COMPARISON BETWEEN MANUAL AND AUTOMATIC SEGMENTATION OF THE WHOLE-BRAIN AND CEREBELLUM

Abstract

Introduction/Justification

18F-FDG PET/CT is widely used to quantify brain metabolic activity and plays a key role in studying various diseases. Segmentation method choice can significantly influence standardized uptake value (SUV) measurements, thereby affecting the accuracy of the analysis. The Beth Israel Plugin in ImageJ allows both manual and automatic segmentation, making it relevant for evaluating differences in brain and cerebellum analysis.

Objectives

This study aims to compare the mean, maximum, and peak SUVs obtained through manual and automatic (Grow Mask) segmentation of the brain and cerebellum, assessing the relative percentage differences and variations between the methods.

Materials and Methods

Seventy-three multiple myeloma (MM) patients who underwent 18F-FDG PET/CT were included in the study, comprising 43 men (58.9%) with a mean age of 64.2 ± 11.4 years. Brain segmentation was performed in FIJI using two methods: (1) manual segmentation (MS) consisting of a spherical volume of interest (VOI) of 6.7 mL for the cerebellum and 377 mL for the brain and (2) auto-segmentation (AS) using a Grown Mask algorithm. Manual cropping of PET images was performed before AS to exclude non-cerebellar regions. The relative percentage difference between the two methods was calculated as (1 - MS/AS). Mean, maximum and peak SUVs (SUVmean, SUVmax and SUVpeak, respectively), as well as maximum and minimum variation ranges of SUVs between MS and AS, were recorded.

Results

For the brain, SUVs were higher for AS compared to MS: SUVmean = 4.19 ± 0.02 (MS) vs. 5.99 ± 0.03 (AS), corresponding to 30.05% difference (range: 10.21% to 41.39%); SUVpeak = 8.07 ± 0.05 (MS) vs. 9.05 ± 0.06 (AS), 10.83% difference (range: 0% to 40.59%); and SUVmax = 10.76 ± 0.06 (MS) vs. 11.75 ± 0.07 (AS), 8.43% difference (range: 0% to 55.46%). For the cerebellum, a greater variability between MS and AS SUVs were found: SUVmean = 6.00 ± 0.03 (MS) vs. 5.47 ± 0.02 (AS), corresponding to -9.69% difference (range: 0% to 53.51%); SUVpeak = 7.06 ± 0.03 (MS) vs. 7.29 ± 0.03 (AS), 3.15% difference (range: 0% to 32.96%); SUVmax = 8.23 ± 0.04 (MS) vs. 9.20 ± 0.04 (AS), 10.54% difference (range: 0% to 42.57%).

Conclusion

The choice of segmentation method significantly impacts SUV values. AS yielded higher brain SUVs, while cerebellum MS showed greater variability due to manual adjustments and VOI selection. The differences between methods stem from segmentation techniques: MS used a spherical VOI, sometimes excluding the highest SUV point, whereas AS encompassed the full structure, capturing the true SUVmax. Thus, spherical VOI is less precise for whole-organ analysis but useful for quick regional calculations. Standardizing segmentation methods is crucial for reliable comparisons in clinical and research settings.