Intelligent Pathological Diagnosis of Melanocytic Lesions Using Deep Learning

Melanocytic lesions occur on the surface of the skin, and melanoma is a malignant type of melanocytic lesion associated with a high mortality rate, posing a serious threat to human health.Histopathological analysis remains the gold standard for diagnosing melanocytic lesions.In this study, a fully automated intelligent diagnosis method based on deep learning is proposed to classify pathological whole slide images (WSIs) of melanocytic lesions.First, color normalization using a CycleGAN neural network was performed on multi-center pathological WSIs.Second, a ResNet-152 neural network-based deep convolutional network prediction model was built using 745 WSIs.Third, a decision fusion model was cascaded to calculate the average prediction probability for each WSI.Finally, the diagnostic performance of the proposed method was verified using internal and external test sets containing 182 and 54 WSIs, respectively.Experimental results showed that the overall diagnostic accuracy of the proposed method reached 94.12% on the internal test set and exceeded 90% on the external test set.Furthermore, the adopted color normalization method was superior to the traditional color statistics-based and staining separation-based methods in terms of structure preservation and artifact suppression.These results demonstrate that the proposed method can achieve high precision and strong robustness in classifying pathological WSIs of melanocytic lesions, highlighting its potential in promoting the clinical application of computer-aided pathological diagnosis.