Automated Bayesian Quantification of Pulmonary Perfusion Alterations in CTEPH via CTPA

Overview

This study developed an automated Bayesian method to quantify the extent and spatial distribution of pulmonary perfusion changes on conventional CT pulmonary angiography (CTPA) in patients with chronic thromboembolic pulmonary hypertension (CTEPH). The approach uses Gaussian mixture modeling of lung attenuation histograms to classify lung tissue into hyperemic, normal, and oligemic compartments, enabling objective assessment of perfusion heterogeneity and centralization.

Background

CTEPH is characterized by persistent obstruction or stenosis of pulmonary arteries due to organized thrombi and microvasculopathy, leading to pulmonary hypertension and right heart failure if untreated. CTPA is a key imaging modality for evaluating pulmonary artery structure and perfusion changes, traditionally assessed subjectively by experts. Mosaic perfusion patterns on CTPA, indicating small vessel disease and poor prognosis, have lacked automated quantification methods. This study addresses this gap by developing a fully automated, reproducible technique to assess perfusion alterations in CTEPH.

Data Highlights

The automated method segments lungs and pulmonary veins using a deep learning network and applies a variational Bayesian Gaussian mixture model to voxel attenuation histograms, classifying lung tissue into three compartments: high attenuation (hyperemic), medium attenuation (normal), and low attenuation (oligemic). Spatial heterogeneity is quantified by entropy of 3D voxel coordinate histograms for high- and medium-attenuation compartments. Centralization of perfusion is assessed by the slope of mean attenuation versus distance from hilum to pleura. The method was applied to CTPA scans of patients diagnosed with CTEPH, with clinical and imaging parameters collected for correlation.

Key Findings

• The automated Bayesian approach successfully decomposed lung attenuation histograms into three compartments representing hyperemic, normal, and oligemic lung tissue.
• Spatial heterogeneity of perfusion compartments was quantified using entropy measures, reflecting dispersion of perfusion abnormalities.
• Centralization analysis quantified the distribution of perfusion changes from central to peripheral lung regions, capturing mosaic perfusion patterns.
• The method provided objective, reproducible quantification of perfusion alterations correlating with expert visual assessments and clinical parameters.
• The approach minimized bias from contrast enhancement variability by fitting patient-specific Gaussian mixture models.

Clinical Implications

This automated method offers clinicians an objective tool to quantify and spatially characterize pulmonary perfusion abnormalities in CTEPH using routine CTPA scans. It may improve diagnostic accuracy, enable standardized assessment of disease severity, and facilitate monitoring of treatment response. The technique reduces reliance on subjective interpretation and interobserver variability in evaluating mosaic perfusion patterns.

Conclusion

The study presents a novel, fully automated Bayesian framework for quantifying the degree and distribution of pulmonary perfusion alterations in CTEPH on CTPA. This method enhances the objectivity and reproducibility of perfusion assessment, with potential to improve clinical evaluation and management of CTEPH patients.

References

1. European Society of Cardiology Guidelines 2022 -- Diagnosis and treatment of pulmonary hypertension
2. Study on CTPA and V/Q scan diagnostic accuracy -- Comparative imaging in CTEPH
3. Mosaic perfusion and prognosis in CTEPH -- Imaging biomarkers of small vessel disease
4. Limitations of subjective CTPA analysis -- Variability in mosaic perfusion interpretation
5. Total Segmentator network for lung segmentation -- Automated lung and vein segmentation
6. Bayesian Gaussian mixture modeling in imaging -- Variational Bayesian estimation for histogram decomposition