Improvement of Basal Ganglia Detectability in Brain Single Photon Emission Computerized Tomography by Wavelet Transformation in Image Processing Domain: A XCAT Phantom Study

Abstract

Purpose: Noise in brain Single Photon Emission Computed Tomography (SPECT) images limits an early diagnosis of Parkinson's Disease (PD). To overcome the limitation, as an image processing approach, wavelet transformation was used to denoising the images also with a segmentation method to differentiate the basal ganglia in brain SPECT. Materials and Methods: The brain scans of the human XCAT phantom through the Simulating Medical Imaging Nuclear Detectors (SIMIND) simulated SPECT system were imported to the MATLAB toolkit for image processing. The reconstructed brain images by iterative reconstruction were de-noised through 9 methods of wavelet transformation at different levels, and then six segmentation methods were applied to differentiate the caudate and putamen. The Dice coefficient, Specificity, and Sensitivity evaluation criteria were calculated based on the adaptive thresholding of the selected images from segmentation. A ground truth image was manually marked by a clinical nuclear medicine specialist. Results: The dice coefficient was obtained in a range from 0.3979 to 0.6299, as well as the specificity criterion from 0.7682 to 0.8168 and the sensitivity from 0.9049 to 0.9871. The results from adaptive threshold segmentation and the evaluation criteria showed that the best levels of the nucleuses detectability were provided by level 7 of Biorthogonal, levels 4 and 7 of Coiflet, level 6 of Daubechies, level 5 of Haar, level 6 of Morlet and level 6 of Symlet methods. Conclusion: Parkinson’s disease may be diagnosed in the early stage by an image processing approach to improve the quality of brain SPECT images.

Keywords

• Parkinson’s Disease
• Single Photon Emission Computed Tomography
• Simulating Medical Imaging Nuclear Detectors Monte Carlo
• Wavelet Transformation
• Segmentation