Predicting Lung, Heart, and Thoracic Cavity Volumes from Subject Demographics Using Machine Learning to Improve Lung Transplant

Overview

Lung transplantation is the standard treatment for end-stage lung diseases. A crucial factor affecting its success is size matching between donor’s lungs and recipient’s thoracic space. Computed tomography (CT) scans can accurately determine recipient’s lung size, but donor’s lung size is often unknown due to the absence of medical images. This study aims to predict donor's lung, thoracic cavity, and heart volumes from only subject demographics to improve the accuracy of size matching. A cohort of 4,610 subjects with chest CT scans and basic demographics, including age, gender, height, weight, race, smoking status, and smoking history, was used. The right and left lungs, thoracic cavity, and heart depicted on chest CT scans were automatically segmented using developed U-Net models, and their volumes were computed. Eight machine learning models were developed to predict the volume measures from subject demographics and validated using the 10-fold cross- validation method. The developed models showed promising performance in predicting thoracic cavity volume, right and left lung volume, total lung volume, and heart volume with R² ranging from 0.430 to 0.628, mean average error (MAE) ranging from 0.075 to 0.736 liters, and mean average percentage error (MAPE) ranging from 10.9% to 15.2%. Our results demonstrate the feasibility of using subject demographics to predict lung, heart, and thoracic cavity volumes. The developed tool may be used to facilitate lung size matching for lung transplant and ultimately improve post-transplant survival.

Competition history

• JSHS 2023 Category not listed

Resources

• National JSHS abstract book ↗

Related projects

Closest projects by meaning, across every fair and year in the corpus.