A randomization approach to handling data scaling in nuclear medicine

Area of Science:

• Medical Imaging
• Nuclear Medicine
• Data Science

Background:

• Medical imaging data often requires scaling for system complexity, like uniformity calibration.
• Conventional scaling methods (truncation, rounding) can alter data distribution, especially at low counts.
• This impacts the reliability of subsequent data analysis and applications.

Purpose of the Study:

• To evaluate the impact of conventional data scaling on medical imaging data.
• To introduce and assess a novel randomization approach for data scaling.
• To compare the effectiveness of randomization versus conventional methods in preserving data integrity.

Main Methods:

• Conventional scaling (truncation, rounding) and a new randomization approach were applied to gated cardiac SPECT studies.
• A randomization technique assigns integer values based on a generated random number, preserving floating-point information probabilistically.
• The performance was statistically analyzed on 50 clinical gated studies using scaling factors of 1.1 and 1.2.

Main Results:

• Conventional scaling noticeably altered myocardial perfusion defect size in an example study.
• The randomization approach resulted in scaled images identical to the original, with minimal data change.
• Significantly fewer quantitative (6% vs. 92%) and visual (2% vs. 58%) differences were observed with randomization compared to rounding for data scaled by 1.2.

Conclusions:

• The proposed randomization approach significantly minimizes local data changes introduced by scaling.
• Conventional scaling methods, particularly rounding, introduce noticeable quantitative and visual alterations.
• The randomization approach is recommended for nuclear medicine data scaling due to its superior data preservation.