Recognition of Cutaneous Melanoma on Digitized Histopathological Slides via Artificial Intelligence Algorithm

Area of Science:

• Oncology
• Pathology
• Computer Science

Background:

• Rising skin cancer incidence and dermatopathologist shortages strain pathology departments.
• High variability in diagnosing melanocytic skin lesions impacts melanoma assessment accuracy.
• Standardized, accurate diagnostic tools are crucial for effective melanoma treatment planning.

Purpose of the Study:

• To develop an artificial intelligence (AI) system for recognizing cutaneous melanoma from histopathological slides.
• To assess the accuracy and efficiency of a deep learning model in differentiating melanoma from healthy tissue.
• To provide a tool that aids pathologists in standardizing diagnoses and improving patient care.

Main Methods:

• A convolutional neural network (CNN) was trained using digital slides of primary cutaneous melanoma.
• The CNN, based on Inception-ResNet-v2, differentiated tumoral from healthy tissue.
• Whole-slide images from 100 melanoma cases were used for training and testing the AI model.

Main Results:

• The AI system achieved high diagnostic accuracy (96.5%), sensitivity (95.7%), and specificity (97.7%).
• The model demonstrated strong performance with a F1 score of 96.5% and Cohen's kappa of 0.929.
• The deep learning system's accuracy was comparable to that of experienced dermatopathologists.

Conclusions:

• A deep learning system can effectively recognize melanoma from histopathological slides.
• This AI approach can enhance diagnostic efficiency and reduce interobserver variability in pathology.
• Further research with larger datasets is needed to explore subclassification of melanoma subtypes using AI.