Estimating hair density with XGBoost

Area of Science:

• Dermatology and Trichology
• Computational Biology
• Medical Imaging Analysis

Background:

• Manual hair density counting is labor-intensive and prone to errors.
• Existing automated methods using image processing and deep learning face challenges in robustness and applicability.
• Accurate hair density estimation is vital for diagnosing and monitoring hair loss conditions.

Purpose of the Study:

• To explore the efficacy of XGBoost for accurate and versatile hair density estimation.
• To develop an automated method that overcomes limitations of manual counting and existing automated techniques.
• To improve the objectivity and efficiency of clinical hair analysis.

Main Methods:

• Utilized 895 scalp images for feature extraction.
• Developed and trained an XGBoost model on 745 images.
• Evaluated model performance on 150 test images, assessing accuracy, error rate, and scatter plot.

Main Results:

• The XGBoost model achieved 89.5% accuracy on the training set and 95.3% accuracy on the test set.
• Outperformed previous methods, including those by Kim et al. (52.4%), Urban et al. (79.6%), and Sacha et al. (88.2%) on the test set.
• Demonstrated high accuracy in estimating hair density from scalp images.

Conclusions:

• XGBoost algorithm is effective for automated hair density estimation with 95.3% test set accuracy.
• The method, focusing on scalp coverage and erosion features, streamlines clinical hair analysis.
• This approach offers improved objectivity and efficiency in dermatological and trichological assessments.