A Monte Carlo approach to handling data scaling in nuclear medicine

In medicine imaging, data scaling is sometimes desired to handle the system complexity, such as uniformity calibration. Since the data are usually saved in short integer, conventional data scaling will first scale the data in floating point format and then truncate or round the floating point data to short integer data. For example, when using truncation, scaling of 9 by 1.1 results in 9 and scaling of 10 by 1.1 results in 11. When the count level is low, such scaling may change the local data distribution and affect the intended application of the data. In this work, we use a gated cardiac SPECT study to illustrate the effect of conventional scaling by factors of 1.1 and 1.2. We then scale the data with the same scaling factors using a Monte Carlo approach in which a random number evenly distributed between 0 and 1 is generated to determine how the floating point data will be saved as short integer data. If the random number is between 0 and 0.9, then 9.9 will be saved as 10, otherwise 9. In other words, the floating point value 9.9 will be saved in short integer value as 10 with 90% probability or 9 with 10% probability. The image reconstructed from the original data showed a perfusion defect at the apex of the myocardium. The defect size was noticeably increased by scaling with 1.1 and significantly decreased by scaling with 1.2 using the conventional approach. Using the Monte Carlo approach, in contrast, the images from the scaled data appeared identical to the original image. In conclusion, conventional approaches for data scaling can lead to unexpected local count variation and significantly affect the clinical usage of the data. The proposed Monte Carlo approach minimizes the scaling-introduced local data variation and is preferred for nuclear medicine data scaling.