Impact of Photon-counting Detector Computed Tomography on a Quantitative Interstitial Lung Disease Machine Learning...

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Interstitial lung disease (ILD) diagnosis relies on imaging and pulmonary function tests (PFTs).
• Quantitative machine learning (QML) models analyze CT scans for ILD assessment.
• Photon-counting detector computed tomography (PCD-CT) offers potential improvements over conventional CT.

Purpose of the Study:

• To compare the performance of PCD-CT versus conventional CT within an ILD QML model.
• To evaluate the impact of PCD-CT on the correlation between QML-derived ILD features and PFT results.

Main Methods:

• A QML model was applied to 52 same-day conventional and PCD-CT scans from patients with suspected ILD.
• Lin's concordance correlation coefficient (CCC) assessed agreement between QML results from both CT types.
• Spearman rank correlation analyzed associations between QML features (reticulation, honeycombing, ground glass, vessel-related structures) and PFTs.

Main Results:

• Good to excellent concordance (CCC ≥0.8) was observed between PCD-CT and conventional CT QML results, except for honeycombing (HC).
• PCD-CT demonstrated significantly stronger correlation with PFTs for HC compared to conventional CT.
• PCD-CT showed similar correlation for reticulation and vessel-related structures but reduced correlation for ground glass extent.

Conclusions:

• PCD-CT and conventional CT QML models for ILD show strong agreement, with notable improvement in HC assessment by PCD-CT.
• PCD-CT enhances HC correlation with PFTs but diminishes GG extent correlation.
• While most quantitative ILD features are largely unaffected, model adjustments are necessary for PCD-CT implementation.