An in-silico lung phantom to assess the performance of pulmonary artery segmentation using angiogram

Pulmonary hypertension (PH) can lead to significant vascular remodeling, resulting in altered pulmonary blood flow. Estimating the patient-specific contributions of each remodeling event is necessary to optimize and individualize clinical intervention strategies. In-silico modeling has emerged as a powerful tool to simulate pulmonary hemodynamics, and one of the primary requirements for robust in-silico modeling is an accurate representation of the pulmonary vasculature structure. Computed tomography (CT) imaging can be used to segment and reconstruct the proximal vasculature. However, contrast-enhanced imaging, such as CT pulmonary angiography, is required to obtain a comprehensive and high-fidelity view of the pulmonary vasculature. The clinical use of CT pulmonary angiography is limited by the complications associated with the injection of contrast agents. Machine learning (ML) approaches have emerged to effectively segment and reconstruct the pulmonary vasculature without the need for contrast-enhanced imaging. We have developed a method to create in-silico pulmonary angiogram phantoms with varying simulated contrast levels. The results indicated that adding simulated contrast can allow for successful segmentation of the pulmonary vasculature. We expect this method to assist with developing and training ML-based segmentation frameworks and aid in their validation, thereby improving the capability to segment and reconstruct pulmonary vasculature without using contrast-enhanced imaging.Clinical relevanceThis study can aid in the generation of synthetic data sets for the training and validation of ML-based segmentation and reconstruction tools, reducing the need for contrast-enhanced imaging.