MULTI-VIEW CNN FOR TOTAL LUNG VOLUME INFERENCE ON CARDIAC COMPUTED TOMOGRAPHY

Area of Science:

• Medical Imaging
• Pulmonary Physiology
• Artificial Intelligence in Medicine

Background:

• Total lung volume (TLV) is crucial for pulmonary function assessment but typically requires full lung computed tomography (CT) scans.
• Cardiac CT scans, which are more accessible, visualize a significant portion of the lungs, presenting an opportunity for TLV estimation.
• Existing methods for TLV estimation from limited imaging data have limitations in accuracy and reproducibility.

Purpose of the Study:

• To develop and validate a novel multi-view convolutional neural network (CNN) model for inferring total lung volume (TLV) from cardiac CT scans.
• To assess the performance of the proposed CNN model against existing regression-based TLV estimation techniques.
• To evaluate the accuracy and reproducibility of the CNN-based TLV estimation compared to traditional full-lung CT methods.

Main Methods:

• A multi-view convolutional neural network (CNN) model was trained using supervised learning.
• The model utilized paired full-lung and cardiac CT scans from the Multi-Ethnic Study of Atherosclerosis (MESA) dataset.
• The CNN was designed to infer total lung volume (TLV) from cardiac CT images, which cover approximately two-thirds of the lung volume.

Main Results:

• The developed CNN model significantly outperformed existing regression models in estimating total lung volume (TLV).
• The accuracy and reproducibility of the TLV estimations derived from cardiac CT scans were comparable to those obtained from full-lung CT scans.
• The model demonstrated robust performance, achieving results similar to the inherent scan-rescan variability of TLV measurements on full-lung CT.

Conclusions:

• A novel multi-view CNN approach enables accurate and reproducible estimation of total lung volume (TLV) from widely available cardiac CT scans.
• This method provides a valuable tool for pulmonary research and epidemiological studies where full lung CT is not feasible.
• The findings suggest that cardiac CT imaging can be leveraged for reliable lung volume assessment, expanding its utility beyond cardiac evaluation.