Dermoscopic Image Classification of Pigmented Nevus under Deep Learning and the Correlation with Pathological...

Area of Science:

• Dermatology
• Artificial Intelligence
• Medical Imaging

Background:

• Pigmented nevi (PN) diagnosis relies on dermoscopy.
• Deep learning offers potential for automated image analysis.

Purpose of the Study:

• To evaluate deep learning for pigmented nevus classification using dermoscopy.
• To compare diagnostic accuracy between deep learning and traditional methods.

Main Methods:

• Utilized a deep learning image recognition algorithm for feature extraction, segmentation, and classification of dermoscopic images.
• Compared performance and accuracy of a fusion classifier.
• Randomly assigned 268 patients into observation (n=134) and control (n=134) groups.

Main Results:

• Optimized classifier achieved 85.82% accuracy with a linear kernel.
• Collaborative training showed significantly higher sensitivity (P < 0.05) than feature or fusion feature training.
• Observation group sensitivity (88.65%) and specificity (90.26%) exceeded the control group (85.65% sensitivity, 84.03% specificity) (P < 0.05).

Conclusions:

• Deep learning with dermoscopy is a viable method to enhance pigmented nevus diagnostic accuracy.
• Dermoscopic features correlate with PN types, supporting auxiliary clinical diagnosis.
• This AI-driven approach shows promise for clinical application.