AI Models Continue to Gain Clinicians’ Attention

A closer look at the latest FDA list of approved AI-enabled devices suggests these digital tools are poised to have a significant impact on patient care.

By John Halamka, M.D., President, Mayo Clinic Platform and Paul Cerrato, MA, senior research analyst and communications specialist, Mayo Clinic Platform

There was a time in the not-too-distant past when physicians and nurses paid little attention to AI/ML-based medical devices, viewing them as novelties not useful bedside tools. That has drastically changed in the last five years. Consider the latest list of such medical devices, updated in May 2024 and published by the FDA. It includes devices that were marketed via 510k clearance, DeNovo request, and premarket approval.

The majority of these digital tools are focused on radiology, with others that meet the needs of clinicians in neurology, dentistry, cardiology, and surgery. One new device that has gained attention is DermaSensor, which was approved in January to help physicians distinguish between normal skin tissue and skin cancer. It’s a prescription device, indicated for the evaluation of skin lesions suggestive of melanoma, basal cell carcinoma, and/or squamous cell carcinoma in patients aged 40 and over, and designed to help determine whether to refer a patient to a dermatologist. The FDA states that the device should be used in conjunction with the totality of clinically relevant information from the clinical assessment, including visual analysis of the lesion, by physicians who are not dermatologists. The device should be used on lesions already assessed as suspicious for skin cancer and not as a screening tool, and it shouldn’t be used as the sole diagnostic criterion nor to confirm a diagnosis of skin cancer.

In the past, we have pointed out the weaknesses of many AI-enabled devices because their approval has been based on retrospective studies. But DermaSensor, which includes a handheld sensor about the size of a cell phone, underwent a multicenter prospective blinded trial to evaluate its ability to detect melanoma.

An FDA-cleared Eko Health stethoscope that has the ability to detect low ejection fraction has also gained attention in the healthcare community. The AI-assisted stethoscope has been on the market for quite some time but in March 2024, the agency cleared ELEFT, the company’s Eko Low Ejection Fraction Tool.  It’s defined as reduced Ejection Fraction machine learning-based notification software; it employs machine learning techniques to suggest the likelihood of a reduced ejection fraction for further referral or diagnostic follow-up for possible detection of heart failure. The technology behind ELEFT was developed in conjunction with Mayo Clinic.

The research supporting the enhanced digital stethoscope was conducted by Patrik Bachtiger, MBBS,  with National Heart and Lung Institute and Centre for Cardiac Engineering, Imperial College London, London, UK, and his colleagues and published in The Lancet Digital Health. It consisted of an observational prospective multicenter study using the same convolutional neural network  and ECG combination that was used by Mayo Clinic investigators in previous clinical trials and made available through Anumana, Inc. Bachtiger et al found that placing the stethoscope at the pulmonary valve position resulted in the best performance, with AUROC of 0.85, sensitivity 85%, and specificity of 69.5%. They concluded: “A deep learning system applied to single-lead ECGs acquired during a routine examination with an ECG-enabled stethoscope can detect LVEF [left ventricular ejection fraction] of 40% or lower.”

Digital stethoscopes and skin cancer algorithms join a long list of devices that are slowly transforming patient care, and attracting serious investment dollars. By one estimate, the software as a medical device (SaMD) market reached $1.02 billion in 2022 and may exceed $3 billion by 2029. There are also models that may help detect breast cancer, including Koios DS Breast, an FDA-cleared software that can be integrated into PACS workstations to improve malignancy risk assessment, Medisafe, designed to help improve patient adherence to their medication regimen, and many more.

These digital tools present both a challenge and an opportunity. On the one hand, they challenge clinicians to assess the worth of AI-enabled tools and find ways to incorporate them into their workflow, but on the other hand, they offer the potential to ease clinicians' workloads and improve patient outcomes.