RESTORATION OF Tc-99m MDP BONE SCAN IMAGES BY CLS FILTERING

1426 Objectives: Scintigraphy images are blurred by collimator response function and often appears noisy due to low counts. The image restoration method recovers original image from the observed image by simultaneously deblurring and reducing the noise. In this study, Tc-99m MDP bone scan images were restored by a constrained least square (CLS) restoration algorithm that uses a regularized filter. Methods: Fifty Tc-99m MDP bone scan images of thorax, knee, pelvis, feet & skull were subjected to CLS restoration algorithm. The algorithm is sensitive to noise power, Lagrange multiplier, and use of different smoothness constraints. Using the optimum value of Lagrange multiplier and Laplacian constraint on image smoothness as a constant, the restored images were obtained interactively by adjusting the value of noise power. Keeping in mind the quality of the sample output image based on consensus of two Nuclear Medicine physicians (NMP), interactive image restoration was performed. The initial value of noise power was estimated from the image. A rectangular region of interest (ROI) was placed on a homogeneous background area of the image ensuring that no structure is included in the ROI. The program (written in MATLAB 2020b) computes the standard deviation of pixel values in the ROI as the initial value of noise power. Noise power values of each image was recorded along with processed image for statistical analysis. Side by side analysis of input and output images were done by NMPs. NMPs were asked about acceptability of processed images. Results: 50 out of 50 restored images using constrained least square algorithm were acceptable to both NMPs. The statistical summary of Noise Power of the processed images are as follows: (Mean = 42.37, 1st Quartile = 17.16, Median = 38.56, 3rd Quartile = 55.38, Minimum = 4.88, Maximum = 119.33, Standard Deviation = 28.98) Figure-1 shows two sample images (observed and restored images). The restored images are smooth and have better resolution compared to input image. For example, margins of lytic lesions are more sharp (FIG: A & B) Conclusions: Tc-99m MDP Bone Scan images can be restored by constrained least square algorithm interactively and result seems to be promising.