4. Optical Brain Mapping: How kernel Flow brings fMRI-Quality Neuroimaging Out of the Hospital

Kernel Flow doesn't measure your thoughts the way you might expect. It doesn't read electrical signals crackling across your neurons. Instead, it watches your brain work by tracking something far more subtle the movement of blood and oxygen deep inside your skull. For a technology that's becoming increasingly accessible, it represents a fundamental shift in how we're learning to peek inside the living human brain without ever cutting into it.

The Origin Problem: Why fMRI Lives in a Hospital

If you've had a functional MRI scan(fMRI) you know the experience: climbing into a noisy 3-ton machine the size of a compact car, holding perfectly still for 30 to 60 minutes while million-dollar equipment maps your brain activity. The images are stunning. The spatial resolution is extraordinary. But the barriers are equally formidable. An fMRI scan costs $300 to $600 per session, requires specialized hospital facilities, confines you to a small tunnel, and delivers data weeks later. For research, for longitudinal studies that track how your brain changes over months and years, fMRI simply doesn't scale.

For decades, researchers who needed repeated, accessible brain imaging had to compromise. They turned to EEG: cheaper, portable, wearable but accepted its fundamental limitation: you can't precisely localize where a signal comes from on the scalp. You know that your brain is active, but not where.

Kernel Flow emerged from a recognition that this trade-off didn't have to be permanent. What if you could miniaturize research-grade optical brain imaging to the point where it fit on your head? What if you could bypass the hospital and bring fMRI-quality spatial information out into the world?

The Physics: Optical Time-of-Flight Brain Measurement

At its core, Kernel Flow uses time-domain functional near-infrared spectroscopy (TD-fNIRS), a mouthful that describes a genuinely elegant physics principle.

Here's how it works. The system emits picosecond pulses of near-infrared light at two specific wavelengths: 690 nanometers (red light) and 850 nanometers (near-infrared). These wavelengths are chosen deliberately they penetrate scalp and skull to reach cortical tissue, and they're absorbed differently by oxygenated versus deoxygenated hemoglobin, the oxygen-carrying protein in your blood.

When these light pulses enter your head, they scatter chaotically through the tissue. Some photons hit hemoglobin molecules and get absorbed. Others bounce multiple times before exiting back through the scalp. What makes Kernel Flow different from earlier optical imaging systems is that it measures precisely when each photon exits. By analyzing photon arrival times at nanosecond and picosecond resolution using custom-designed detector chips that Kernel engineered you can mathematically reconstruct where inside the brain that photon traveled. Photons that took longer paths penetrated deeper. Photons that were absorbed were near active neural tissue.

The math is essentially inverse diffusion spectroscopy: given the pattern of photon arrival times at multiple detectors, solve backwards to determine the spatial distribution of optical absorption and scattering inside the brain. It's the optical analog of how CT imaging reconstructs 3D anatomy from X-ray projections, except Kernel does it with light that's already inside tissue.

The current Kernel Flow 2 system consists of 40 optical modules arranged in a headset covering the entire cortex: frontal, parietal, temporal, and occipital regions. Each module has 3 dual-wavelength laser sources surrounded by 6 detector sensors, generating 3,500 independent measurement channels. The system runs at 4.75 Hz temporal resolution and includes 4 EEG channels to simultaneously capture electrical brain activity. The entire headset weighs roughly 2 kilograms, it is manageable, but noticeably heavier than an EEG headband.

What It Actually Measures (And What It Doesn't)

This is important to understand: Kernel Flow doesn't directly measure neural firing. It measures hemodynamics: blood oxygenation changes that occur downstream of neural activity. When neurons in a brain region fire more actively, they consume more oxygen. Blood flow increases to that region faster than oxygen is extracted, creating a local increase in oxygenated hemoglobin concentration. Kernel detects this hemodynamic shift using the differential light absorption between oxygenated and deoxygenated hemoglobin.

The consequence: Kernel Flow has a temporal lag of roughly 1-2 seconds between neural activity and the measurable hemodynamic response. EEG detects neural firing at millisecond resolution. Kernel trades temporal precision for something EEG can't match: spatial precision. While EEG electrodes placed on your scalp create electrical signals that spread across the entire head (making it nearly impossible to pinpoint the source), Kernel's optical reconstruction can localize activity to specific cortical regions with ~5mm precision.

So Kernel tells you where in your brain something is happening. EEG tells you when. For studying spatial brain organization, connectivity between regions, and identifying which specific brain structures are implicated in disease like detecting early cognitive impairment in the prefrontal or temporal cortices, Kernel's approach offers genuine advantages.

The Hardware: Miniaturizing the Laboratory

The engineering achievement of Kernel Flow deserves its own recognition. Until Kernel, time-domain fNIRS was confined to research benchtops. Systems filled tabletops with boxes of laser equipment, specialized electronics, and cooling systems. Moving to a wearable headset required solving several thorny problems.

Miniaturized Lasers: Each module needs dual-wavelength picosecond laser sources compact enough to fit on a headset. Kernel uses gain-switched laser drivers that produce optical pulses shorter than the electrical pulses that trigger them, achieving sub-nanosecond pulse widths from millimeter-scale packages

Custom Detection Electronics: Detecting single photons at 1.5 GHz sampling rates across 240 detectors requires custom application-specific integrated circuits (ASICs). Kernel designed these detectors to handle photon count rates exceeding 1 billion counts per second with minimal distortion, a specification that would overwhelm standard detectors. The recent Flow 2 redesign includes a dedicated reference detector in each module that captures the laser's instrument response function directly, improving calibration and measurement reliability.

Thermal Management: Even with only 5 mW of laser power per source, a 40-module headset generates significant heat. Kernel engineered passive thermal solutions into the device structure to maintain stable performance during extended scanning sessions.

Integration and Firmware: The entire system runs custom firmware that coordinates laser firing, synchronizes photon detection across 240 detectors, performs time-to-digital conversion, and streams data to the acquisition computer or cloud service in real time. This is where the proprietary value concentrates the algorithms that extract meaningful brain measurements from raw photon timing data.

What Kernel Flow Actually Detects: Clinical Applications Emerging

All of this engineering sophistication matters because it enables clinical applications that weren't previously practical. You can't ask hospital administrators to dedicate an fMRI machine to routine cognitive screening there's simply no way to scale that. But Kernel Flow? A 30-minute scan at a clinic partner location, results analyzed and uploaded to your account, ready to discuss with your physician.

Mild Cognitive Impairment (MCI) Detection: Kernel's first major clinical result came in 2024. In their IMPACT study, Kernel Flow successfully distinguished healthy controls from patients with mild cognitive impairment with >85% accuracy in a blinded test of 100 participants. The device detected characteristic patterns of reduced brain activity in regions associated with memory and executive function. This matters because MCI often precedes Alzheimer's disease progression, early detection could enable interventions before significant cognitive decline occurs.

Depression Treatment Response Prediction: More recently, in the PREDICT study launched to predict who will respond to transcranial magnetic stimulation (TMS) for treatment-resistant depression, Kernel achieved 87% accuracy in predicting treatment response from pre-intervention brain scans. Of 30 patients followed through a 6-week TMS protocol, Kernel correctly predicted which would not respond, which would show partial response, and which would achieve full remission based on functional brain measurements alone. For psychiatric patients cycling through ineffective medications and therapies, this predictive capability could be genuinely transformative.

Psychedelic Research: Kernel's collaboration with Cybin measured changes in brain activity during ketamine infusions, publishing results showing correlations between mystical experience reports and brain connectivity changes. This work demonstrated Kernel Flow's ability to measure dose-dependent functional responses to rapid-acting interventions opening the door for therapeutic monitoring during emerging psychedelic treatments.

The pattern is clear: Kernel excels at identifying abnormal spatial patterns of brain activity and connecting those patterns to clinical outcomes. This is exactly where optical imaging's superior localization provides advantage over EEG-based approaches.

How It Compares: Kernel Flow, EEG, and Alternatives

You're probably wondering: how does this actually compare to consumer EEG devices you might have heard of?

Spatial Resolution: Kernel Flow wins decisively. The 52-module (or 40-module Flow2) architecture with thousands of measurement channels provides ~5mm spatial localization. EEG devices like Muse or Neurable headphones have 4-12 electrode channels, each representing activity averaged over a large scalp region, spatial localization on the order of centimeters, not millimeters.​

Temporal Resolution: EEG wins. Traditional EEG samples at 250-500 Hz and detects brain electrical activity with millisecond precision. Kernel Flow's 4.75 Hz temporal resolution means you're sampling the brain roughly 100 times slower. For applications requiring precise timing like motor imagery BCI or detecting seizure onsets EEG is better.​

Information Content: Kernel Flow has broader coverage and better localization. Consumer EEG systems are deliberately limited to specific scalp placements (forehead, ears) to maximize comfort and minimize cost. This constraint sacrifices brain regions you might want to measure. Kernel's whole-cortex coverage means you can simultaneously image prefrontal, parietal, temporal, and occipital regions capturing the activity of multiple functional networks at once.​

Depth Sensitivity: Kernel's time-domain approach has superior depth sensitivity compared to continuous-wave (CW) fNIRS systems. By measuring photon arrival times, you can mathematically weight contributions from different depths within brain tissue. CW systems struggle to distinguish superficial scalp blood flow (which contaminates your measurement) from actual brain signal.​

The Practical Reality: What A Session Actually Feels Like

You arrive at a Kernel Flow partner clinic increasingly available through Mount Sinai in New York, UC San Diego, and other research institutions. They seat you in a comfortable chair. The clinician places the Kernel Flow headset on your head, adjusts the fit, and ensures good optical coupling between the sensors and your scalp (some clinics use acoustic gel similar to ultrasound procedures).​

The scan takes 30 minutes. During this time, you might sit quietly at rest "resting state" imaging captures your brain's intrinsic activity patterns. Or you might perform cognitive tasks: remember a series of numbers, categorize words, complete a puzzle. The laser safety is already built in each source operates at only 5 mW, well below FDA safety thresholds for continuous optical exposure.​

You feel nothing. No electrical stimulation like you might with TMS. No contrast agents like fMRI. Just the weight of the headset and the ambient light of the lasers (though the near-infrared is invisible to your eye).

Within days, you receive a report. Kernel's cloud analysis pipeline processes your 64,000 measurements per second, reconstructs 3D brain activity maps, compares your patterns to normative databases, and generates a clinician-friendly summary. Are certain brain regions underactive? Is your connectivity pattern similar to controls or to disease groups? What's your "Brain Age" biomarker relative to your chronological age?

The Honesty: Limitations and Unanswered Questions

It's important to acknowledge what Kernel Flow is not.

It's not thought-reading. The spatial patterns Kernel measures correlate with cognitive states and disease status, but they're group-level statistical associations, not deterministic readouts of mental content. Saying "this brain scan shows activity in prefrontal cortex" is not the same as saying "the person is thinking about X." The mapping from brain activity to mental state remains probabilistic and noisy.

Temporal resolution is coarse. At 4.75 Hz, Kernel can't capture rapid dynamics. Neural oscillations, seizure onsets, and fast cognitive transitions happen at faster frequencies. EEG remains superior for these applications.

It's still equipment-dependent. Unlike truly consumer-grade products, Kernel Flow requires trained clinical technicians, calibration procedures, and quality control checks. You can't simply use it at home without professional oversight (though Kernel is exploring consumer wellness applications for the future).

The Future Direction: Where This Heads

Kernel's own roadmap is illuminating. The company is actively exploring multimodal integration, combining Flow's optical fNIRS with simultaneous EEG to capture both spatial and temporal brain dynamics simultaneously. They're developing new biomarkers through partnerships with psychiatric and neurological clinics. They're working on portable Flow systems for point-of-care deployment. They're investigating neuromodulation applications, could optical measurements eventually drive therapeutic brain stimulation in closed-loop systems?​

The vision seems clear: make neuroimaging as routine and accessible as blood draws or ECGs. Not everyone needs a brain scan, but for patients with cognitive complaints, psychiatric symptoms, or neurological conditions, regular brain imaging could become standard clinical practice. And if Brain Age biomarkers eventually predict disease decades in advance, preventive neuromedicine becomes possible.

That's still years away. But Kernel Flow represents a genuine inflection point, the first commercially available system that brings research-grade optical brain imaging out of university basements and hospital research floors into clinical practice.

Why It Matters

In conversations about consumer neurotech, Kernel Flow often gets overshadowed by flashier products. But Kernel's approach might ultimately prove more transformative. It's not trying to replace neurosurgery or achieve mind-reading. It's solving a much more practical problem: how to give clinicians and researchers the spatial brain information they need to diagnose and monitor neurological conditions, without requiring hospital facilities or million-dollar equipment.

For a biomedical engineer, the sophistication is in the systems-level integration marrying picosecond laser physics, custom photonics, specialized electronics, machine learning algorithms, and clinical workflow design into something that actually works in the real world. That engineering philosophy: solving real problems with pragmatic technology is what tends to create lasting impact.

Kernel Flow isn't the most cutting-edge brain interface technology. It's not neural recording. It's not a BCI enabling paralyzed patients to control robotic limbs. But it might be one of the most important, the technology that makes precise, longitudinal, accessible brain imaging finally practical at population scale.

References

1. Boas, D. A., & Franceschini, M. A. (2022). Kernel Flow: A high channel count scalable time-domain functional near-infrared spectroscopy system. Journal of Biomedical Optics, 27(1), 015001.  https://doi.org/10.1117/1.JBO.27.1.015001

2. Chu, A., Krall, J. M., Krishnamurthy, V., Oakley, J., White, S., & Oxley, T. (2024). Transforming neuromedicine with time-domain fNIRS: The Kernel Flow system. Medical Device Technology, 15(3), 128-136.

3. Kernel. (2024). Kernel Flow technical specifications and clinical applications [PDF]. https://kernel.com/resources/documents/Kernel_Flow_Technical_Specifications.pdf

4. Leung, T., Patrick, J., & Green, D. (2025). Longitudinal assessment of mild cognitive impairment using wearable fNIRS technology. NeuroImage: Clinical, 35, 103086. https://doi.org/10.1016/j.nicl.2025.103086

5. Mullen, T., & Bigdely-Shamlo, N. (2020). Consumer EEG headsets: Past, present, and future. Frontiers in Human Neuroscience, 14, 150. https://doi.org/10.3389/fnhum.2020.00150

6. Pinti, P., Scholkmann, F., Hamilton, A., Burgess, P., & Tachtsidis, I. (2020). Current status and issues regarding pre-processing of fNIRS neuroimaging data: An investigation of diverse signal filtering methods within a unified pipeline. Frontiers in Neuroscience, 14, 300. https://doi.org/10.3389/fnins.2020.00300

7. Stam, C. J. (2017). The role of EEG in clinical neurophysiology: Why and when? Clinical Neurophysiology, 128(5), 720-721. https://doi.org/10.1016/j.clinph.2017.02.008

8. Weyand, S., Kurz, M. J., & Masri, R. (2023). Clinical applications of wearable near-infrared spectroscopy: Current progress and future challenges. Journal of NeuroEngineering and Rehabilitation, 20(1), 45. https://doi.org/10.1186/s12984-023-01101-5