Towards fully automatized [177Lu]Lu-PSMA personalized dosimetry based on 360° CZT whole-body SPECT/CT: a proof-of-concept.

Abstract

Background: The advent of 360° CZT gamma-cameras allows to conceive personalised dosimetry studies from whole-body SPECT/CT data. We aimed to demonstrate the proof-of-concept of an automated personalized dosimetry pipeline for [¹⁷⁷Lu]Lu-PSMA organ dosimetry, called SimpleDose, and to compare to other dosimetry approaches.

Methods: The organ segmentation is based on a nnU-Net framework that was trained to allow for the segmentation of 23 organs and structures over all the body. The method implemented to model the energy deposition is the collapsed-cone-superposition (CCS) taking into account non-uniform activity and density distributions. Ten patients with metastatic castration resistant prostate cancer treated [¹⁷⁷Lu]Lu-PSMA-617 were included. All SPECT/CT acquisitions were performed on a VERITON-CT 200 (Spectrum Dynamics®, Caesarea, Israel) from head to mid-thigh with 5 min per bed. The absorbed-dose-rates were computed with SimpleDose and compared with organ-level MIRD approach and local-deposition-method (LDM) for bone marrow, kidneys, liver, lungs, pancreas, salivary glands and spleen. Finally, an example of multi-time-point and single-time-point dosimetry is given.

Results: The median (IQR) calculation time with SimpleDose (SD), for segmentation, computation of dose-rates and descriptive statistics was 161 (23) seconds at a resolution of 2.46 × 2.46 × 2.46 mm³ (Intel Xeon 20 × 3.70 GHz CPU computer). The median (IQR) differences between SD and MIRD and LDM, were respectively 1.8 (61) % and  - 16 (76) % in bone marrow, 2.4 (1.5) % and  - 93.1 (0.4) % in kidneys, 2.9 (3.4) % and  - 9.2 (3.0) % in liver, 21 (13) % and 13 (13) % in lungs, 11 (3.3) % and  - 11 (3.0) % in pancreas, 1.1 (12) % and 3.8 (8.4) % in salivary glands, 4.0 (4.3) % and  - 10.0 (4.5) % in spleen. For the clinical example, the multi-time-point dosimetry with 4 time-points took 14 min, while the single-time-point approach took 3.5 min from day 1 dataset and 3.3 min from day 3.

Conclusion: The SimpleDose platform demonstrated its capability to compute organ-absorbed-dose rates in a simple and fast manner with close results to the standard MIRD approach for soft-tissues organs. SimpleDose is freely available for demonstration purpose as a Software as a Service (SaaS) at https://oncometer3d.com .