Single time-point kidney dosimetry in177Lu-PSMA therapy: A comparison between AI-based and manual segmentation approaches
Issued Date
2026-01-01
Resource Type
eISSN
25396056
Scopus ID
2-s2.0-105016529425
Journal Title
Journal of Associated Medical Sciences
Volume
59
Issue
1
Start Page
9
End Page
17
Rights Holder(s)
SCOPUS
Bibliographic Citation
Journal of Associated Medical Sciences Vol.59 No.1 (2026) , 9-17
Suggested Citation
Chaiwongsa T., Charoenphun P., Chamroonrat W., Chuamsaamarkkee K. Single time-point kidney dosimetry in177Lu-PSMA therapy: A comparison between AI-based and manual segmentation approaches. Journal of Associated Medical Sciences Vol.59 No.1 (2026) , 9-17. 17. doi:10.12982/JAMS.2026.002 Retrieved from: https://repository.li.mahidol.ac.th/handle/123456789/114730
Title
Single time-point kidney dosimetry in177Lu-PSMA therapy: A comparison between AI-based and manual segmentation approaches
Author's Affiliation
Corresponding Author(s)
Other Contributor(s)
Abstract
Background: Single time-point (STP) dosimetry has become a practical and efficient approach for personalised radioligand therapy (RLT), with 48-hours post-injection identified as optimal for kidney dose estimation in ¹⁷⁷Lu-PSMA therapy for prostate cancer. However, segmentation accuracy remains a critical factor affecting dosimetry reliability. AI-based segmentation has recently been integrated into commercial software to improve efficiency and reduce variability. Objectives: This study aims to quantify kidney absorbed doses in patients receiving ¹⁷⁷Lu-PSMA therapy using STP dosimetry and to compare the accuracy and consistency of AI-based segmentation versus manual segmentation techniques. Materials and methods: Eight treatment cycles from 5 patients of ¹⁷⁷Lu-PSMA were retrospectively analysed. In this work, whole-body SPECT/CT imaging was performed approximately 48 hours post-injection. Then, kidney dosimetry was calculated using voxel-based STP (Hänscheid method) within MIM SurePlan™ MRT software. Kidney volumes of interest (VOIs) were segmented using three approaches: 1) AI-based automatic segmentation, 2) AI-based with manual refinement, and 3) fully manual segmentation. Mean absorbed doses and VOI volumes were compared across methods. Statistical analyses included ANOVA, Dice Similarity Coefficient (DSC), and Jaccard Similarity Coefficient (JSC). Results: No significant differences in mean kidney absorbed doses were found across segmentation methods (p=0.964), while kidney VOI volumes showed significant variation (p<0.05). AI-based segmentation achieved high concordance with manual delineation (DSC: 0.898±0.019; JSC: 0.816±0.031). Conclusion: AI-based segmentation provides comparable absorbed dose results to manual segmentation, with reduced time and inter-observer variability.