Artificial intelligence in computed tomography for quantifying lung changes in the era of CFTR modulators

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Pulmonology

Background:

• Chest computed tomography (CT) is the standard for assessing cystic fibrosis (CF) airway disease.
• Current visual scoring systems are time-consuming, require specialized training, and have limited reproducibility.
• There is a need for automated, objective methods to quantify CF lung disease severity.

Purpose of the Study:

• To validate a fully automated artificial intelligence (AI)-driven scoring system for CF lung disease severity.
• To compare AI-based quantification with traditional visual scoring and pulmonary function tests.
• To assess the AI system's reproducibility and clinical validity in patients undergoing CFTR modulator therapy.

Main Methods:

• A 2D convolutional neural network algorithm was trained on CT scans from 78 CF patients for semantic labeling of airway abnormalities.
• The AI system was tested against ground-truth labels using CT scans from 36 patients.
• Clinical validity was assessed in an independent cohort of 70 patients, including those treated with lumacaftor/ivacaftor.

Main Results:

• The AI system achieved good overall pixelwise similarity (Dice 0.71) with ground-truth labels.
• AI-driven volumetric quantifications showed moderate to very good correlations with visual scoring and pulmonary function tests.
• Significant changes in airway disease markers were detected in patients with and without lumacaftor/ivacaftor treatment, with near-perfect reproducibility (Dice >0.99).

Conclusions:

• Fully automated volumetric quantification of CF-related lung modifications is feasible using AI.
• This novel AI scoring system offers a robust outcome measure for CF lung disease.
• The AI system is well-suited for monitoring disease progression and treatment response in the era of CFTR modulator therapies.