The Role Of Artificial Intelligence In Dermatology: Enhancing The Accuracy Of Skin Cancer Diagnosis

Abstract

Skin cancer is still a major health issue and a correct and timely diagnosis plays a crucial role in the management of skin cancer. Artificial intelligence (AI) has become the focus as an important tool to improve the accuracy of diagnosis in dermatology. However, much more needs to be done for its real-world usability and the optimal outcome in various populations. This study involved a two-part design, a review of 10,000 dermoscopic images in the archive, and a study of 500 clinical cases. An AI diagnostic system was trained and validated on a diverse dataset with an emphasis on diagnostic sensitivity, specificity, and diagnostic concordance across demographic subgroups. The inclusion criteria included patients with clinically suspicious skin lesions and the exclusion criteria excluded cases with missing data or prior treatment. Descriptive statistics compared the AI and dermatologist diagnoses, and the results were compared across countries. The AI system had a sensitivity of 90.1% and a specificity of 87.6% which was comparable to dermatologist performance. The diagnostic accuracy was similar for all types of lesions but slightly lower for the dark skin tone and lesions in complex anatomic areas. Statistically relevant relationships (p < 0.05) were found between AI predictions and histopathological results. AI showed good results in the diagnosis of skin cancer, which may be useful as an auxiliary in dermatology. But, combating dataset biases and extending such research to more extensive audiences is crucial.