Improved 3D automatic segmentation and measurement of pleural effusions

Pleural effusions are accumulations of fluid in the pleural space, usually associated with atelectasis of the adjacent lung. We have previously presented an automated method to measure the volume of pleural effusions on chest CT images. This paper presents an improved version of the same method, which adds 3D surface modeling and additional propagation of the segmentation in the inferior direction. The improved method is also more robust to noise. We compared this method to manual segmentations and the previous method by applying it to 15 chest CT scans. The new segmentation, on average, increased estimated effusion volume by 11%, bringing it closer to the expected average. In addition, the correlation between manual and automatic effusion volumes increased from .59 to .81 (p = .13), indicating a better segmentation.