Image enhancement using Stochastic Resonance in ^99mTc-MDP bone scan images

Background: Stochastic resonance (SR) is a technique which enhances the weak signal by addition of noise. The classic signature of SR is the output signal-to-noise ratio (SNR) being greater than the input SNR when an appropriate amount of noise (usually Gaussian) is added to weak periodic input signals. In the current study, it was hypothesized that after adding an appropriate amount of noise to the half-time ^99mTc-MDP bone scan image, the resultant image will have more counts, hence the noise will reduce. However, to optimize the amount of noise and type of noise to add to the images is important. The aim of the study was to investigate whether a half-time ^99mTc-MDP bone scan image can be transformed into a full-time bone scan image using the principle of stochastic resonance. Materials & Methods: Stochastic resonance (SR) based image processing technique was developed and implemented using MATLAB R2013b on personal computer operating with Microsoft windows platform. One hundred one (101) set of bone scan patients referred to the nuclear medicine department for routine investigation were included in this prospective study. Each study included two scans: one was acquired with normal standard protocol and another one with exact half time of standard protocol acquisition time on a Siemens dual head gamma camera using low energy all purpose collimator. Gaussian noise (mean 0 and variance from 0.1 to 0.9) was added to the phantom images but in case of uniform noise, mean was kept zero and amplitude of noise varied from -a to +a, the value of ‘a’ was in steps 0.1 from 0.1 to 0.9. A power point presentation with four images (the input image, its corresponding half time image and 2 output images one with added gaussian noise and second with added uniform noise) was reviewed by Nuclear Medicine physician.

Results: Nuclear Medicine Physician on review, found that there was no perceptible loss in details of skeletal lesions and other high count areas like growth plates, even in half time images in 80 out of 100 images (80%). Some clinical problems like loss of demarcation in intervertebral space, intercostal spaces, femur in obese individuals occured in half time images in comparison to the normal standard time image. Only change that was consistently reported was that overall images appeared less bright with increase in noise and reduction in bone to background contrast. The rest of the 20 images were processed by a different method called dynamic stochastic resonance. This method was able to improve 15 images out of 20 images (75%) and cause minimal or equivocal improvement in 4/20 (20%) images. In only one image with initial low count statistics, deterioration was seen.

Conclusions: Half time ^99mTc-MDP image can be made equivalent to the normal standard bone scan image using stochastic resonance and dynamic stochastic resonance.