Julia Chivu ’23
The current burn injury evaluation methods available for clinical use are ineffective and outdated. These circumstances are problematic as the initial injury assessment is necessary to ensure the best clinical treatment plan for the patient. Burns can lead to deep tissue damages, necrosis, and severe scarring. Without a proper course of action, the injury can negatively impact the patient’s quality of life, mental health, and mortality. Terahertz time-domain spectroscopy is a technique that can be used in burn examination by delivering short pulses of terahertz radiation to examine the quality of the damaged skin. While this technique can rapidly estimate the severity of a burn in a non-invasive manner, the current machinery associated with this technique is impractical in a clinical setting. Therefore, scientists in the Department of Biomedical Engineering at Stony Brook University have created a new device that utilizes terahertz time-domain spectroscopy in conjunction with an artificial neural network to assess burn injuries and predict their outcomes.
The current terahertz time-domain spectroscopy machinery is often challenging to utilize since the expensive devices are large, bulky, and require difficult optical alignments. To combat these inconveniences, the research team created a handheld terahertz time-domain spectroscopy machine, which they named the Portable Handheld Spectral Reflection (PHASR) scanner. This device allows for more efficient clinical use on patients by providing high speed imaging measurements of a 37 x 27 mm2 field within a few seconds. To account for the double Debye theory (which describes how the skin’s terahertz reflectivity can become altered due to physical changes such as swelling or the alteration of collar fibers or proteins due to high heat exposure), the research team built a two layer artificial neural network by using five parameters from the double Debye dielectric model to automatically assess burn severity. Together, the neural network and the PHASR device were able to capture spectroscopic images and determine the permittivity of burns, which measures how much energy the affected skin can store when under the influence of an electric field. The research team discovered that their neural network classification algorithm was able to estimate the severity of burns with 84.5% accuracy. The model also predicted the healing outcome of burns with 93% accuracy. In addition, the research team concluded that their neural network requires less training data to provide significant results, which can be useful for other algorithm training.
With further testing of this device, the PHASR scanner and the neural network can potentially be used in clinical trials. Ultimately, this innovative technology may allow for better burn healing outcomes due to enhanced and specialized treatment plans.
 M. E. Khani, et al., Triage of in vivo burn injuries and prediction of wound healing outcomes using neural networks and modeling of the terahertz permittivity based on the double debye dielectric parameters. Biomedical Optics Express 14, 918-931 (2023). doi: 10.1364/BOE.479567.
 Image retrieved from: https://unsplash.com/photos/Skf7HxARcoc