Dive Brief:
- Spectral AI’s Deepview System has outperformed burn physicians in identifying non-healing tissue, the company said Monday.
- Deepview achieved 86.6% sensitivity when identifying non-healing tissue at an image-wise level of assessment, while the clinical judgment annotation of burn physicians scored 40.8%. The study also conducted assessments at the pixel-wise level.
- BTIG analysts called the result a “major milestone” in a Monday note to investors. Spectral AI plans to file the results with the Food and Drug Administration by mid-2025 and launch the product in 2026.
Dive Insight:
Spectral AI’s system uses a physical device to capture images. The device sends multiple wavelengths of light and detects the light reflected back by the tissue. By applying artificial intelligence algorithms to the images, Spectral AI’s system predicts whether tissue will heal from a burn or not.
The burn validation study assessed images from 164 patients, including 49 children, treated for burns in the U.S.
The trial compared Deepview to the clinical judgment of burn physicians. At the pixel-wise level, sensitivities for Deepview and physicians were 81.9% and 38.8%, respectively.
Deepview also delivered better dice scores, a method for comparing the similarity of two samples. The AI-enabled system performed at 68.5%, compared with 39.2% for physicians. Spectral AI said the results represented “improved pixel-wise evaluation between predicated and true segmented wound areas.”
Specificity, a measure of the identification of true negatives, was one area where physicians performed better than Deepview. Image-wise specificity in segmenting non-healing wound areas was 61.2% for the AI-enabled system, compared with 79.1% for physicians. Spectral AI said the result reflected physicians’ conservative assessment of burn areas.
“We believe these results validate [Spectral AI’s] thesis and development efforts and provide a major proof-point for investors on the potential of the DeepView technology in a variety of addressable markets,” BTIG analysts said. “Overall, [Spectral AI] is progressing against key milestones.”
Spectral AI plans to apply for FDA authorization via the de novo pathway by the end of June and receive a decision by early next year. The company said its current manufacturing arrangements position it to start commercializing the product as soon as it receives approval.
Michael DiMaio, chair of the Spectral AI board, outlined the company’s plans for commercialization on an earnings call in November.
“The [Biomedical Advanced Research and Development Authority] contract actually has some clauses in there to help us begin to kickstart the sales force for the device,” DiMaio said. The contract also positions Spectral AI “to get approximately 200 devices via the government pathways into the burn centers across the United States,” DiMaio added.