Image-based kidney dosimetry for [177Lu]Lu-DOTATATE treatments Development, validation, and assessment of quantification accuracy and precision

Abstract

This thesis aims to develop, validate, and evaluate methods for accurate and precise image-based kidney dosimetry in patients with neuroendocrine tumours (NETs) treated with [177Lu]Lu-DOTATATE. Special focus is placed on quantifying and reducing uncertainties in kidney dose estimation. In paper I, an automatic segmentation method for the whole kidney parenchyma (WKP) was developed using a convolutional neural network (CNN). The CNN-based segmentation demonstrated performance comparable to manual (CT-based) segmentation, with a strong correlation in activity concentration measurements (r > 0.96, p < 0.01) indicating its suitability as an alternative for clinical dosimetry workflows. In paper II, the small-volume-of-interest (SV) method was optimized by adjusting VOI size and increasing the number of VOIs per kidney. This reduced dose estimation uncertainty from 14% when using a single 4 mL SV to 8.3% when using five 2 mL SVs, compared to the WKP method used as the reference. Papers II & III established a clinically feasible routine for calculating patient-specific recovery coefficients (RCs) using Monte Carlo simulations, reducing RC-related uncertainty by up to 4% compared to the fixed RC of 0.85 recommended by EANM guidelines. Paper III quantified anatomical changes in kidney volume (up to 10.77%) during treatment due to amino acid infusion. These changes were shown to result in absorbed dose overestimation of up to 2.5% when a single-timepoint CT segmentation was propagated across serial SPECT scans. In paper IV, a novel framework was introduced that uses an empirical power-law model to quantify the accuracy of WKP-based kidney dosimetry using SV data. The model demonstrated a relative precision of 7.27% for 2 mL SV and 6.92% for 0.6 mL SVs. In conclusion, this work presents a systematic evaluation of image-based kidney dosimetry methods, identifying key sources of uncertainty and demonstrating how they can be mitigated through optimized VOI strategies, patient-specific recovery correction, and CNN-based segmentation. A quantitative framework for estimating relative accuracy was also introduced, supporting the development of robust and clinically feasible dosimetry practices for individualized molecular radiotherapy in NET patients.