Terahertz Time-Domain Spectroscopy Gains
How THz-TDS paired with neural network can assess burn severity with 93 percent accuracy
The first triage of a burn injury is crucial in formulating successful clinical treatment plans. But with accuracy of burn depth assessment at 60–75 percent, researchers at Stony Brook University decided to develop a more accurate device. The result? The Portable Handheld Spectral Reflection (PHASR) Scanner, which relies on terahertz time-domain spectroscopic (THz-TDS) imaging to assess burn injuries. We asked the lead author, Hassan Arbab, from Stony Brook’s Department of Biomedical Engineering, to tell us more about the project.
How are burn injuries currently assessed?
Clinical evaluation and triage of burn injuries at emergency departments involve subjective visual and tactile examinations. It is very difficult for clinicians to determine which burn wounds will heal on their own and which will require excision and grafting. This method has proved to be unreliable – with accuracy rates hovering around 60–75 percent despite standardization attempts. Various modalities have been developed to address this known issue, but long acquisition times, technical challenges, and high costs have limited their use in clinical settings. And that’s why we decided to seek an objective and accurate technological solution.
How did you approach the problem?
We had proposed the emergence of terahertz time-domain spectroscopy (THz-TDS) as a promising technique for non-invasively sensing biological tissues – including the assessment and monitoring of burn injuries. But early demonstrations of THz-TDS for diagnosing burn severity were limited to point spectroscopy measurements, which did not account for heterogeneity and spatial variations in burns. Additionally, typical THz spectroscopy setups are bulky, expensive, and require cumbersome optical alignments – making them impractical for clinical use in real-world settings. To address these challenges, we developed the PHASR (Portable Handheld Spectral Reflection) Scanner – a fast fiber-coupled device for hyperspectral imaging of in vivo burn injuries using THz-TDS.
Your research also generated an artificial neural network classification algorithm…
THz-TDS provides access to the full dielectric constant of the tissue in an unexplored part of the spectrum – between 0.1 and 3 THz. There are three main physical changes from burn injuries that affect THz reflectivity. The first of these is the formation of interstitial edema – a standard inflammatory tissue response that contains the spread of the thermal insult. It is mostly made of water and associated with increased THz reflectivity and absorption in deep layers of skin. The second change is the presence of dermal adnexal structures, scattering THz waves and correlating with the severity of burns and THz reflectivity measurements. The last change occurs in the chemical makeup and dielectric properties of biomolecules, such as proteins and collagen fibers. This can also affect THz reflectivity, but is difficult to isolate from the other two sources of THz signal contrast.
To investigate the mechanisms that change the complex dielectric function of cutaneous burns in terahertz frequencies, we employed the double Debye theory – a successful method that explains the interaction of THz radiation with various biological tissue types. Here, we used it to explain the contrast in refractive indices between burns of different depths. We showed that the five parameters of the double Debye model can be used to create an artificial neural network classification algorithm, which predicts the ultimate healing outcome of in vivo burns with 93 percent accuracy. Our findings suggest that the Debye dielectric parameters offer a physics-based approach for extracting biomedical diagnostic markers from broadband THz pulses – reducing the dimensionality of THz training data for AI models. Indeed, our technique reduced the number of input variables in the neural network model – what is usually 100s and 1000s of data points per pixel of a terahertz full-spectroscopic image is now only five Debye parameters, improving the efficiency of machine learning algorithms for processing large data sets obtained over large clinical trials. In short, our approach allows complex analysis without complex machine learning.
How long before we could see your device being used by clinicians?
Well, we are continuing to improve the speed and accuracy of our imaging system and working towards a pilot human study and an expanded clinical study in the coming years. A new version of our PHASR Scanner will debut this year, which can image the same field of view in 2–3 seconds. We are also working on other skin conditions and cancer imaging, as well as diagnosing diseases in the cornea that similarly change the spectroscopic information due to changing hydration of the tissue.
Outside of the clinic, an important future project for our lab involves non-destructive testing in industrial applications, such as in-line quality control. Our compact and handheld PHASR Scanners are ideally suited to bring terahertz spectral imaging to analytical control of fabrication processes and quality control of products in chemical manufacturing and the pharmaceutical industry.
Credit: Conny Schneider / Unsplash.com
Associate Editor, The Analytical Scientist
Georgia Hulme is Associate Editor at The Analytical Scientist