A hybrid approach for reconstructing 3D model from oral contrast-enhanced CT colonography images

Abstract

This dissertation proposes a hybrid framework for 3D model reconstruction from oral contrast-enhanced CT colonography images. The framework is composed of three main parts which are colon cleansing, colon wall detection, and colon segmentation and 3D model reconstruction. The first part is colon cleansing which is performed by applying K-means clustering and morphological operations. The average Gaussian low pass filter of two different sizes combined with median filter are employed to reconstruct colon wall to be as realistic as possible. The examination was performed on four datasets. The results evaluated by an expert radiologist revealed that the accuracy on colon cleansing is satisfactory. The second part is colon wall segmentation which applies enhanced gradient vector flow and hybrid edge to assist colon wall detection. The examination was performed on eight datasets and compared with the existing techniques. The results evaluated by two expert radiologists revealed that the proposed method gives better results even in difficult cases. Afterward, colon segmentation based on anatomical structures and volume analysis, is applied to images obtained from the previous parts. Finally, 3D models are reconstructed from twenty datasets.