[Feasibility of using Statistical tests in evaluation of non-uniformity]

Non-uniformity test is essentially the only required daily QC procedure in nuclear medicine practice. Noise creates statistical variation or random error in a flood image. Non-uniformity on the other hand does not have statistical nature and may be regarded as systemic error. The present methods of non-uniformity calculation do not distinguish between these two types of error. The Jarque-Bera and Kolmogorov-Smirnov tests were examined as alternative methods in calculation of non-uniformity in flood test images. Using the Monte carol method, uniform and non-uniform flood images of different matrix sizes and count density were generated. The uniformity of the images was calculated using the present and proposed methods. The results were also tested using 1300 planar images of 128x128 matrix size. The proposed methods were more accurate and sensitive to non-uniformity at low count density. However, their precisions were less than the conventional methods. There were no significant differences between these procedures at high count density. The integral and differential uniformity do not distinguish between noise always present in the data or in occasional non-uniformity. In a uniform intact flood image, the difference between maximum and minimum pixel count [the value of integral uniformity] is much more than recommended values for non-uniformity. After filtration of image this difference decreases but still remains high. The proposed methods are more sensitive to non-uniformity at low count density and may be used as alternative methods in daily uniformity test

Alternative methods for evaluation of non-uniformity in nuclear medicine images

Non-uniformity test is the most essential in daily quality control procedures of nuclear medicine equipments. However, the calculation of non-uniformity is hindered due to high level of noise in nuclear medicine data. Non-uniformity may be considered as a type of systematic error while noise is certainly a random error. The present methods of uniformity evaluation are not able to distinguish between systematic and random error and therefore produce incorrect results when noise is significant. In the present study, two hypothetical methods have been tested for evaluation of non-uniformity in nuclear medicine images. Using the Monte Carol method, uniform and non-uniform flood images of different matrix sizes and different counts were generated. The uniformity of the images was calculated using the conventional method and proposed methods. The results were compared with the known non-uniformity data of simulated images. It was observed that the value of integral uniformity never went below the recommended values except in small matrix size of high counts [more than 80 millions counts]. The differential uniformity was quite insensitive to the degree of non-uniformity in large matrix size. Matrix size of 64'64 was only found to be suitable for the calculation of differential uniformity. It was observed that in uniform images, a small amount of non-uniformity changes the p-value of Kolmogorov-Smirnov test and noise amplitude of fast fouries transformation [FFT] test significantly while the conventional methods failed to detect the non-uniformity. The conventional methods do not distinguish noise, which is always present in the data and occasional non-uniformity at low count density. In a uniform intact flood image, the difference between maximum and minimum pixel count [the value of integral uniformity] is much more than the recommended values for non-uniformity. After filtration of image, this difference decreases, but remains high. Both proposed methods were more sensitive to the non-uniformity at a much lower count density