Design and comparison of computationally efficient uniaxial stress–strain models of the lung parenchyma for real-time applications

• Published in: Computers in biology and medicine Vol. 180; p. 108928
• Main Authors: Courreges, Fabien, Melloni, Boris, Absi, Joseph
• Format: Journal Article
• Language: English
• Published: United States Elsevier Ltd 01.09.2024; Elsevier Limited
• Subjects: Atelectasis; Biomechanical computational model; Biomechanics; Computational efficiency; Computer applications; Error analysis; Floating point arithmetic; Localization; Lung cancer; Lung parenchyma; Lungs; Models evaluation; Parameter identification; Parenchyma; Radiation therapy; Real time; Real-time application; Robotic surgery; Strain; Synthetic data; Tumors; Biomechanical computational model; Models evaluation; Real-time application; Lung parenchyma
• DOI: 10.1016/j.compbiomed.2024.108928
• ISSN: 0010-4825; 1879-0534

Real-time clinical applications such as robotic lung surgery, tumor localization, atelectasis diagnosis, tumor motion prediction for radiation therapy of lung cancer, or surgery training are in need of biomechanical models of lungs, not necessarily highly accurate, but with good computational properties. These properties can include one or several of the following: low computation time, low memory resource requirement, a low number of parameters, and ease of parameter identification in real-time. Among the numerous existing models of lung parenchyma, some may be well suited for real-time applications; however, they should be extensively assessed against both accuracy and computational efficiency criteria to make an informed choice depending on the requirements of the application. After demonstrating how to derive a real-time compliant force-indentation model from a unixial stress–strain model with rational expression, the core purpose of this paper is to propose such an evaluation of selected models in fitting human lung parenchyma experimental and synthetic data of uniaxial tension. Furthermore, new uniaxial stress–strain models are developed based on an empirical observation of the volumetric behavior of the lungs along with an emphasis on computational performance. These new proposed models are competitive with existing one in terms of computational efficiency and compliance with experimental and synthetic data. One of them reduces the prediction error by 2 compared to other investigated models while maintaining an excellent adjusted coefficient of determination between 0.999 and 1 across various datasets. It exhibits excellent real-time capabilities with an explicit rational expression, only 3 parameters and linear numerator and denominator in the parameters. It is computed with only 20 floating point operations (flops) while another proposed model even requires as few as 2 flops. •Automation of lung tumor palpation needs real-time force-indentation models.•Evaluation method assesses models for accuracy and computational efficiency.•New stress-strain models based on empirical observation of lung volumetric behavior.•Proposed models’ competitiveness shown against existing ones and data fitting.•Proposed models excel in performance with low complexity.