The impact of angular separation on the performance of biplane correlation imaging for lung nodule detection
In this paper, we evaluate the performance of biplane correlation imaging (BCI) using a set of off-angle projections acquired from an anthropomorphic chest phantom. BCI reduces the effect of anatomical noise, which would otherwise impact the detection subtle lesions in planar images. BCI also minimizes the number of false positives (FPs) when used in conjunction with computer aided diagnosis (CAD) applied to a set of coronal chest x-ray projections by eliminating non-correlated nodule candidates. In BCI, two digital images of the chest are acquired within a short time interval from two slightly different posterior projections. The image data are then incorporated into the CAD algorithm in which nodules are detected by examining the geometrical correlation of the detected signals in the two views, thus largely "canceling" the impact of anatomical noise. Seventy-one low exposure posterior projections were acquired of an anthropomorphic chest phantom containing tissue equivalent lesions with small angular separations (0.32 degree) over a range of 20 degrees, [-10°, +10°], along the vertical axis. The data were analyzed to determine the accuracy of the technique as a function of angular separation. The results indicated that the best performance was obtained when the angular separation of the projection pair was greater than 6 degrees. Within the range of optimum angular separation, the number of FPs per image, FPpI, was ∼1.1 with average sensitivity around 75% (supported by a grant from the NIH R01CA 109074).
