Deep Learning-based Outcome Prediction in Progressive Fibrotic Lung Disease Using High-Resolution Computed Tomography

Area of Science:

• Radiology
• Artificial Intelligence
• Pulmonary Medicine

Background:

• Predicting outcomes in fibrotic lung disease using high-resolution computed tomography (HRCT) is challenging.
• Current methods lack consistent prognostic accuracy.

Purpose of the Study:

• To assess the prognostic accuracy of the deep learning algorithm SOFIA (Systematic Objective Fibrotic Imaging Analysis Algorithm).
• To compare SOFIA's performance against radiologist assessments in predicting outcomes for patients with progressive fibrotic lung disease.

Main Methods:

• SOFIA identified usual interstitial pneumonia (UIP)-like features on HRCT, generating UIP probability scores.
• Scores were converted to Prospective Investigation of Pulmonary Embolism Diagnosis (PIOPED)-based categories.
• Cox proportional hazards modeling assessed prognostic utility, adjusting for clinical and imaging variables.

Main Results:

• SOFIA's UIP probability categories independently predicted survival in multivariable analysis.
• SOFIA remained significant for indeterminate cases and predicted mortality after adjusting for histology in biopsy-proven cases.
• Deep learning-based prediction outperformed radiologist evaluation and histologic patterns.

Conclusions:

• Deep learning analysis of HRCT offers superior outcome prediction in fibrotic lung disease compared to traditional methods.
• SOFIA shows potential as a decision support tool for multidisciplinary fibrotic lung disease characterization.
• Further investigation is needed for its use in settings with limited ILD expertise.