News
Article
Author(s):
A prospective study evaluating the efficacy of an artificial intelligence (AI)–based clinical support system further demonstrated the reliability of AI or machine learning as a diagnostic tool in skin cancer.
An artificial intelligence (AI)–based clinical decision support tool showcased great accuracy in the prospective diagnosis of cutaneous melanoma, according to a study recently published in British Journal of Dermatology.
Innovations in AI and machine learning have altered the landscape of skin cancer diagnosis. These models, the present authors mention, have demonstrated impressive accuracy when it comes to detecting and differentiating cancerous skin lesions. Particularly in cases of cutaneous melanoma, early identification is tremendously important for patients’ prognostic outcomes and survival.
Multiple studies have shown the diagnostic power of AI models in diagnosing skin cancer; however, prospective studies on their accuracy have not been sufficiently conducted—especially in primary care, which is where many patients’ initial assessments are conducted. Primary care physicians (PCPs), the present authors add, typically diagnose melanomas with less accuracy compared with dermatologists. Therefore, furthering the study and integration of AI in primary care could provide great benefits to clinicians and patients at large. With this potential in mind, researchers conducted a study to evaluate the diagnostic performance of an AI-based smartphone application (app) that PCPs used to prospectively assess suspicious skin lesions.
The AI-based decision tool, Dermalyser, developed in Sweden, was used by 138 PCPs and implemented across 36 primary care centers (PCCs) throughout the southern regions of Sweden. Study participants were recruited from May to December 2022 and were eligible if they presented at a PCC with 1 or multiple skin lesions suspected to be melanoma. Lesions thought to be basal cell or squamous cell carcinoma were not included in the analysis.
Overall, 253 lesions in 228 patients were included. The mean age of participants was 54 years. Among the presented lesions, there were 194 lesions managed with a dermatologist referral, 54 with a primary excision conducted at the PCC, and 5 with an excision referral to a surgical clinic. In total, 21 melanomas (10 melanomas in situ and 11 invasive melanomas) in 20 patients were detected, most commonly on the posterior torso.
The accuracy of the app to differentiate nonmelanoma and melanoma lesions was measured with the receiver operating characteristic (ROC) curve. The area under the ROC (AUROC) revealed that the capacity of the app to distinguish lesions was 0.960 (95% CI, 0.929-0.980), with an estimated 95.2% sensitivity and a specificity of 84.5%. Additionally, the app had a positive predictive value of 35.9% and a negative predictive value (NPV) of 99.5%. In the detection of invasive melanomas, the app’s AUROC was 0.988 (95% CI, 0.965-0.997) with a sensitivity of 100% and a specificity of 92.6%. The authors mentioned that there were 2 in situ melanomas that were missed by the app; however, these lesions also had a low suspicion from PCPs.
Their results suggest that the vast majority of benign lesion cases, as well as over half of the total lesions, would not have warranted extra assessments or referrals to dermatologists or excisions if PCPs were able to implement the app’s guidance into their practice.
“The app was strictly applied on lesions with some degree of suspicion for melanoma. We believe this enhances the significance of this study’s outcome, as this scenario corresponds to the real-life setting in primary care. Although it is known from previous studies that some melanomas are initially missed by the standard clinical assessment (in primary care, as well as by dermatologists), the solution is not to apply this type of app thoughtlessly on all of a patient’s skin lesions, as this could lead to an unmanageable number of false-positive results. However, the app’s high NPV and ease of use might promote use on lesions with a lower degree of suspicion for melanoma that might best be managed with lesion monitoring,” the authors concluded.
Reference
Papachristou P, Söderholm M, Pallon J, et al. Evaluation of an artificial intelligence-based decision support for detection of cutaneous melanoma in primary care - a prospective, real-life, clinical trial. Br J Dermatol. 2024:ljae021. doi:10.1093/bjd/ljae021
Expert Insights on How Utilization Management Drives Physician Burnout
Clinical Features Vary Across Myasthenia Gravis Foundation Classes