Combining deep learning and X-ray imaging for enhanced quantification of lung fibrosis progression and therapeutic response in small animals

Computed tomography (CT) represents a cornerstone of modern medical imaging, providing detailed anatomical information through X-ray–based tomographic reconstruction. Among its modalities, high-resolution computed tomography (HRCT) has become indispensable for assessing pulmonary structure and pathology, enabling precise visualization of interstitial and alveolar abnormalities without contrast enhancement. Quantitative and texture-based HRCT analyses have recently advanced from qualitative interpretation toward reproducible, data-driven biomarkers capable of tracking disease progression and therapeutic response in chronic lung disorders such as idiopathic pulmonary fibrosis (IPF), chronic obstructive pulmonary disease (COPD), and connective tissue disease–associated interstitial lung disease (ILD). Despite these advances, the clinical implementation of quantitative imaging remains constrained by technical and standardization challenges. In preclinical research, micro-computed tomography (µCT) has emerged as the analogue of HRCT, offering high-resolution, non-invasive, longitudinal imaging in small-animal models of pulmonary disease. µCT allows quantitative evaluation of lung aeration, volume, and tissue density, and when coupled with respiratory-gated functional imaging, it can approximate spirometric parameters in vivo. However, traditional image-processing workflows relying on manual or semi-automatic segmentation remain labor-intensive and prone to variability, particularly in diseased lungs with reduced air content. To address these limitations, this work integrates classical algorithms and artificial intelligence (AI)-based methods into the µCT analysis pipeline at Chiesi Farmaceutici. The approach aims to enhance longitudinal and regional assessments of pulmonary disease progression and therapeutic response, which may contribute to a reduction of the translational gap between preclinical and clinical research. Within this framework, several automated pipelines have been developed, including a YOLO-based respiratory gating algorithm for 4D µCT reconstruction, a tool for fully-automated lobe-level lung segmentation and analysis, a new strategy for regional investigation, and a semi-automatic tool for linking in vivo µCT data with ex vivo histological findings.