Fuzzy Hybrid Method for the Reconstruction of 3D Models Based on CT/MRI Data

Abstract

This research proposes a hybrid method for improving the segmentation accuracy of reconstructed 3D models from computed tomography/magnetic resonance imaging (CT/MRI) data. A semi-automatic hybrid method based on combination of Fuzzy C-Means clustering (FCM) and region growing (RG) is proposed. In this approach, FCM is used in the first stage as a preprocessing step in order to classify and improve images by assigning pixels to the clusters for which they have the maximum membership, and manual selection of the membership intensity map with the best contrast separation. Afterwards, automatic seed selection is performed for RG, for which a new parameter standard deviation (STD) of pixel intensities, is included. It is based on the selection of an initial seed inside a region with maximum value of STD. To evaluate the performance of the proposed method, it was compared to several other segmentation methods. Experimental results show that the proposed method overall provides better results compared to other methods in terms of accuracy. The average sensitivity and accuracy rates for cone-beam computed tomography CBCT 1 and CBCT 2 datasets are 99 %, 98.4 %, 47.2 % and 89.9 %, respectively. For MRI 1 and MRI 2 datasets, the average sensitivity and accuracy values are 99.1 %, 100 %, 75.6 % and 99.6 %, respectively. The average values for the Dice coefficient and Jaccard index for the CBCT 1 and CBCT 2 datasets are 95.88, 0.88, 0.6, and 0.51, respectively, while for MRI 1 and MRI 2 datasets, average values are 0.96, 0.93, 0.81 and 0.7, respectively, which confirms the high accuracy of the proposed method.