A burn depth detection system based on near infrared spectroscopy and ensemble learning

Area of Science:

• Biomedical Optics
• Medical Spectroscopy
• Machine Learning in Medicine

Background:

• Near-infrared (NIR) spectroscopy is a non-invasive technique for assessing tissue properties.
• Determining burn depth accurately is crucial for effective clinical treatment.
• Traditional analysis of NIR spectral data for burn depth lacks precision due to complex signal-tissue relationships.

Purpose of the Study:

• To develop and validate a machine learning model for accurate burn depth assessment using NIR spectroscopy.
• To establish a robust relationship between NIR spectral signals and burn depth.
• To provide a reliable tool for clinicians in burn management.

Main Methods:

• Extraction of optical properties from NIR spectral signals based on diffuse reflection theory.
• Application of a chained-agent genetic algorithm (CAGA) to optimize support vector regression (SVR) for modeling.
• Development of a CAGA-SVR integrated inversion model correlating optical properties with burn depth.
• Validation of the model using a porcine burn model.

Main Results:

• The proposed CAGA-SVR model effectively established a regression relationship between optical properties and burn depth.
• The integrated inversion model demonstrated accurate prediction of burn depth in the porcine model.
• The method provides a significant improvement over simpler data analysis techniques for NIR spectral data.

Conclusions:

• The CAGA-SVR integrated inversion model offers a precise and reliable method for assessing burn depth using NIR spectroscopy.
• This approach provides valuable quantitative data for clinical decision-making in burn treatment.
• The study highlights the potential of advanced machine learning techniques in medical diagnostics.