Exploring artificial intelligence for differentiating early syphilis from other skin lesions: a pilot study

Area of Science:

• Dermatology
• Medical Imaging
• Artificial Intelligence

Background:

• Early diagnosis of syphilis is crucial for effective disease control.
• Distinguishing syphilis from other skin lesions can be challenging.
• Radiomics offers a novel approach for analyzing medical images.

Purpose of the Study:

• To develop an Artificial Intelligence (AI) diagnostic model for early syphilis detection.
• To utilize radiomics features for differentiating syphilis from other skin lesions.
• To evaluate the performance of machine learning classifiers in syphilis diagnosis.

Main Methods:

• Collected 260 skin lesion images (115 syphilis, 145 other infections).
• Extracted 102 radiomics features from manually segmented Regions of Interest (ROIs).
• Trained and tested 11 classifiers, including Gradient Boosted Decision Trees (GBDT), with cross-validation and a hold-out set. Investigated Wavelet filters and used SHAP for interpretation.

Main Results:

• The GBDT model with Wavelet filter achieved a cross-validation AUC of 0.832 and accuracy of 0.735.
• The model demonstrated 0.792 AUC and 0.750 accuracy on the hold-out test set.
• SHAP analysis identified 2D sphericity as the most predictive feature for distinguishing syphilis.

Conclusions:

• The AI diagnostic model based on radiomics features shows potential for distinguishing early syphilis.
• The model achieved a diagnostic accuracy of 75.0% in the test set.
• Radiomics and AI offer a promising tool for improving early syphilis diagnosis.