Quantitative Analysis of Radiation-Associated Parenchymal Lung Change

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Radiation-induced lung damage (RILD) is a common side effect of thoracic radiotherapy.
• Accurate assessment of RILD evolution and its correlation with clinical outcomes is crucial for patient management.
• Current methods for evaluating RILD are often subjective and lack quantitative detail.

Purpose of the Study:

• To develop and validate a novel deep learning system for classifying parenchymal features of RILD.
• To quantitatively describe the temporal evolution of RILD up to 24 months post-radiotherapy.
• To correlate RILD changes with radiotherapy dose and respiratory outcomes.

Main Methods:

• A deep learning network was developed to delineate five classes of parenchymal textures in CT scans.
• CT scans from 46 non-small cell lung cancer patients treated with chemoradiotherapy were analyzed.
• Volumetric changes in RILD classes were quantified and correlated with dosimetric and respiratory parameters (FVC, MRC dyspnea scores).

Main Results:

• The deep learning network successfully segmented 230 CT scans into five distinct parenchymal classes.
• Distinct temporal patterns were observed for each of the five RILD classes.
• Moderate correlations were found between RILD class volume changes and radiotherapy dose (e.g., V30) and respiratory function decline (e.g., FVC).
• A strong dose-dependent relationship was observed between local radiation dose and tissue class changes.
• Radiological RILD correlated with spirometry and MRC dyspnea scores.

Conclusions:

• The developed deep learning system provides a quantitative and detailed analysis of RILD evolution.
• This approach enables a deeper understanding of the morphological and functional changes associated with RILD.
• The findings highlight the potential of AI in improving the assessment and management of radiation-induced lung toxicity.