Portable dual-modal infrared-RGB imaging for automated classification of skin lesions

Skin cancer is one of the most prevalent malignancies worldwide, making early diagnosis crucial for optimal treatment outcomes. While traditional dermoscopy remains valuable diagnostic tools, they may present challenges in standardization and objective assessment, particularly in primary care settings. This study presents an intelligent diagnostic system combining portable dual-modal infrared-RGB imaging for automated classification of basal cell carcinoma (BCC), seborrheic keratosis (SK), and intradermal nevus. Fifty-nine subjects were recruited, comprising 16 BCC patients, 10 SK patients, and 33 intradermal nevus patients, with lesion areas simultaneously captured using a portable infrared-RGB device for RGB and thermal images. Subsequently, feature extraction included texture, morphology, and color characteristics from RGB images, alongside temperature distribution patterns and local gradient changes from thermal image analysis. These features were then processed through principal component analysis for dimensionality reduction and support vector machine classification for automated diagnosis. To validate system performance, three dermatologists with varying experience levels independently diagnosed identical images as controls. Results demonstrated that for three-category classification, the intelligent system achieved 66.15% accuracy, comparable to physicians' average performance of 62.64% accuracy. More notably, in the clinically challenging BCC-SK diagnosis, the system significantly outperformed clinical physicians with 76.44% versus 54.67% accuracy. These findings indicate that portable infrared-RGB imaging demonstrates potential for skin lesion diagnosis, particularly in challenging clinical cases, thereby offering promising screening capabilities for primary healthcare settings and potentially improving diagnostic standardization in dermatological practice.