Manual Versus Artificial Intelligence-Based Segmentations as a Pre-processing Step in Whole-body PET Dosimetry...

Area of Science:

• Nuclear Medicine
• Radiology
• Medical Imaging

Background:

• Accurate radiation dose estimation is crucial for novel radiotracers.
• Organ segmentation is vital for dosimetry but is time-consuming and prone to interobserver variability.

Purpose of the Study:

• To evaluate the impact of AI-based automated organ segmentation on whole-body effective dose calculations.
• To compare the time efficiency and variability of AI segmentation versus manual segmentation in dosimetry.

Main Methods:

• PET/CT data from six patients imaged with ⁸⁹Zr-pembrolizumab were analyzed.
• Manual organ segmentations were compared with segmentations from an AI tool (RECOMIA).
• Whole-body effective doses were calculated using OLINDA/EXM based on residence times derived from both segmentation methods.

Main Results:

• AI-based segmentation reduced analysis time per scan from ~4 hours to ≤30 minutes.
• Median Jaccard similarity between manual and AI segmentations ranged from 0.05 (pancreas) to 0.78 (lungs).
• Whole-body effective doses showed minimal differences, with a median variation of 0.52% between the two methods.

Conclusions:

• Automated AI segmentation is a fast and effective pre-processing step for whole-body dosimetry.
• AI tools can help standardize dosimetry calculations and reduce interobserver variability.
• This approach supports optimized radiotracer administration and patient radiation safety.