Artificial intelligence-driven body surface area (BSA) estimation using computed tomography: Comparative evaluation...

Area of Science:

• Medical Imaging and Radiology
• Biostatistics
• Clinical Pharmacology

Background:

• Total body surface area (BSA) is crucial for drug dosing and predicting metabolic mass, often preferred over body weight due to its reduced sensitivity to abnormal adipose tissue.
• Existing BSA estimation formulas primarily rely on 2D measurements and may use outdated statistical methods, potentially limiting their accuracy.
• The need for precise BSA estimation is critical in clinical medicine for safe and effective therapeutic interventions.

Purpose of the Study:

• To develop and validate novel formulas for estimating total body surface area (BSA) using 3D imaging data.
• To improve the accuracy of BSA estimation by incorporating patient demographics such as sex and race, alongside height and weight.
• To compare the performance of the newly developed BSA formulas against established methods.

Main Methods:

• Analysis of 3D whole-body positron emission tomography/computed tomography (PET/CT) scans from 698 patients to derive BSA measurements.
• Utilized TotalSegmentator software on CT components for accurate 3D BSA quantification.
• Employed multivariable ridge regression with 5-fold cross-validation to create new BSA estimation formulas, adjusting for sex, race, height, and weight.

Main Results:

• Established a strong positive correlation between BSA and both height (R=0.49) and weight (R=0.83).
• Developed an optimal formula for BSA estimation: [Formula].
• Demonstrated that the new formulas significantly outperformed all previously existing BSA formulas in accuracy, with lower mean absolute error and mean squared error.

Conclusions:

• The newly derived BSA estimation formulas offer superior accuracy compared to traditional methods.
• These advanced formulas provide a more reliable tool for clinical applications, particularly in drug dosing.
• The study highlights the importance of utilizing 3D imaging data and robust statistical techniques for precise physiological measurements.