Measurement of Body Surface Area for Psoriasis Using U-net Models

Area of Science:

• Dermatology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate psoriasis body surface area (BSA) measurement is vital for assessing disease severity and guiding treatment.
• Subjective visual evaluation by physicians lacks reliability for precise BSA assessment.
• Objective and automated methods are needed for consistent psoriasis severity evaluation.

Purpose of the Study:

• To develop and evaluate a machine learning model for automated psoriasis lesion segmentation and BSA measurement.
• To assess the performance of a U-net based artificial neural network for this task.
• To provide an objective tool for evaluating psoriasis severity.

Main Methods:

• A U-net convolutional neural network architecture was employed for psoriasis lesion segmentation.
• The model was trained on 255 high-resolution images of psoriasis lesions.
• Performance was evaluated using metrics like average residual and interclass correlation coefficient against dermatologist assessments.

Main Results:

• The U-net model demonstrated high accuracy in segmenting psoriasis lesions and estimating BSA.
• The average residual between predicted and ground truth BSA was approximately 0.033.
• An interclass correlation coefficient of 0.966 indicated strong agreement with dermatologist segmentations.

Conclusions:

• The proposed U-net model achieves dermatologist-level performance in estimating psoriasis-involved BSA.
• Automated segmentation and BSA measurement using deep learning offer a reliable alternative to subjective visual assessment.
• This AI-driven approach can enhance objective evaluation of psoriasis severity.