Contactless vital sign technology has a demographic problem. A 2025 study published in npj Digital Medicine reviewed the most widely used public rPPG datasets and found a pronounced bias toward subjects with lighter skin tones, with darker Fitzpatrick types severely underrepresented across nearly every major dataset. This isn't a footnote — it's a fundamental limitation that undermines the technology's ability to serve the populations that could benefit from it most.
Circadify is taking direct action on this problem. We are announcing that Travor Nkolo has joined the company as Head of Global Health Research, a role created to lead our field deployment efforts in Uganda and drive the data collection work necessary to build an rPPG model that performs equitably across all skin tones.
Travor Nkolo — Head of Global Health Research
"The gap in rPPG performance across skin tones is not a technology problem — it's a data problem. You can't build equitable algorithms from inequitable datasets." — npj Digital Medicine, Demographic Bias in Public rPPG Datasets (2025)
Why this role matters now
The evidence for demographic bias in camera-based vital sign measurement has been building for years. Verkruysse et al. noted signal attenuation from melanin absorption in their foundational 2008 rPPG paper. Nowara et al. (2020) quantified the performance gap, showing heart rate mean absolute error 2-4 times worse for darker skin tones compared to lighter ones. And a 2025 multi-site field study by Dasa et al. in Nigeria — testing rPPG on 306 participants with predominantly Fitzpatrick type V and VI skin — reported a hypertension detection sensitivity of just 0.04. Nearly every case was missed.
The root cause is well documented. The ICCS 2025 review on underrepresentation of dark skin tones in medical imaging datasets found that training data across the field is overwhelmingly composed of lighter-skinned subjects. When algorithms learn from biased data, they produce biased results. Ba et al. (2023) demonstrated the inverse: deep learning rPPG models trained on genuinely diverse datasets narrow the performance gap substantially.
Collecting that diverse data requires boots on the ground. It requires working directly with communities, building trust, navigating local healthcare systems, and capturing physiological measurements under real-world conditions — not controlled lab environments. That's what Travor's role is designed to do.
The Uganda deployment
Circadify has been conducting field work in communities outside Kampala, Uganda, deploying our rPPG model as a limited-release Android APK for contactless vital sign screening. This deployment serves two purposes: providing accessible health screening to communities with limited clinical infrastructure, and collecting the diverse Fitzpatrick skin tone data needed to improve model equity.
Uganda represents a critical testing ground for several reasons:
- Demographic diversity — The population spans Fitzpatrick types IV through VI, precisely the skin tones most underrepresented in existing rPPG datasets
- Healthcare access gaps — With a physician density of roughly 0.1 per 1,000 people (compared to 2.6 in the United States), communities outside major cities often lack basic screening infrastructure
- Mobile penetration — Smartphone adoption is rising rapidly across East Africa, making phone-based health tools a realistic delivery mechanism
- Existing relationships — Circadify's prior field work in the region has established community partnerships and local operational knowledge
Travor will lead this effort on the ground, coordinating data collection protocols, managing community health worker relationships, and ensuring that the data captured meets the quality standards required for model training and validation.
The data collection challenge
| Factor | Controlled Lab Setting | Field Deployment (Uganda) |
|---|---|---|
| Lighting conditions | Standardized, consistent | Variable — indoor, outdoor, mixed |
| Subject movement | Minimal, instructed | Natural, unpredictable |
| Camera hardware | Research-grade | Consumer smartphone cameras |
| Skin tone representation | Typically Fitzpatrick I-III | Fitzpatrick IV-VI |
| Ground truth devices | Clinical-grade reference | Portable reference devices |
| Sample size feasibility | Tens to low hundreds | Potentially thousands |
| Environmental noise | Controlled | Ambient — heat, dust, humidity |
| Cultural context | Neutral | Requires trust-building and local engagement |
Collecting rPPG data in field conditions is harder than collecting it in a lab. The lighting varies. People move naturally rather than sitting still on command. Smartphone cameras introduce sensor variability that research-grade cameras don't. And ground truth measurements from reference devices need to be captured simultaneously under the same imperfect conditions.
But this is exactly the point. An rPPG model that only works under perfect conditions in a controlled lab is not useful for the communities that need it. The field data — messy, variable, and real — is what produces algorithms that actually function in deployment.
Building toward equity
The work Travor is leading fits into a broader trajectory in the field. Researchers worldwide are recognizing that rPPG equity requires intentional effort:
- Diverse dataset initiatives — Groups at UCLA, TU Eindhoven, and several other institutions are building more representative rPPG datasets, though most still skew lighter than the global population
- Skin-tone-adaptive algorithms — Wang et al.'s POS algorithm and newer deep learning approaches can partially compensate for melanin-related signal differences when trained on appropriate data
- Regulatory attention — The FDA has signaled increased scrutiny of demographic performance claims for digital health devices, and the EU AI Act explicitly addresses algorithmic fairness
Circadify's approach is straightforward: go where the data gap is, collect responsibly, and use it to build technology that doesn't leave anyone out. Travor's appointment as Head of Global Health Research is the organizational commitment behind that approach.
What comes next
Travor's immediate priorities include expanding the Uganda field deployment, establishing standardized data collection protocols for diverse skin tones, and building partnerships with local health organizations to ensure the work benefits the communities involved — not just the model.
Longer term, the data collected under his leadership will feed directly into Circadify's model development pipeline, with the goal of demonstrating validated performance across all six Fitzpatrick skin types. The rPPG field has spent years documenting the equity problem. The next phase is solving it — and that requires people willing to do the work where it matters.
Welcome to the team, Travor.
Frequently Asked Questions
What is Travor Nkolo's role at Circadify?
Travor Nkolo serves as Head of Global Health Research, leading Circadify's rPPG deployment efforts in Uganda and overseeing data collection initiatives to improve model performance across diverse Fitzpatrick skin types.
Why does rPPG need diverse skin tone data?
Camera-based vital sign measurement relies on detecting subtle light absorption changes caused by blood flow beneath the skin. Melanin content affects this signal, and most existing rPPG datasets are heavily skewed toward lighter skin tones. Without representative data from Fitzpatrick types IV-VI, algorithms cannot be validated or optimized for darker-skinned populations.
Where is Circadify deploying its rPPG technology?
Circadify has been conducting field work in communities outside Kampala, Uganda, deploying its rPPG model as a limited-release Android application for contactless vital sign screening in low-resource health settings.
How does diverse data collection improve rPPG accuracy?
Training and validating rPPG algorithms on datasets that proportionally represent all Fitzpatrick skin types allows the model to learn melanin-related signal variations and compensate for them. Research shows that models trained on diverse data significantly narrow the performance gap between lighter and darker skin tones.