A pilot study: Quantify lung volume and emphysema extent directly from two-dimensional scout images

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Pulmonary Medicine

Background:

• Computed tomography (CT) scans are crucial for diagnosing lung conditions like chronic obstructive pulmonary disease (COPD).
• Traditional analysis requires volumetric CT data, which is time-consuming and involves higher radiation exposure.
• Planar scout images, routinely acquired during CT, offer a potential alternative for rapid assessment.

Purpose of the Study:

• To investigate the feasibility of calculating emphysema volume metrics from standard CT scout images.
• To enhance the diagnostic utility of planar medical images through computational analysis.
• To develop and validate artificial intelligence models for emphysema quantification.

Main Methods:

• Utilized a cohort of patients with chronic obstructive pulmonary disease (COPD).
• Trained two convolutional neural networks (CNNs), VGG19 and InceptionV3, on 1,446 CT scout images.
• Used volumetric CT data as the ground truth for lung volume and emphysema percentage.
• Evaluated model performance using R-square (R2) and mean absolute difference (MAD).

Main Results:

• CNN models demonstrated significant linear correlation between scout image estimations and volumetric CT data (R2 up to 0.977 for lung volume).
• VGG19 achieved R2 of 0.934 for lung volume and 0.751 for emphysema percentage.
• InceptionV3 achieved R2 of 0.977 for lung volume and 0.775 for emphysema percentage.
• Mean absolute differences for lung volume and emphysema percentage were within acceptable clinical ranges.

Conclusions:

• Demonstrated the feasibility of inferring volumetric metrics of emphysema from planar CT scout images.
• Convolutional neural networks show promise for efficient and accurate emphysema quantification.
• This approach could significantly increase the diagnostic potential of routinely acquired planar medical images.