Artificial intelligence-driven automated lung sizing from chest radiographs.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Transplantation Surgery

Background:

• Accurate donor-recipient lung size matching is critical for successful lung transplantation outcomes.
• Current size matching methods have limitations and lack a superior, standardized approach.

Purpose of the Study:

• To develop and validate an automated system using deep learning and computer vision for standardized lung size measurements from portable chest radiographs.
• To improve accuracy, reduce variability, and streamline the donor/recipient matching process in lung transplantation.

Main Methods:

• A two-step framework was developed: lung mask extraction from chest radiographs followed by feature point detection.
• The system generated six distinct lung height and width measurements.
• Measurements were validated against those of two expert radiologists for 50 lung transplant recipients.

Main Results:

• The automated system achieved less than 2.5% error (<7.0 mm) in lung size measurements.
• Demonstrated robust interrater and intrarater agreement compared to expert radiologist review.
• Performance was validated even on challenging radiographs with consolidations, effusions, and patient rotation.

Conclusions:

• Artificial intelligence shows significant potential for reproducible lung size assessment in clinical settings.
• This technology can facilitate large-scale studies on the impact of lung size matching on transplant outcomes.
• Further validation in a larger patient cohort is warranted.