Evaluation of an Artificial Intelligence-Assisted Videodermoscopy System for Cutaneous Lesion Classification: A Retrospective Study
Abstract
Introduction: Skin cancers represent some of the most common cancers worldwide. [1] Its incidence is constantly rising. Early detection is essential to reduce morbidity and improve survival. [11] In recent years, artificial intelligence (AI)-based software has increasingly been integrated into 2D/3D total-body-photography (TBP) systems, to support clinicians in diagnosing cutaneous lesions.[18] Objectives: To evaluate VIDIX 4.0 diagnostic performance using histopathology as the reference standard and assess the ability of the DEXI score to discriminate between malignant and benign lesions. Methods: This monocenter retrospective study included 204 lesions from 165 patients who underwent VIDIX 4.0 videodermoscopy, followed by excision and histopathological examination. Diagnostic performance was evaluated in terms of sensitivity, specificity, predictive values, and accuracy, with 95% confidence intervals calculated using the Wilson method. Malignant and benign lesions were compared after excluding actinic keratoses. DEXI scores were analyzed using Kruskal–Wallis and Mann–Whitney U tests with Holm adjustment. Receiver operating characteristic (ROC) analysis assessed DEXI discrimination using the area under the curve (AUC) with 95% confidence intervals. Results: VIDIX sensitivity was 93.3% for nevi, 81.8% for basal cell carcinoma (BCC), 63.2% for melanoma, 44.4% for squamous cell carcinoma (SCC), 33.3% for actinic keratosis (AK), and 61.1% for seborrheic keratosis (SK). For malignant versus benign lesions, sensitivity, specificity, and accuracy were 80.3%, 92.7%, and 88.9%, respectively. DEXI scores differed significantly among histopathological categories (P<0.0001), showing good discrimination between malignant and benign lesions (AUC, 0.939; 95% CI, 0.896–0.974). Conclusion: VIDIX, demonstrating good diagnostic performance, particularly for benign melanocytic lesions and BCC, and moderate results for melanoma, may represent a useful adjunctive tool for cutaneous lesion assessment.