AI-Powered Elastic Scattering Spectroscopy Device Improves Melanoma Evaluations

Primary care physicians significantly improved their accuracy in identifying potentially malignant skin lesions when using a handheld, noninvasive electrical impedance spectroscopy (ESS) device powered by artificial intelligence (AI), new data suggest.1

These findings resulted from a recent web-based reader study, authored by investigators such as Elizabeth V. Seiverling, MD, from the Department of Dermatology at the University of Pittsburgh in Pennsylvania. Seiverling and coauthors highlighted recent AI technology advances made in melanoma detection and such technologies’ potential for primary care physicians.2

“Here, we describe the findings of a multireader multicase (MRMC) study to assess the referral performance of PCPs when evaluating lesions suggestive of melanoma with and without the aid of the ESS device,” Seiverling et al wrote.1 “Since previous studies were not adequately powered or designed to demonstrate the device impact on melanoma management, this study was accomplished with intentional inclusion of more melanoma cases.”

Design and Findings on ESS Device Use

The investigators looked at a handheld, non-invasive device under investigation, utilizing ESS combined with machine learning (ML) to assist clinicians in the assessment of suspicious skin lesions. Development of the device’s algorithm had implemented a dataset comprising more than 10,000 ESS recordings gathered from over 2000 skin lesions. These included both malignant and benign lesions.

The spectral data used for algorithm training, Seiverling and coauthors highlighted, were entirely separate from the dataset utilized in the current study. By doing this, the investigators ensured none of the training recordings would be included in the testing phase. The ESS device produces, for each lesion analyzed, a binary classification result: lesions exhibiting malignant features are labeled “investigate further,” and those showing benign characteristics are given a “monitor” notice.

In the case of “investigate further” lesions, the AI-powered device was also designed to generate a spectral similarity score with a range from 1 - 10. In this range, higher scores correspond to a stronger resemblance between the lesion’s spectral profile and those of lesions known to be malignant from the algorithm’s development set.

In Seiverling and colleagues’ investigation, they used a web-based, multi-reader, multi-case (MRMC) reader study design aimed at the detection of melanoma. Primary care physicians took part in a total of 200 readings per physician, and this was done across 100 unique lesion cases. Each such case was presented 2 times, with the initial only involving standard clinical information and digital images, and the second with the addition of the ESS device output.

Two major aspects of decision-making among clinicians were explored in this analysis: first, whether a lesion needs to be referred to a dermatologist for any additional assessments, and second, whether a lesion is judged to be malignant or benign. There were 118 board-certified internal and family medicine physicians who took part in this analysis, with each assessing 50 malignant and 50 benign lesions.

Seiverling et al’s study yielded a total of 5900 assessments conducted without ESS data and another 5900 conducted with the ESS device’s assistance. In addition to diagnostic performance, those evaluated as patients in the study were surveyed by the team regarding their confidence in clinical decision-making when implementing this AI-powered tool.

The analysis successfully met its primary endpoint, with the area under the receiver operating characteristic curve (AUROC) for physicians utilizing the device reached .671 (95% confidence interval [CI]: .611– .732), as opposed to an AUROC of .630 (95% CI: 0.582–0.678) when unaided by the device. This was noted by Seiverling and colleagues to be a statistically significant improvement (P = .036).1

Overall, these findings demonstrated enhancements, as a result of access to the ESS device data, in terms of the diagnostic accuracy of physicians. When asked whether this device provided additional value to clinical decision-making, 91.5% of physicians taking part either agreed or strongly agreed, suggesting a high level of perceived clinical utility.1 In short, the data suggest the integration of ESS technology with machine learning may significantly aid primary care physicians in the early identification and management of lesions potentially indicative of melanoma.

“Given the rising rates of skin cancer in the US and limited access to dermatologic care, this device has the potential to make a significant impact in skin cancer detection in the primary care setting,” Seiverling and colleagues concluded.1

References

1. Seiverling E, Shah A, Siegel D, et al. Enhancing Diagnostic Precision in Primary Care: A Multireader Multicase (MRMC) Study of an AI-Powered Handheld Elastic Scattering Spectroscopy Device for Informed Referral Decisions in Melanoma Evaluation. J Clin Aesthet Dermatol. 2025;18(10):59–65.

2. Gareau DS, Correa da Rosa J, Yagerman S, et al. Digital imaging biomarkers feed machine learning for melanoma screening. Exp Dermatol. 2017;26(7):615-8.