Effects of Interpolation on the Volume Estimation of Pulmonary Nodules
The aim of this project is to explore the effects of various 3-dimensional interpolation methods on the segmentation and volume calculation of pulmonary nodules in a 3-dimensional image. Given a 3D image representing a CT scan of a human lung, the algorithm interpolates the input to a user-specified voxel size; the goal is to increase the accuracy of the automatic segmentation. The current preferred method for this purpose is trilinear interpolation, but tricubic interpolation may yield better accuracy. To this effect, three interpolation algorithms were implemented using the VisionX toolkit: nearest neighbor, trilinear and tricubic. Test synthetic and CT images are first interpolated and then converted to a binary image through various filtering and thresholding operations. The volume of the lung nodule may then be calculated, and is the primary metric of accuracy for the experiment. The hypothesis is that a tricubic-interpolated will yield a more accurate segmentation and volume calculation than its trilinear- or nearest neighbor-interpolated counterpart. The results indicate that tricubic interpolation provides a slight improvement in volume calculation accuracy over currently used methods.
Please use the navigation links above to explore various aspects of the project in more detail. The full report can be downloaded here, a poster outlining the report can be found here, and the full source code for the project can be found here.