Computed tomography-based machine learning for donor lung screening before transplantation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pulmonary Medicine

Background:

• Donor lung selection is currently subjective and lacks standardized criteria, leading to suboptimal utilization.
• The increasing demand for donor lungs necessitates objective screening methods to match the evolving donor pool.
• Ex vivo computed tomography (CT) imaging offers a potential solution for objective lung assessment.

Purpose of the Study:

• To investigate the efficacy of a CT-based machine learning algorithm for objective ex vivo donor lung screening.
• To develop an AI tool that can assist in identifying donor lungs suitable for transplantation and predict recipient outcomes.
• To enhance the accuracy of donor lung selection and reduce post-transplant complications.

Main Methods:

• A prospective clinical trial collected CT scans and clinical data from 100 donor lung cases.
• A supervised machine learning algorithm (dictionary learning) was trained on CT images to classify lung suitability.
• The algorithm's performance was evaluated against standard clinical assessments and recipient clinical outcomes.

Main Results:

• The machine learning algorithm successfully detected pulmonary abnormalities on donor lung CT scans.
• Among transplanted lungs, the algorithm identified recipients at significantly higher risk (19x) of chronic lung allograft dysfunction within two years.
• The algorithm's findings correlated with extended intensive care unit stays for certain recipients.

Conclusions:

• A novel strategy for ex vivo donor lung screening using CT and machine learning has been developed.
• Objective screening techniques are crucial for accurately assessing donor lungs and identifying high-risk recipients.
• This AI-driven approach aims to improve donor lung utilization and mitigate post-transplant complications.