Tech & Gear General Endurance · · 10 min read

Muscle Oxygenation (SmO2) for Threshold Training: Can a NIRS Sensor Replace a Lactate Meter?

A NIRS sensor finds your second threshold with ICC 0.80 in cycling — but 48% of athletes show zero SmO2 breakpoint on a treadmill, and limits of agreement span ±38 W. Here's the honest verdict.

AO
AthleteOS Data Science
TL;DR — The Answer

Portable NIRS sensors (Moxy, Train.Red) can reliably detect the second threshold in cycling — pooled ICC 0.80 across 344 athletes — but fail for the first threshold (ICC 0.53) and in running (ICC as low as 0.23). Nearly half of athletes produce no identifiable SmO2 breakpoint at all. Use NIRS to confirm and monitor your second threshold; confirm with at least one blood lactate test first.

About half the athletes who try a NIRS sensor on a treadmill see no breakpoint at all. Not a blurry one. Not an ambiguous one. Nothing. That’s the finding most NIRS marketing skips over — and it’s the most important thing to understand before you spend $700 on a Moxy.

The answer to the title question is: partly. For roughly 50–65% of cyclists doing a step test, a NIRS sensor can detect the second threshold with good reliability. For the first threshold, running, and a large subset of athletes, it can’t.

What SmO2 Actually Measures

SmO2 (skeletal muscle oxygen saturation) is the percentage of hemoglobin and myoglobin in your muscle that is carrying oxygen at any given moment. The technology is called NIRS (near-infrared spectroscopy) — the sensor shines near-infrared light through your skin into the muscle beneath, and reads how much is absorbed by oxygenated vs deoxygenated blood.

Think of it like a fuel gauge for your muscle’s oxygen supply. When demand outpaces delivery, the needle drops. A sharp drop at a specific intensity is the signal coaches look for.

That signal sounds clean. In practice, it isn’t always.

How Two SmO2 Breakpoints Map to LT1 and LT2

In a step test on trained cyclists and triathletes (N=40), SmO2 drops in three distinct stages as intensity climbs. From baseline to Fatmax, it falls about 16%. From Fatmax to VT1, it falls another 16%. Then from VT1 to VT2, it plunges 45% — the biggest drop of all (Guerrero-Calderón et al., 2023).

That 45% cliff is the signal. When you see it on a graph, you’ve found your second threshold.

ROC analysis on the same cohort put the cut-offs at:

Those are population averages for trained cyclists. Your numbers may differ. Body composition, muscle fiber type, and sensor placement all shift the absolute value.

SmO2 Response During a Cycling Step Test (Stylized) 7 25 43 61 79 SmO2 (%) RestZ1Z2FatmaxVT1VT2Max SmO2 (% saturation)
Stylized values based on Guerrero-Calderón et al., 2023 (n=40 cyclists/triathletes). The steep drop between VT1 and VT2 is the visual signature of second-threshold detection.

What the Meta-Analysis Numbers Really Mean for SmO2

A 2023 systematic review pooled 15 studies and 344 athletes (Sendra-Pérez et al.). The headline numbers:

ICC of 0.80 sounds strong. Here’s where it gets honest.

That meta-analysis carries 86% heterogeneity — meaning the results vary enormously depending on sport, device, and method. An ICC from cycling studies alone runs 0.91–0.97. For running, it falls as low as 0.23. Average the two and you get 0.80. That average hides the sport-specific split.

The individual limits of agreement matter more for training prescription. In NCAA Division I female rowers (Eserhaut et al., 2025), SmO2 breakpoint 2 vs. lactate threshold 2 showed a group bias of just −5.76 W — which looks fine. But individual limits of agreement spanned −38.52 to +22.25 W. That’s a total spread of about 61 W.

For zone prescription, 61 W is the difference between a Zone 3 tempo and a VO2max interval.

This doesn’t make NIRS useless. It means you can’t skip the one blood lactate co-validation.

NIRS Reliability by Sport (Second Threshold ICC) Cycling (step test) 0.91–0.97 Nordic skiing ~0.85 Rowing 0.67 Running (variable) 0.23–0.92 Pooled ICC across sports for second-threshold (LT2/VT2) detection. Cycling is excellent; running varies widely. Sources: Sendra-Pérez et al. 2023, Eserhaut et al. 2025, Forot et al. 2026.

The Sport-Specific Split: Cycling vs Running

NIRS works in cycling because your legs stay in roughly the same position throughout the effort. The sensor sits on your vastus lateralis, motion artifact is low, and the signal is clean.

Running is different. Every footstrike moves the sensor. Sweat loosens the strap. The gastrocnemius fires explosively instead of smoothly. These factors drive SmO2 variability at high intensity to a coefficient of variation of 43.8% — versus just 11% at low intensity (Feldmann et al., 2023). At exactly the moment you need a clean signal (near threshold), the noise is worst.

NIRS is a cycling tool that also works in rowing and skiing. For running, treat the data as directional, not prescriptive.

The Athlete Who Sees Nothing

Here’s the finding almost no review article covers: 10 of 21 subjects in a treadmill study (Baiget et al., 2023) showed no identifiable SmO2 breakpoint at all. Not at VT1. Not at VT2. The curve just drifted downward with no inflection.

Zero.

That’s 48% of participants with completely uninterpretable data.

In a separate cycling study, 3 of 10 athletes also showed no clear second-threshold inflection. The technology is blind for a meaningful minority of athletes.

Why? Two main causes. First, adipose tissue thickness. The layer of subcutaneous fat between the sensor and the working muscle absorbs the NIRS signal before it reaches muscle. Fat doesn’t desaturate during exercise the way muscle does. A thicker layer means the sensor reads a mixture of fat and muscle — blurring the breakpoint. Adipose tissue explains about 80% of SmO2 variance at peak exercise (Feldmann et al., 2016). If you have more than 10–12 mm of fat at the sensor site, your readings are significantly compromised.

Second, individual physiology. Some athletes’ respiratory and cardiac systems become the limiting factor at threshold before the quadriceps desaturate sharply. The breakpoint exists physiologically but doesn’t show up at the vastus lateralis.

A multi-muscle approach helps. World-class Nordic skiers (N=52) achieved 96.1% breakpoint detection rates when sensors were placed on both biceps femoris and biceps brachii simultaneously (Forot et al., 2026).

Moxy vs Train.Red: What’s Actually Different

The two popular consumer NIRS devices aren’t reporting the same thing.

FeatureMoxy MonitorTrain.Red FYER 2.0Train.Red PLUS
Price (approx.)$650–$850~€699~€1,799
Wavelengths4 (680/720/760/800 nm)1 (850 nm)1 (850 nm)
Output typeAbsolute SmO2 (0–100%)TSI (relative index)TSI (relative index)
Reporting rateStandard10 Hz100 Hz (6 spatial channels)
Peer-reviewed studiesMany (most sport science NIRS studies)FewFew
Key strengthValidated absolute scale; most studiedDeeper hardware pedigree (Artinis)High-resolution spatial data

The unit difference matters. Moxy reports absolute SmO2 on a validated 0–100% scale. Train.Red reports a Tissue Saturation Index derived from a single 850 nm wavelength. They trend together but the absolute numbers don’t match. You can’t apply Moxy’s 26% VT2 cut-off to Train.Red data. Each device needs its own calibration.

How to Run the Step Test

The 5-1 cycling step test is the validated protocol for NIRS threshold detection.

Setup:

The rest periods allow partial resaturation. This “refresh” between stages makes the SmO2 curve step-shaped rather than one smooth descent — and clean steps make breakpoints easier to identify visually.

Analyze SmO2 averaged over the final 60–90 seconds of each stage. Apply double linear regression or look for the inflection visually.

Avoid ramp tests for NIRS threshold detection. SmO2 lags behind power output. On a 20 W/min ramp, the curve never has time to stabilize, and the breakpoint shifts earlier than the true threshold.

A Mini Case Study: What the Data Looks Like in Practice

James is 41, a Category 4 cyclist with five years of structured training. He did a blood lactate step test in March: LT1 at 215 W, LT2 at 285 W. He bought a Moxy and repeated the protocol in April.

His SmO2 curve showed a clear inflection around 230 W (step 4) and a sharp drop through step 5, bottoming near 26% at 280 W. The NIRS-detected second threshold was 280 W versus 285 W from blood lactate — 5 W off.

Eight weeks of base work later, he ran the same step test. Now the sharp drop came at 310 W. SmO2 at the same old threshold power of 285 W had risen to 33%, meaning he was using less oxygen at that intensity. His muscle delivery capacity had improved.

That’s where NIRS earns its keep: not the single snapshot, but the trend over weeks. The watch shows the number. The trend shows the adaptation.

Where SmO2 Genuinely Replaces a Lactate Meter

Use NIRS without co-validation when:

Don’t rely on NIRS alone when:

How AthleteOS Uses Your SmO2 Data

AthleteOS ingests live SmO2 streams from Moxy and Train.Red sensors via its ANT+/BLE pipeline. During session analysis, it overlays SmO2 on concurrent power and heart rate data from the same workout. A two-breakpoint algorithm runs on step-test data to detect SmO2BP1 (aligned with LT1) and SmO2BP2 (aligned with LT2) — no finger prick needed.

Detected thresholds feed directly into your zone model and update upcoming plan workouts. If you’ve validated your NIRS thresholds with blood lactate and the numbers diverge, AthleteOS flags the discrepancy and prompts a manual override.

That’s the model the research supports: NIRS as a high-frequency monitoring layer, with periodic blood lactate as the ground truth. They’re complementary. Neither replaces the other entirely.

For more on threshold methods, read FTP testing protocols and which one matches your event and how Zone 2 builds the aerobic base. If you’re a triathlete, power meter selection for multisport athletes covers the hardware side of threshold tracking. Start your free AthleteOS account to connect your Moxy or Train.Red and see your SmO2 overlaid on power in your next session analysis.

Frequently Asked Questions

Can a Moxy or Train.Red sensor replace a blood lactate test for finding my FTP?

Not as a first-pass test. Pooled ICC for second-threshold detection is 0.80 in cycling — good, but individual limits of agreement span roughly ±38 W. Best practice: do one blood lactate co-validation first, then use NIRS for ongoing monitoring.

What SmO2 percentage corresponds to Zone 2 ceiling and threshold?

In trained cyclists and triathletes, ROC analysis places VT1 at roughly ≤34% SmO2 and VT2 at roughly ≤26% SmO2 (Guerrero-Calderón et al., 2023). These are population averages; individual calibration is essential.

Why does my SmO2 never show a clear breakpoint?

Nearly 50% of athletes produce no identifiable inflection in treadmill studies (Baiget et al., 2023). Main causes: adipose tissue masking the signal, sensor placement, or individual physiology where respiratory muscles become the limiter before leg muscles desaturate sharply.

How does body fat at the sensor site affect my readings?

Adipose tissue thickness explains up to 80% of SmO2 variance at peak exercise. If you have more than 10–12 mm of subcutaneous fat at the sensor site, your readings will underestimate the true desaturation response near threshold.

Is a Moxy reading directly comparable to a Train.Red reading?

No. Moxy reports absolute SmO2 on a validated 0–100% scale using a 4-wavelength model. Train.Red FYER reports TSI (Tissue Saturation Index) from its 850 nm system — a relative measure. You can't apply Moxy cut-offs to Train.Red data without device-specific calibration.

Does NIRS work better for cycling than running?

Yes, significantly. Cycling ICC for second-threshold detection runs 0.91–0.97. Running ICC ranges from 0.23 to 0.92 — often poor for the first threshold. Motion artifact is the main culprit in running.

#SmO2#NIRS#lactate-threshold#threshold-training#Moxy#Train-Red#cycling-zones

See your SmO2 and power curves on the same chart

AthleteOS overlays your Moxy or Train.Red SmO2 stream on concurrent power and heart rate to detect your second threshold without a finger prick. Connect your sensor and let the data find your zones.

Generate Your Free AI Plan
14-day free trial · No credit card required