure.
Conclusion and Future Work
This work provides a clear path toward portable, high-fidelity neural decoding. By identifying the 1-cm resolution requirement and demonstrating the technical superiority of Time-Domain fNIRS, the authors have set the stage for the next generation of wearable BCIs. Future efforts will focus on refining hardware components, such as laser drivers for narrower pulse widths, and collecting large-scale fNIRS datasets to train reconstruction models directly on optical signals.
Summary updated