Near‑infrared spectroscopy, or fNIRS, offers a way to monitor brain activity without surgery or radiation by tracking changes in blood flow and oxygenation. Light sources placed on the scalp send near‑infrared light into the head, and detectors measure the light that scatters back. Because this light must pass through the scalp and skull before reaching the brain, the measured signal always includes a mix of superficial and cerebral contributions. Separating those signals has long been a central challenge for fNIRS researchers.
In a study published in Biophotonics Discovery , researchers from the Tufts University Diffuse Optical Imaging of Tissue Laboratory show that combining a specific source–detector geometry with a simple, anatomically informed tissue model can substantially improve how fNIRS data are interpreted. By accounting for how light travels through layered head structures, the approach makes it possible to better isolate brain‑specific signals without relying on complex imaging systems or subject‑specific MRI scans.
Why geometry and modeling matter
Traditional fNIRS measurements often use a single distance between a light source and a detector. While easy to implement, this setup is highly sensitive to blood flow changes in the scalp and skull. More advanced methods can separate surface and brain signals, but they often require many light sources and detectors, dense sensor arrays, and heavy computation.
The approach tested in this study uses a special configuration with two light sources and two detectors placed at different distances on the scalp. This arrangement produces multiple measurements, each with a different balance of sensitivity to superficial and deep tissue. By combining these measurements—a method known as the dual‑slope approach—the technique reduces the influence of surface tissue and enhances sensitivity to the brain.
Even with this improved geometry, the data still need a model to translate measured light changes into meaningful estimates of brain activity. Many fNIRS studies rely on a simplified “homogeneous” model that treats the head as a single, uniform tissue. While convenient, this assumption ignores known anatomical layers and can distort the results.
Testing layered head models
To address this limitation, the researchers compared two simplified but more realistic head models: a two‑layer model and a three‑layer model. In the two‑layer case, the head is represented as superficial tissue overlying brain tissue. In the three‑layer case, an additional middle layer represents the cerebrospinal fluid that surrounds the brain.
The team used Monte Carlo simulations to track how light propagates through these layered structures. The simulations covered a wide range of tissue thicknesses and optical properties that reflect biological variability. The goal was to see which model could reproduce the patterns observed in real fNIRS measurements.
To validate the simulations, the researchers also collected in vivo data from healthy volunteers. During the experiment, participants viewed a visual stimulus designed to activate the occipital cortex. The researchers measured changes in light intensity and phase using frequency‑domain fNIRS with the dual‑slope source–detector configuration. Ultrasound imaging was used to estimate each participant’s scalp and skull thickness, providing an independent measure of superficial anatomy.
The role of cerebrospinal fluid
The results showed a clear advantage for the three‑layer model. Only when the model included a low‑scattering, low‑absorbing layer representing cerebrospinal fluid did the simulated data reproduce the main qualitative features of the human measurements.
This middle layer plays an outsized role in shaping how light travels through the head. Even though it is thin, cerebrospinal fluid changes the relative sensitivity of different fNIRS measurements to brain tissue. In the three‑layer model, differences between subjects could be explained primarily by variations in scalp and skull thickness, which aligns with known anatomy.
By contrast, the two‑layer model could match the experimental data only by assuming large and biologically unlikely differences in tissue scattering properties between individuals. This finding suggests that explicitly representing cerebrospinal fluid is important for realistic modeling of light transport in the head.
Separating surface and brain signals
Beyond reproducing experimental trends, the three‑layer model enabled the researchers to estimate how much of the measured signal came from superficial tissue versus the brain. When applied to the visual stimulation data, the analysis showed that the detected responses were dominated by cerebral changes, with minimal contribution from the scalp.
This result is consistent with prior work showing that visual tasks produce strong, localized brain responses with relatively small systemic effects at the surface. More importantly, it demonstrates that a modestly complex model can help distinguish these contributions using standard fNIRS measurements.
Toward practical brain monitoring
The authors emphasize that layered models are still simplifications of real head anatomy. They do not capture fine spatial detail or localized brain activity patterns. However, the study shows that moving from a homogeneous model to a three‑layer model provides a meaningful improvement without greatly increasing complexity.
By pairing a two‑source, two‑detector configuration with a three‑layer tissue model, the approach avoids the need for large sensor arrays or subject‑specific MRI scans. This makes the method more practical for settings where simplicity, portability, and ease of use are critical.
In practical terms, the work points toward more reliable noninvasive brain monitoring in clinical environments, at the bedside, or in naturalistic settings outside the laboratory. By clarifying how light interacts with layered head tissues, the study helps bring fNIRS closer to its goal of accessible, accurate measurement of brain activity.
For details, see the original Gold Open Access article by J. Frias et al., “ Functional dual-slope frequency-domain near-infrared spectroscopy data interpreted with two- and three-layer models, " Biophoton. Discovery 3(2), 025003 (2026), doi: 10.1117/1.BIOS.3.2.025003
Biophotonics Discovery
Not applicable
Functional dual-slope frequency-domain near-infrared spectroscopy data interpreted with two- and three-layer models
24-Mar-2026