Accuracy Meets Efficiency: How AI Can Enhance Dermatologists' Diagnostic Performance

Original Article: Assessment of Diagnostic Performance of Dermatologists Cooperating with a Convolutional Neural Network in a Prospective Clinical Study

What are the key takeaway points of this article?

The use of artificial intelligence (AI) in healthcare is rapidly advancing, with one promising application being the use of convolutional neural networks (CNNs) in the field of dermatology. CNNs are a type of AI model specifically designed to process data with a grid-like topology, such as an image, making it an ideal tool for tasks like detecting patterns and abnormalities in medical imaging. A recent study conducted by researchers at the University of Heidelberg sought to investigate whether dermatologists could benefit from cooperation with a market-approved CNN in classifying melanocytic lesions. The results were impressive.

Overall, the investigators involved 22 dermatologists who assessed 228 suspicious melanocytic lesions in 188 patients. Among the lesions, 190 were nevi, and 38 were found to be melanomas. In this study, the CNN achieved a higher specificity and diagnostic accuracy, as well as a comparable sensitivity, compared to dermatologists alone in classifying melanocytic lesions. When dermatologists cooperated with the CNN, their diagnostic performance significantly improved, reducing unnecessary excisions of benign nevi by 19.2%.

The CNN was able to assess one dermoscopic image per lesion, obtaining a mean sensitivity of 81.6% and specificity of 88.9%. Without access to CNN results, dermatologists achieved a mean diagnostic accuracy of 74.1%, which significantly improved to 86.4% when cooperating with the CNN. This highlights the potential benefits of using AI in healthcare, particularly in reducing the rate of misdiagnosis and unnecessary treatment.

Moreover, the study found that when given malignancy scores for suspicious lesions, dermatologists indicated significantly lower malignancy scores for benign nevi after receiving CNN results. The mean malignancy scores of melanomas were also higher when dermatologists cooperated with the CNN compared to when they were evaluated alone. These results suggest that with the assistance of CNN, dermatologists were better able to distinguish between benign nevi and malignant melanomas, which may lead to more timely and appropriate treatments.

Although the sensitivities of both the CNN and dermatologists alone were considerably lower, dermatologists with CNN support did not miss any melanomas in the prospective setting of the study. Excitingly, this study indicates that integrating CNN results into the diagnosis process can significantly improve dermatologists' diagnostic performance, highlighting the potential for AI to improve healthcare outcomes.

Ultimately, the study's findings demonstrate the potential revolutionary benefits of using CNN as a tool to aid dermatologists in diagnosing melanocytic lesions. By reducing unnecessary excisions and improving diagnostic accuracy, using CNN could potentially lead to better patient outcomes and reduce healthcare costs associated with unnecessary procedures.

Publication Date: May 23, 2023

Reference: Winkler JK, Blum A, Kommoss K, et al. Assessment of Diagnostic Performance of Dermatologists Cooperating with a Convolutional Neural Network in a Prospective Clinical Study: Human With Machine. JAMA Dermatol. Published online May 3, 2023. doi:10.1001/jamadermatol.2023.0905

Summary By: Parsa Abdi