Organ volume estimation from magnetic sensor based 3D ultrasound data: application in gastric emptying

Abstract

The study of gastric emptying requires measuring the change of stomach volume with time after a meal. A series of cross-sectional images of the stomach were acquired with conventional ultrasound imaging. The position and orientation of each image in 3D space was registered by using a magnetic sensor which was attached on the ultrasound scanhead. The borders of the stomach on the images were outlined to identify a set of contours representing the surface of the stomach. The contours often intersect one another because of the curvature of the abdomen surface, the deep position of the stomach inside the abdomen, and the tilting of ultrasound scanhead.Two algorithms were developed for computing the volume of stomach from intersected contours. The first algorithm built a wireframe by sorting the connections between contours with vector dot-product of the connections and the central long axis of the stomach. The second algorithm deformed a pre-built wireframe to approximate the stomach of interest based on least square criterion. Both algorithms used Green and Gauss's theorem to compute the stomach volume, which is equivalent to dividing the whole volume into tetrahedra with respect to a spatial point and summing their signed volumes.The accuracy of the methods was evaluated by scanning an in vitro porcine stomach filled with different volumes. Thereafter, fourteen male volunteers were studied at various time point before and after ingestion of a 500-ml broth. The mean absolute difference between the estimated volumes of the porcine stomach and the true volumes in water displacement was 0.7% $\pm$ 0.4% (SD) with the dot-product sorting algorithm. Both methods were applied to the gastric emptying data. The average stomach volume after ingestion was expressed as (462.23 m1)-(11.97 ml) $\times$ (minutes after ingestion), r = 0.986, SEE = 22.65 ml, with half emptying time of 21.9 $\pm$ 4.5 minutes. The mean absolute difference between the two algorithms is 4.1% $\pm$ 2.8% (SD). The deformation algorithm performs slower and less accurately than the dot-product sorting algorithm, however, it has the potential of handling more irregular shaped organs. Both algorithms are also applicable to compute the volumes of other organs scanned with freehand 3D ultrasound.