ISSUE 020 · Oct 22, 2026 · 9 min read

If we knew today which burns would not heal within 21 days, could we change care and transfer sooner?

Move multispectral AI beyond depth classification into a pathway linking burn-specialist review, transfer, early excision or grafting, and safe observation—then test whether decisions come sooner and avoidable transfers fall without more missed nonhealing burns.

01Clinical problem · Evidence

Indeterminate partial-thickness burns sit between delay and overtreatment.

Early visual assessment is subjective and burn wounds evolve. Late recognition of nonhealing tissue can delay grafting and increase infection, stay, and scar burden; overcalling can create unnecessary transfer, surgery, and donor-site harm.

In non-burn centers, access to specialist judgment is itself a bottleneck.

02Technology stack · Proposal

Design what happens after the scan—not merely the scan.

Combine standardized wound preparation, noncontact multispectral imaging, AI 21-day healing prediction, structured photography and TBSA, remote burn-specialist review within 30 minutes, 24–48-hour reassessment, and reserved transfer capacity. A negative result never ends clinical observation; a positive result never orders surgery.

03Research question · Hypothesis

Does an accuracy signal translate into better timing and transfer?

In indeterminate partial-thickness thermal burns, test whether AI plus specialist review shortens time to the correct management plan—adjudicated by healing or surgery through day 21—and reduces tertiary transfer among wounds that heal by day 21, while keeping any increase in missed nonhealing wounds within 3 percentage points.

04First study · Proposal

Twenty referring EDs, eight burn centers, 900 patients, cluster crossover.

Enroll adults and children within 72 hours of an indeterminate partial-thickness thermal burn covering 0. 5–15% TBSA.

Assign eight-week blocks to standard tele-burn care or the same resources plus multispectral AI. Co-primary endpoints are superiority in time to the ultimately correct management plan and noninferiority within 3 percentage points for missed nonhealing wounds.

Measure transfer among wounds healed by day 21, surgery and donor-site harm, time to epithelialization, infection, 90-day scar and function, cost, and performance by skin tone, age, and site.

05Defensible data · Hypothesis

The moat is not the spectrum—it is the timeline of healing and changed decisions.

Link raw spectra, calibration and device, objective skin color, site, depth and TBSA, serial photographs, AI regions, specialist decisions and override reasons, transfer, surgery and pathology, epithelialization, scar, and cost across pixel, patient, and site. Keep independent trial adjudication separate from model-training labels.

06Venture thesis · Assumption

The first customers are burn networks and referring emergency departments.

Pair an annual hospital subscription—capture quality control, specialist queue, transfer SLA, outcome audit, and training—with per-scan pricing rather than selling hardware alone. The value proposition is fewer avoidable transfers and revisits plus earlier treatment of nonhealing wounds.

Payer outcome contracts remain a hypothesis until safety and total-cost savings are shown prospectively.

Key numbers

Between an authorized adjunct and an unproven care pathway

May 21, 2026

FDA DEN250028 De Novo authorization date

≤21 days

Authorized healing-potential horizon under conservative care

80.7% / 95.5%

Sensitivity / specificity in a 2025 UK multicenter evaluation

58% vs 90%

MSI versus laser Doppler accuracy in a separate 31-wound pilot

Evidence and interpretation

Authorization opens the door; it does not prove pathway effectiveness

Established: FDA granted Class II De Novo authorization to the DeepView AI System on May 21, 2026. Its software highlights deep partial- and full-thickness thermal burn tissue and areas unlikely to heal within 21 days with conservative treatment alone, but it is an adjunct to clinical assessment—not a standalone diagnosis or a precise excision-margin tool.

A 2025 UK multicenter observational evaluation reported 80. 7% sensitivity and 95.

5% specificity, with wide uncertainty and no randomized test of pathway or patient outcomes. In a separate 31-wound pilot, MSI matched clinical management in 58% of cases versus 90% for laser Doppler imaging, underscoring device- and setting-dependence and the need for independent validation.

Registered study NCT06131203 collected device-validation data; it was not a randomized test of a service model improving transfer, surgery, scar, or cost. The proposed study size, time and transfer effects, and business model are ANM hypotheses.

FDA De Novo authorization letter UK multicenter original article MSI–laser Doppler comparison article U.S. burn validation registry

Failure conditions and safety, ethics, and regulatory risks

Clinical and economic failure

Do not scale if time to the correct plan falls by less than 24 hours, missed nonhealing wounds cross the 3-point noninferiority margin, or avoidable transfer among wounds healed by day 21 falls by less than 20%. Imaging accuracy alone is not success.

False negatives and delayed care

A false negative can delay grafting and worsen infection or scarring. Maintain specialist escalation regardless of output when pain, perfusion, infection, wound evolution, or clinician concern warrants it, with reassessment at 24–48 hours.

False positives and unnecessary surgery

A highlighted region is not an excision boundary. Prohibit AI-only transfer or surgery; require burn-surgeon confirmation, an option for conservative reassessment, and audit of surgical and donor-site harms.

Skin-tone, site, and capture bias

Prespecify performance by skin tone, age, anatomical site, residual dressings, blistering, eschar, lighting, and distance. Failed quality checks trigger preparation and recapture rather than inference; eye-adjacent and other unvalidated complex sites follow separate protocols.

Consent, image privacy, and access inequity

Wound images may contain identifying body marks and pediatric data. Separate consent for care, audit, and model training; minimize data and define encryption and deletion periods.

Audit transfer and outcomes by language, distance, insurance, and geography so deployment does not benefit only large networks.

Regulatory scope, change control, and accountability

DEN250028 authorizes adjunctive wound assessment, not autonomous surgery or transfer. New anatomical sites, sensors, models, alert language, automated prioritization, or treatment recommendations may require additional regulatory review.

Contracts must define version lock, drift surveillance, standard care during downtime, and the accountable clinician.

Primary sources

References

Established facts are separated from research and venture hypotheses that still require validation.

01
U.S. Food and Drug Administration. DeepView AI System, De Novo Classification Order.
Primary regulatory source for the 21-day healing-potential output, Class II classification, adjunctive use, and limitations against standalone diagnosis or excision-margin use.DEN250028 · May 21, 2026
02
Tan P, Nyeko-Lacek M, Walsh K, Sheikh Z, Lewis CJ. Artificial intelligence-enhanced multispectral imaging for burn wound assessment: Insights from a multi-centre UK evaluation. Burns. 2025;51(8):107550.
A prospective multicenter observational evaluation at two UK burn centers. It suggests performance potential but has limited sample precision and does not test pathway effectiveness.PMID 40858093 · DOI 10.1016/j.burns.2025.107550
03
Perusseau-Lambert A, Akram M, Frew Q, Barnes D. Comparison between multispectral imaging and laser Doppler imaging to predict burn wound requirements for surgery. Burns. 2025;51(8):107650.
A 31-wound pilot reporting 58% accuracy for MSI and 90% for LDI, supporting the need for independent validation and clinician integration.PMID 40834474 · DOI 10.1016/j.burns.2025.107650
04
ClinicalTrials.gov. Burn Validation Study: DeepView SnapShot Portable.
A registered adult and pediatric thermal-burn study collecting training and validation data. Registration and device validation do not establish effectiveness of the proposed treatment and transfer pathway.NCT06131203

All issues