SPECT Lung Delineation via True 3D Active Contours

In this paper, we developed an automated three-dimensional (3D) lung delineation method that is truly 3D in all aspects capable of handling single photon emission computed tomography (SPECT) lung scans with normal/low maximum count value (MCV) and/or total count value (TCV), defective contours, and/or extraordinary high counts due to hotspots. Four datasets consisting of (1) two sets of 50 randomly selected Monte Carlo simulations and real subjects with normal maximum and/or total count values, and (2) 90 simulations with low MCV and/or TCV and 35 real subjects with similar-ranged MCV/TCV were used as the basis of this study. A fast method was also developed to mass generate simulations with artificial hotspots, and the resulting set of 30 hotspot-infected simulations was also include in our dataset. After removing background noise using dual adaptive exponential thresholding (DUET), 3D Gaussian filter and 3D Sobel kernels are then used for edge enhancement, followed by final contour delineation via 3D active contours. Both quantitative validation and qualitative verification were implemented to evaluate the method. We achieved above 90% congruency overall for both simulations and subject scans that have low/normal MCV/TCV and hotspots.