Evaluation of Automatic Segmentation Tools on Low-Dose and Ultra-Low-Dose CT Images in PET/CT Scans

The use of Ultra-Low-Dose computed tomography (ULD-CT) in PET/CT imaging is gaining attention due to its potential to significantly reduce radiation exposure, which is crucial for patient safety. Accurate organ segmentation is essential for treatment planning, particularly in radiotherapy where it aids in delineating target volumes and organs at risk. This study investigated the reproducibility of organ segmentation on ULD-CT in PET/CT imaging. In a cohort of 28 patients undergoing [18F]FDG PET/CT scans, segmentation outcomes were compared between standard low-dose CT (LD-CT) and ULD-CT images. Segmentations were performed using the TotalSegmentator and MOOSE tools, and accuracy was quantified using metrics such as the Dice coefficient, Jaccard index, and Hausdorff and Average distance. Results indicated that ULD-CT, despite its significantly reduced radiation dose and increased image noise, achieved comparable segmentation accuracy to LD-CT for most anatomical structures, with Dice coefficients and Jaccard indices maintaining high values across segmented organs. However, lower reproducibility was observed for structures like the kidneys, and aorta, potentially due to the presence of adjacent structures with similar intensities, making accurate boundary delineation more challenging, especially in the noisy ULD-CT images. Quantitative PET analyses, represented by SUVmean and SUVmax, also showed consistent results between LD-CT and ULD-CT reconstructions. However, the segmentation tools exhibited some errors even on the LD-CT images, that means there are areas for further improvement in these tools. Overall, the study emphasizes that ULD-CT does not compromise the use of automatic segmentation tools like TotaSegmentator and Moose, maintaining the clinical utility of PET/CT scans for most anatomical structures while reducing radiation exposure.