Clinical validation of perfusion imaging with pulmonary function test data using Voronoi-based discretization.

Area of Science:

• Medical Imaging
• Pulmonary Medicine
• Computational Biology

Background:

• Accurate lung function assessment is crucial for diagnosing and managing respiratory diseases such as COPD, pulmonary emboli, and lung cancer.
• Single-photon emission computed tomography (SPECT) and CT-based perfusion (CT-P) offer 3D functional imaging but face limitations like low resolution, radiation, and cost.
• Pulmonary function tests (PFTs) provide rapid, noninvasive global lung function measures, but correlating them with perfusion imaging is challenging due to complex blood flow and data summarization issues.

Purpose of the Study:

• To develop a framework that integrates regional perfusion imaging with global pulmonary function assessment.
• To improve the quantitative comparison between SPECT and CT-based perfusion maps by capturing local spatial information.
• To establish a noninvasive tool for combining imaging and physiological data for lung function evaluation.

Main Methods:

• Proposed a framework using lung discretizations based on Voronoi diagrams to analyze raw-valued and percentile-scaled perfusion maps (SPECT and CT-P).
• Computed hierarchical descriptive statistics at intra-subvolume, inter-subvolume, and left-right lung levels to derive a single global value per patient.
• Validated the approach using Spearman correlations with PFT measures (DLCO, FEV1, FEV1/FVC) and logistic regression for distinguishing normal from abnormal lung function.

Main Results:

• Discretizing perfusion maps using Voronoi subvolumes consistently yielded stronger Spearman correlations with PFTs compared to non-discretized methods.
• Achieved strong correlations (0.636 ≤ ρ ≤ 0.843 for raw-valued maps, 0.590 ≤ ρ ≤ 0.789 for percentile-scaled maps) with PFT measures.
• Demonstrated reliable distinction between normal and abnormal lung function with high Area Under the Curve (AUC) values (0.865–0.937 for raw-valued maps, 0.877–0.933 for percentile-scaled maps).

Conclusions:

• The proposed Voronoi diagram-based framework effectively bridges regional perfusion imaging and global pulmonary function assessment.
• The method enables meaningful quantitative comparisons between different perfusion imaging modalities (SPECT and CT-P).
• This approach offers a noninvasive tool for integrating imaging and physiological data, supporting broader clinical and AI-driven applications in lung function evaluation.