Comparison of physician and computer-aided image analysis in dermoscopy of melanocytic lesions

ABSTRACT COMPARİSON OF PHYSİCİAN AND COMPUTER-AİDED IMAGE ANALYSİS İN DERMOSCOPY OF MELANOCYTİC LESİONS Objective: Our study aimed to compare the dermoscopic evaluation by physicians with the diagnostic analysis of a digital dermoscopy device (Moleanalyzer-3TM) in patients with melanocytic lesions who had undergone histopathological examination and digital dermoscopy recording within a 10-year period. We sought to determine whether we should incorporate device analysis more frequently in our clinical practices by comparing their concordance with histopathology, which is considered the gold standard for diagnosis. Materials and Methods: The study included 216 lesions from patients diagnosed with junctional nevus, compound nevus, intradermal nevus, dysplastic nevus, lentigo, lentigo maligna, and melanoma following histopathological examination at Aydın Adnan Menderes University Medical Faculty Hospital Pathology Department between January 1, 2014, and October 31, 2023. These patients were examined at the Dermatology Clinic, and their lesions were recorded with a digital dermoscope prior to histopathological examination. The dermoscopic images were evaluated using the ABCD score and revised pattern analysis methods by a dermatology resident blinded to the histopathology results. Subsequently, these images were analyzed and scored by the Moleanalyzer-3TM image analysis application of the Fotofinder® device. Results: When the lesions were classified as benign or melanoma based on pattern analysis, the sensitivity, specificity, and diagnostic accuracy for melanoma diagnosis were calculated as 97.3%, 92.7%, and 93.5%, respectively. Good concordance was found between pattern analysis and histopathology for melanoma diagnosis (κ=0.798, p<0.001). When the ABCD score threshold was set at 5.45 for melanoma diagnosis, resulting in a binary classification of benign and melanoma; the sensitivity, specificity, and diagnostic accuracy for melanoma diagnosis were calculated as 71.4%, 93.7%, and 82.9%, respectively. Moderate concordance was found between the ABCD score and histopathology for melanoma diagnosis (κ=0.590, p<0.001). When the threshold ABCD score was set at 4.75, the sensitivity, specificity, and diagnostic accuracy were calculated as 81.0%, 77.7%, and 78.1%, respectively. Low concordance was found between the ABCD score and histopathology for melanoma diagnosis at this threshold score (κ=0.338, p<0.001). To evaluate the performance of the Moleanalyzer test in classifying lesions as benign or malignant, ROC analysis was performed, yielding an AUC value of 0.847, indicating good diagnostic accuracy for melanoma diagnosis. According to the ROC curve, the score with the highest combined sensitivity and specificity was found to be 0.65. For a threshold score of >0.65, the sensitivity, specificity, and diagnostic accuracy were calculated as 83.8%, 77.3%, and 78.4%, respectively. The Moleanalyzer showed moderate concordance with histopathology at this threshold score (κ=0.448, p<0.001). Conclusion: The Moleanalyzer did not demonstrate superiority over the ABCD score or pattern analysis. Therefore, as dermatologists, we can rely on our knowledge and evaluate lesions using the ABCD scoring or pattern analysis, except in cases where we are uncertain and seek a second opinion. Furthermore, the pattern analysis method was found to be the most successful method. Thus, we recommend the primary use of pattern analysis in the evaluation of melanocytic lesions. We believe that the Moleanalyzer could be beneficial for melanoma screening in primary healthcare facilities where physicians lack dermoscopy experience. Key Words: Computer-Aided Image Analysis, Dermoscopy, Melanoma, Nevus.