Featured video: Watch the full WHS Education Committee webinar on YouTube
Presenter:
- Shawn Tejiram, MD | Burn Surgeon, The Burn Center at MedStar Washington Hospital Center; Assistant Professor of Surgery, Georgetown University School of Medicine
Moderator:
- Shannon Sapp Layton, PhD | Associate Professor, School of Nursing, University of Alabama at Birmingham; Member, WHS Education Committee
Why This Webinar Matters for Wound Care Clinicians
In part 1 of this series {link}, Dr. Tejiram covered the fundamentals of burn assessment, resuscitation, and transfer criteria. This follow-up moves into the tools and technologies that are reshaping how burn providers evaluate injuries, make surgical decisions, and manage fluid resuscitation.
The central problem? Even experienced burn surgeons correctly assess burn depth only about 75% of the time. For providers who don’t see burns regularly, that number drops to roughly 50%. Inaccurate depth assessment in either direction carries clinical consequences: underestimation delays surgery and prolongs healing, while overestimation results in excision of tissue that may have recovered without intervention.
That accuracy gap is what the diagnostic tools covered in this session are designed to close. Some are already in clinical use, while others are still in trials, though all of them point toward a future where burn assessment becomes more objective, more accessible, and less dependent on having a seasoned burn surgeon at the bedside.
Burn Depth and Size: A Quick Review
Burn depth determines the clinical path. First-degree burns affect the epidermis only and resolve with supportive care. Second-degree or partial-thickness burns extend into the dermis, produce blisters, remain painful, and typically heal within two weeks. Third-degree or full-thickness burns destroy all skin layers including adnexal structures (hair follicles, sweat glands), resulting in dry, insensate, leathery tissue that requires surgical excision. Fourth-degree burns extend past skin into muscle, fascia, and bone, most commonly seen in electrical injuries, and frequently require amputation.
Burn size is expressed as a percentage of total body surface area (TBSA) and drives two critical decisions: how aggressively to resuscitate and what mortality risk a patient is facing.
Two standard methods for estimating TBSA:
- Rule of Nines: Each arm accounts for 9%, the head for 9%, and everything else (anterior torso, posterior torso, each leg) for 18%. Genitalia account for 1%.
- Palmer method: A patient’s palm is estimated at 1% TBSA. This is more practical for patchy, discontiguous burns where the Rule of Nines is hard to apply cleanly.
Burn Wound Conversion: The Zone of Stasis Problem
Every burn wound has three concentric zones of injury:
- Zone of coagulation: The center of the injury, in direct contact with the heat source. This tissue is dead and must be removed.
- Zone of stasis: The surrounding area with compromised but potentially viable tissue. This is the clinical target, because it can either recover or convert to necrosis depending on how well it is supported.
- Zone of hyperemia: The outermost zone, characterized by increased blood flow from the inflammatory response. This area generally recovers.
If the zone of stasis converts to necrosis, the wound becomes larger which increases morbidity and mortality. The entire purpose of the imaging modalities discussed in this webinar is to identify which tissue is at risk of conversion before it happens, so providers can intervene earlier and more precisely.
Laser Doppler Imaging (LDI)
Laser Doppler imaging has been in clinical use for decades and is probably the most established advanced burn diagnostic tool currently available. It works by scanning a wound bed with a laser to measure blood perfusion. The output is a color-mapped image: red indicates high perfusion (tissue likely to heal within two weeks), while yellow, green, or blue indicates low perfusion (tissue unlikely to heal, likely needing surgery).
Dr. Tejiram’s group at MedStar has published on protocolized LDI workflows that incorporate the imaging into assessment, debridement, hydrotherapy, dressing decisions, and admission or discharge planning.
Key clinical details:
- LDI is FDA-cleared for use between 48 hours and 5 days after injury.
- Use within the first 48 hours may be unreliable because the hyperinflammatory response immediately following burn injury can distort perfusion readings.
- The device requires a dedicated workstation, trained operators, and workflow integration, all of which carry cost and labor implications.
There are however some limitations, such as financial cost of the workstation, training requirements for off-hours coverage (the 3:00 a.m. admission scenario where the burn surgeon is not in-house), and the learning curve for interpreting results accurately.
ICG Fluorescence Angiography
Indocyanine green (ICG) fluorescence angiography is an intraoperative tool. A clinician injects ICG dye and uses immunofluorescence imaging to map tissue perfusion at the wound bed during surgery. The technique is similar in concept to ICG cholangiography used in biliary surgery.
The application in burns is to guide excision margins in real time: if an area is well-perfused, it may not need to be excised. If it is poorly perfused, it likely does. After excision and grafting, the same technique can be used to track wound bed perfusion over time.
This is not a bedside tool. It requires operating room infrastructure and ICG-compatible imaging equipment.
Infrared Thermography
Infrared thermography is a non-invasive modality that maps oxygen levels and temperature across the wound bed. The concept is similar to LDI in that well-oxygenated tissue correlates with adequate blood flow and healing potential, while poorly oxygenated tissue may be at risk.
Dr. Tejiram described a portable device connected to an iPad with associated software, making it more accessible than an LDI workstation. Published literature shows it has demonstrated early detection of tissue at risk of necrosis, which supports earlier intervention.
Spectral AI’s DeepView System
This system is a handheld AI-powered device that scans burn wounds to assess both depth and TBSA, designed specifically to improve burn assessment in settings where experienced burn providers are not available.
A note on regulatory status: At the time of this webinar (November 2025), the device’s regulatory status was described as recently FDA-cleared. The product, developed by Spectral AI (formerly Spectral MD), received FDA Breakthrough Device Designation in 2018 and submitted its De Novo 510(k) application in June 2025. Providers interested in this technology should verify its current regulatory status before clinical use.
MedStar Washington Hospital Center was one of the institutions involved in the training study, where burn providers used the device and participated in “truthing sessions” to validate and train the AI’s assessments.
Ultrasound-Guided Burn Depth Assessment
Dr. Mohamed El Masry, who chairs the WHS Education Committee and was present during the webinar, has published research on ultrasound-based burn depth assessment. His group’s approach uses two ultrasound modalities:
- B-mode (standard ultrasound): Standard grayscale imaging of tissue structure.
- Elastography: Measures tissue elasticity. Necrotic or fibrotic tissue is stiff and less elastic. Inflamed tissue is softer and more elastic. The elastography output produces a color map similar in concept to LDI, where color patterns correspond to tissue condition across skin layers.
Dr. El Masry’s group validated their AI model against histological biopsy (the gold standard for burn depth) and reported strong diagnostic accuracy, as measured by area under the curve, for identifying burn depth and predicting surgical need. His initial work used a knowledge-driven reasoning framework called CODER, and subsequent work integrated digital photographs with Doppler imaging into a multimodal AI agent.
This remains early-stage research that needs larger multicenter datasets for clinical validation, but it represents a direction where widely available equipment (ultrasound) could be paired with AI to provide burn depth assessment without specialized imaging devices.
Large Language Models for Burn Depth Assessment
Dr. Tejiram’s group tested whether commercially available large language models (LLMs) could assess burn depth from deidentified photographs. Using a 92-image dataset, they compared clinician assessments to ChatGPT assessments delivered through both the standard web interface (GUI) and the application programming interface (API).
Findings:
- Burn clinicians were still more accurate overall than the LLM.
- The API interface was more accurate than the GUI interface for burn depth classification.
- The results suggest that prompting strategy matters significantly. How you frame the question to the LLM affects the quality of the answer, a finding consistent with the broader AI literature.
This is not a clinical recommendation to use ChatGPT for burn diagnosis. It is a proof-of-concept study exploring whether widely accessible AI tools could eventually serve as decision support in settings without burn specialists.
AI for Burn Resuscitation Guidance
The second major AI application Dr. Tejiram discussed was using LLMs to guide fluid resuscitation, specifically as an alternative to specialized tools like Burn Navigator in resource-limited settings.
Burn Resuscitation Background
Burns exceeding roughly 15 to 20% TBSA trigger a systemic inflammatory response that overwhelms the body’s compensatory mechanisms. The inflammation causes vascular permeability (leaky blood vessels), loss of intravascular fluid and protein into the interstitial space, reduced cardiac output, and inadequate organ perfusion, a condition known as burn shock.
The American Burn Association consensus formula (modified Brooke formula) calls for 2 mL/kg/% TBSA over 24 hours, which is half the volume of the older Parkland formula (4 mL/kg/% TBSA). The calculated volume is divided into an hourly starting rate and then titrated up or down based on end points of resuscitation, primarily urine output, but also base deficit and lactate at some institutions.
Burn Navigator
Burn Navigator is a clinical decision support tool originally developed by the U.S. military (U.S. Army Institute of Surgical Research) to help non-burn-experienced providers manage fluid resuscitation during transport from austere environments. It received 510(k) FDA clearance in 2013.
The tool provides hourly fluid rate recommendations and tracks cumulative fluid volume against benchmark curves: the consensus formula line, the Parkland formula line, and the IV index (the threshold above which complications like compartment syndrome, pulmonary edema, and ARDS become significantly more likely).
Published multicenter studies led by Jose Salinas, PhD, and colleagues have shown that Burn Navigator use was associated with lower incidence of burn shock, comparable 28-day mortality, and overall lower fluid volumes compared to institutional protocols without decision support.
LLM as a Resuscitation Tool
Dr. Tejiram’s group developed a prompting strategy where they fed hourly burn diagnostics (TBSA, weight, presence of shock, blood pressure, glucose levels) into ChatGPT and compared the LLM’s fluid titration recommendations to Burn Navigator’s recommendations.
Results from a Bland-Altman analysis showed that approximately 95% of the LLM’s hourly fluid recommendations were statistically close to Burn Navigator’s recommendations. The greatest discordance occurred at moderate fluid volumes (500 mL to 1 L per hour), and the team attributes this to areas where their prompting strategy needs refinement.
An important caveat: LLM versions change frequently (the study was conducted on GPT-4.0; newer versions may produce different results), and there is no way to pin results to a specific model version on a specific date. Reproducibility across LLM updates remains an open problem.
This work is being presented at a future American Burn Association meeting.
Omics and Biomarker Research
The final section of the webinar moved into the earliest-stage research: applying multi-omics analysis (transcriptomics, proteomics) to burn care with the goal of identifying patient-specific biomarkers that could guide individualized treatment.
Dr. Tejiram’s group has examined transcriptomic (RNA) pathways in patients with severe burns (20% TBSA or greater) and found modulation of both immunologic and inflammatory pathways that varied with injury severity. The hope is that identifying specific shutdown or activation pathways could lead to targeted interventions for acute management and long-term outcomes, including burn wound sepsis.
A separate group from China identified a specific protein, S100, as predictive of burn sepsis in a multi-omics analysis.
Challenges identified:
- Omics analysis requires very large datasets, and burn centers individually do not generate enough data.
- The analysis depends on machine learning and deep learning techniques that require significant compute infrastructure.
- Integration with electronic medical records, data governance, HIPAA compliance, and informed consent are all unresolved barriers.
- Most proof-of-concept work to date has been in animal models. Large-scale randomized controlled human studies are needed before clinical application.
The current state of omics in burns today is where AI and LLMs were a few years ago: promising but very early.
The Data Problem in Burn AI
During the Q&A, a question that provoked one of the most candid exchanges of the webinar was asked: where is the data going to come from?
Dr. Tejiram identified several obstacles to building the large, validated image datasets that burn AI tools need:
- HIPAA and consent: Burn images carry higher identification risk than other clinical data because visible features like tattoos, body location, and skin characteristics can identify patients even in deidentified datasets.
- Image quality: Clinical photographs are taken by residents, nurses, and research staff, not professional photographers. Image quality is inconsistent.
- Lack of context: Some studies crop images to just the burn, which removes information about the overall affected area and limits TBSA estimation.
- Existing datasets are unreliable: Dr. Tejiram’s group attempted to use publicly available deidentified burn image databases and found images mislabeled as burns that were actually rashes or other conditions.
- No centralized repository exists: There is no burn-specific equivalent of large standardized medical imaging datasets. Dr. Tejiram suggested the American Burn Association could potentially create one, similar to their existing BQCP registry data.
In any case, multicenter data collaboration will be essential. Without it, individual institution datasets are too small to train AI tools with the accuracy clinical use demands.
What MedStar Is Using Today
When asked directly which of these technologies are part of his current clinical practice, Dr. Tejiram listed:
- LDI: Used when there is a question about burn depth, particularly for admission/discharge decisions.
- Burn Navigator: Used for essentially all burn resuscitations at this point.
- Ultrasound: Used in research for various burn diagnostics, but not yet for routine depth and size assessment.
- Spectral AI (DeepView): Participated in the training study but has not introduced it into regular clinical practice.
Watch the Full Webinar
The full recording, including slide visuals, imaging examples, and the extended Q&A with Dr. El Masry and Dr. Layton, is available on the WHS YouTube channel and in member-only accounts on the WHS website.
This article is a summary of a WHS Education Committee webinar presented November 2025, featuring Dr. Shawn Tejiram, moderated by Dr. Shannon Sapp Layton. Dr. Mohamed El Masry, chair of the WHS Education Committee, contributed to the discussion. It is intended as a clinical reference for wound care professionals and does not replace formal medical training or clinical judgment. The WHS Education Committee does not endorse specific commercial products.