Automated psoriasis lesion segmentation from unconstrained environment using residual U-Net with transfer learning.

Area of Science:

• Medical image analysis
• Computational dermatology
• Artificial intelligence in healthcare

Background:

• Automatic segmentation of psoriasis lesions is challenging due to imaging variability.
• Existing methods have limitations including manual feature dependence and preprocessing needs.
• Deep learning offers a promising avenue for automated lesion segmentation.

Purpose of the Study:

• To develop a fully automatic deep learning model for psoriasis lesion segmentation.
• To leverage transfer learning for improved segmentation performance.
• To enable objective area assessment of psoriasis lesions.

Main Methods:

• A U-Net based deep learning architecture was employed.
• Transfer learning utilized a pre-trained residual network as the encoder backbone.
• The model was retrained on a custom psoriasis dataset with lesion annotations.

Main Results:

• The proposed model achieved an average Dice Similarity Index of 0.948 and Jaccard Index of 0.901.
• Transfer learning enhanced performance by 4.4% (Dice) and 7.6% (Jaccard) over training from scratch.
• Five-fold cross-validation confirmed robust performance.

Conclusions:

• The developed deep learning framework demonstrates promising performance for psoriasis lesion segmentation.
• Comparative analysis validates its effectiveness against state-of-the-art methods.
• This approach provides a foundation for objective psoriasis lesion area measurement.