PT  - JOURNAL ARTICLE
AU  - David Mirando
AU  - Yuni Dewaraja
AU  - Stephen Graves
AU  - Katherine Krawiec
AU  - Aaron Nelson
TI  - Minimizing the clinical requirements for kidney dosimetry in 177Lu-DOTATATE with reduced timepoint imaging and neural network segmentation
DP  - 2020 May 01
TA  - Journal of Nuclear Medicine
PG  - 530--530
VI  - 61
IP  - supplement 1
4099  - http://jnm.snmjournals.org/content/61/supplement_1/530.short
4100  - http://jnm.snmjournals.org/content/61/supplement_1/530.full
SO  - J Nucl Med2020 May 01; 61
AB  - 530Objectives: Despite an increasing body of evidence demonstrating the predictive power of dosimetry for molecular radiotherapies and its potential utility in optimizing treatment planning, clinical adoption has been largely stunted by the cost and logistical burden to patients and the clinic. Both multi-time point imaging with 3-4 SPECT/CT timepoints (TPs) and manual image segmentation for patient specific dosimetry are associated with high demands on clinic resources. The goal was to evaluate fast and practical renal dose estimation tools implemented in a commercial software package with reduced (1 or 2) SPECT/CT imaging time points and a convolutional neural network (NN) based kidney auto-segmentation on CT. Methods: Serial SPECT/CT images were acquired after a total of 8 cycles of 177Lu-DOTATATE administered to 5 patients. Each serial dataset consisted of 4 TPs (25 min acquisitions), including one acquired ~4 hours post-therapy and one acquired ~4 days post-therapy. Images were quantitatively reconstructed using a commercial OSEM algorithm. Kidneys were segmented on the ~4-day scan both manually and using a NN. Adjusted NN contours were also generated by quickly reviewing the NN-generated contours and fixing gross errors. For dosimetry methods involving multiple imaging TPs, SPECT images were rigidly aligned to the ~4-day scan for each kidney independently. Each SPECT TP was converted into an absorbed dose rate map using convolution of a large voxel S value kernel [1]. Finally, a mono- or bi-exponential equation was fitted to the dose rate data using an automatic selection algorithm and integrated over time [2]. For the 2TP approach, the ~4-hour and ~4-day TPs were used, and monoexponential curve fitting was performed. Based on findings from a prior report [3], the ~4-day TP was used for the single-TP approach. The 4TP approach with manual kidney regions was used as the reference method for comparisons for right and left kidney absorbed doses. Results: The Dice coefficient score for NN-based kidney segmentation had median values of 0.86 (range: [0.71, 0.93]) and 0.90 (range: [0.81, 0.93]) for the right and left kidneys, respectively. The average time to review and adjust NN-generated kidneys for gross errors was 3.7 min (range: [0.4, 6.1]) for the right kidney and 2.4 min (range: [0.1, 4.3]) for the left kidney, for a total of 6.2 min of kidney adjustment per patient. Average absolute deviations for the kidney absorbed dose from the reference method (AVGDIFF) are in Table 1. Notably, 2TP methods had smaller AVGDIFF compared to 1TP methods, and quick adjustments to the NN-generated kidneys led to a large improvement in AVGDIFF and removal of the ~40% outlier. None of the 16 kidneys analyzed exceeded 17% AVGDIFF for the 1TP method or 11% for the 2TP method with NN-adjusted contours. View this table:Table 1: Average [range] absolute deviations in absorbed dose from the reference method Conclusions: A protocol using a single SPECT/CT and requiring minimum user intervention with NN-generated kidney contours may provide reasonably accurate estimates of kidney absorbed doses for the majority of patients. Although the contours should be reviewed and edited as needed because a few cases showed larger deviations, this process is much faster than manual contouring alone. Use of 2 imaging timepoints provides improved dose estimation accuracy compared with using a single TP and only requires an additional early scan before the patient leaves the clinic post-therapy. Furthermore, this may be a viable option at many clinics due to recent availability of CPT codes for reimbursement of post-treatment imaging for up to 2 SPECT/CT scans after each cycle of therapy. Future work includes expanding the NN training dataset with more diseased kidneys and more variability in abdominal fat and expanding the cohort of testing patients.