Deep learning-driven pulmonary arteries and veins segmentation reveals demography-associated pulmonary vasculature anatomy

Overview

• Pulmonary artery-vein segmentation is crucial for diagnosing pulmonary diseases and surgical planning.
• Traditionally, this is done using Computed Tomography Pulmonary Angiography (CTPA), which uses contrast agents that can have adverse health effects.
• The paper proposes a new framework called High-abundant Pulmonary Artery-vein Segmentation (HiPaS) that can accurately segment arteries and veins from both non-contrast CT and CTPA scans.
• HiPaS achieves this by first normalizing the CT scans using a super-resolution module, then iteratively segmenting vessels at different branch levels using the low-level segmentation as a prior.
• The framework was trained and validated on a large dataset of 1,073 annotated CT volumes, and demonstrated superior performance compared to existing methods.
• Further experiments showed that HiPaS segmentation on non-contrast CT was non-inferior to segmentation on CTPA.
• The paper also presents an anatomical study of pulmonary vasculature on over 10,000 participants, revealing new associations between vessel abundance, sex, and age.

Plain English Explanation

Pulmonary artery and vein segmentation is crucial for doctors to diagnose lung diseases and plan surgeries. Traditionally, this is done using a special type of CT scan called Computed Tomography Pulmonary Angiography (CTPA), which requires injecting a contrast agent into the patient's body. However, these contrast agents can have negative health effects.

The researchers in this paper developed a new method called HiPaS that can accurately identify arteries and veins in the lungs using regular CT scans, without the need for contrast agents. HiPaS first improves the quality of the CT scans using a super-resolution module, then segments the vessels in a stepwise fashion, using the results from earlier steps to guide the later ones.

The researchers trained and tested HiPaS on a large dataset of over 1,000 annotated CT scans, and found that it outperformed existing methods. Importantly, they also showed that HiPaS produces similar results when used with regular CT scans compared to the CTPA scans.

In addition, the researchers used HiPaS to study the anatomy of the pulmonary vasculature in over 10,000 people. They discovered that women tend to have more blood vessels in their lungs than men, and that the number of vessels decreases slightly with age, even after accounting for differences in lung size.

Overall, this new HiPaS framework provides a promising non-invasive way for doctors to diagnose lung conditions and understand pulmonary physiology, without exposing patients to the risks of contrast agents.

Technical Explanation

The proposed High-abundant Pulmonary Artery-vein Segmentation (HiPaS) framework aims to achieve accurate artery-vein segmentation on both non-contrast CT and Computed Tomography Pulmonary Angiography (CTPA) scans across various spatial resolutions.

HiPaS first performs spatial normalization on the raw CT scans using a super-resolution module, which improves the quality of the input data. It then iteratively segments vessels at different branch levels, utilizing the low-level vessel segmentation as a prior for high-level vessel segmentation. This stepwise approach allows HiPaS to accurately delineate the pulmonary vasculature.

The researchers trained and validated HiPaS on a large multi-centric dataset of 1,073 CT volumes with manual annotations. Both quantitative experiments and clinical evaluation demonstrated the superior performance of HiPaS, achieving a dice score of 91.8% and a sensitivity of 98.0%.

Further experiments showed that the segmentation results from HiPaS on non-contrast CT were non-inferior to those from CTPA, indicating that HiPaS can provide accurate artery-vein segmentation without the need for contrast agents.

The researchers also leveraged HiPaS to conduct an anatomical study of pulmonary vasculature on a large cohort of 10,613 participants across five sites in China. This study revealed a new association between pulmonary vessel abundance, sex, and age: vessel abundance is significantly higher in females than in males, and slightly decreases with age, after controlling for lung volumes.

Critical Analysis

The HiPaS framework presents a promising solution for accurate pulmonary artery-vein segmentation without the use of contrast agents, which can have adverse health effects. The iterative segmentation approach, combined with the super-resolution module, appears to be an effective way to leverage the information available in non-contrast CT scans to achieve results comparable to CTPA.

However, the paper does not provide much detail on the specific architecture or training procedure of the HiPaS model, making it difficult to fully assess the technical implementation. Additionally, the dataset used for training and validation, while large, was limited to a specific Chinese population, so further testing on more diverse cohorts would be needed to establish the broader applicability of the method.

The anatomical study conducted using HiPaS is an interesting contribution, but the underlying reasons for the observed associations between vessel abundance, sex, and age are not explored in depth. Further research would be needed to understand the physiological mechanisms driving these differences and their potential implications for clinical practice.

Overall, the HiPaS framework represents an important step forward in enabling non-invasive pulmonary artery-vein segmentation, with the potential to improve diagnostic capabilities and reduce patient risk. However, additional technical details and validation across diverse populations would strengthen the confidence in the method's robustness and generalizability.

Conclusion

The HiPaS framework proposed in this paper provides a promising solution for accurate pulmonary artery-vein segmentation using non-contrast CT scans, overcoming the limitations of traditional CTPA methods that rely on contrast agents. The iterative segmentation approach, combined with spatial normalization, allows HiPaS to achieve results comparable to CTPA without exposing patients to the potential health risks of contrast agents.

The large-scale anatomical study conducted using HiPaS also yielded interesting new insights into the relationships between pulmonary vessel abundance, sex, and age, opening up avenues for further research into the underlying physiological mechanisms. These findings could have important implications for our understanding of pulmonary physiology and the development of personalized diagnostic and treatment approaches.

While the technical details of the HiPaS framework could be further elaborated, and its performance validated across more diverse populations, this research represents a significant step forward in enabling non-invasive, contrast-free assessment of the pulmonary vasculature. Continued development and refinement of this approach could lead to improvements in the diagnosis and management of a wide range of pulmonary diseases, ultimately benefiting patients and clinicians alike.