Predicting the severity of postoperative scars using artificial intelligence based on images and clinical data

Area of Science:

• Medical image analysis
• Artificial intelligence in dermatology
• Postoperative scar assessment

Background:

• Objective scar evaluation is essential for effective treatment planning.
• Current methods lack a definitive gold standard for scar severity assessment.
• Postoperative scars, particularly after thyroidectomy, require reliable assessment tools.

Purpose of the Study:

• To develop and validate an artificial intelligence (AI) model for predicting postoperative scar severity.
• To utilize deep neural networks with patient imaging and clinical data.
• To compare the AI model's performance against dermatologists' assessments.

Main Methods:

• A deep neural network was trained on 1043 post-thyroidectomy scar images and clinical data.
• The model was validated on an independent external dataset of 240 patients.
• Performance was quantified using the area under the receiver operating characteristic curve (ROC-AUC) and compared to 16 dermatologists.

Main Results:

• The image-based AI model achieved an ROC-AUC of 0.931 internally and 0.896 externally.
• Combining clinical data with images improved the ROC-AUC to 0.938 internally and 0.912 externally.
• The AI model's performance was comparable to that of experienced dermatologists on the internal test set.

Conclusions:

• A deep neural network model effectively predicts postoperative scar severity using imaging and clinical data.
• The AI model demonstrates high accuracy and reliability in scar assessment.
• This AI tool has the potential to aid clinicians in scar management and timely treatment initiation.