Novel dosimetric validation of a commercial CT scanner based deep learning automated contour solution for prostate radiotherapy.

PURPOSE: OAR delineation accuracy influences: (i) a patient's optimised dose distribution (PD), (ii) the reported doses (RD) presented at approval, which represent plan quality. This study utilised a novel dosimetric validation methodology, comprehensively evaluating a new CT-scanner-based AI contouring solution in terms of PD and RD within an automated planning workflow. METHODS: 20 prostate patients were selected to evaluate AI contouring for rectum, bladder, and proximal femurs. Five planning 'pipelines' were considered; three using AI contours with differing levels of manual editing (nominally none (AIStd), minor editing in specific regions (AIMinEd), and fully corrected (AIFullEd)). Remaining pipelines were manual delineations from two observers (MDOb1, MDOb2). Automated radiotherapy plans were generated for each pipeline. Geometric and dosimetric agreement of contour sets AIStd, AIMinEd, AIFullEd and MDOb2 were evaluated against the reference set MDOb1. Non-inferiority of AI pipelines was assessed, hypothesising that compared to MDOb1, absolute deviations in metrics for AI contouring were no greater than that from MDOb2. RESULTS: Compared to MDOb1, organ delineation time was reduced by 24.9 min (96 %), 21.4 min (79 %) and 12.2 min (45 %) for AIStd, AIMinEd and AIFullEd respectively. All pipelines exhibited generally good dosimetric agreement with MDOb1. For RD, median deviations were within ± 1.8 cm3, ± 1.7 % and ± 0.6 Gy for absolute volume, relative volume and mean dose metrics respectively. For PD, respective values were within ± 0.4 cm3, ± 0.5 % and ± 0.2 Gy. Statistically (p < 0.05), AIMinEd and AIFullEd were dosimetrically non-inferior to MDOb2. CONCLUSIONS: This novel dosimetric validation demonstrated that following targeted minor editing (AIMinEd), AI contours were dosimetrically non-inferior to manual delineations, reducing delineation time by 79 %.

Comparison of different calculation techniques for absorbed dose assessment in patient specific peptide receptor radionuclide therapy.

AIM: The present work concerns the comparison of the performances of three systems for dosimetry in RPT that use different techniques for absorbed dose calculation (organ-level dosimetry, voxel-level dose kernel convolution and Monte Carlo simulations). The aim was to assess the importance of the choice of the most adequate calculation modality, providing recommendations about the choice of the computation tool. METHODS: The performances were evaluated both on phantoms and patients in a multi-level approach. Different phantoms filled with a 177Lu-radioactive solution were used: a homogeneous cylindrical phantom, a phantom with organ-shaped inserts and two cylindrical phantoms with inserts different for shape and volume. A total of 70 patients with NETs treated by PRRT with 177Lu-DOTATOC were retrospectively analysed. RESULTS: The comparisons were performed mainly between the mean values of the absorbed dose in the regions of interest. A general better agreement was obtained between Dose kernel convolution and Monte Carlo simulations results rather than between either of these two and organ-level dosimetry, both for phantoms and patients. Phantoms measurements also showed the discrepancies mainly depend on the geometry of the inserts (e.g. shape and volume). For patients, differences were more pronounced than phantoms and higher inter/intra patient variability was observed. CONCLUSION: This study suggests that voxel-level techniques for dosimetry calculation are potentially more accurate and personalized than organ-level methods. In particular, a voxel-convolution method provides good results in a short time of calculation, while Monte Carlo based computation should be conducted with very fast calculation systems for a possible use in clinics, despite its intrinsic higher accuracy. Attention to the calculation modality is recommended in case of clinical regions of interest with irregular shape and far from spherical geometry, in which Monte Carlo seems to be more accurate than voxel-convolution methods.

Partial volume effect of SPECT images in PRRT with 177Lu labelled somatostatin analogues: A practical solution.

BACKGROUND: At present activity quantification is one of the most critical step in dosimetry calculation, and Partial Volume Effect (PVE) one of the most important source of error. In recent years models based upon phantoms that incorporate hot spheres have been used to establish recovery models. In this context the goal of this study was to point out the most critical issues related to PVE and to establish a model closer to a biological imaging environment. METHODS: Two different phantoms, filled with a 177Lu solution, were used to obtain the PVE Recovery Coefficients (RCs): a phantom with spherical inserts and a phantom with organ-shaped inserts. Two additional phantoms with inserts of various geometrical shapes and an anthropomorphic phantom were acquired to compare the real activities to predicted values after PVE correction. RESULTS: The RCs versus volume of the inserts produced two different curves, one for the spheres and one for the organs. After PVE correction, accuracy on activity quantification averaged over all inserts of three test phantoms passed from -26% to 1.3% (from 26% to 10% for absolute values). CONCLUSION: RCs is a simple method for PVE correction easily applicable in clinical routine. The use of two different models for organs and lesions has permitted to closely mimic the situation in a living subject. A marked improvement in the quantification of activity was observed when PVE correction was adopted, even if further investigations should be performed for more accurate models of PVE corrections.

Effect of image registration on 3D absorbed dose calculations in 177Lu-DOTATOC peptide receptor radionuclide therapy.

Peptide receptor radionuclide therapy (PRRT) is an effective MRT (molecular radiotherapy) treatment, which consists of multiple administrations of a radiopharmaceutical labelled with 177Lu or 90Y. Through sequential functional imaging a patient specific 3D dosimetry can be derived. Multiple scans should be previously co-registered to allow accurate absorbed dose calculations. The purpose of this study is to evaluate the impact of image registration algorithms on 3D absorbed dose calculation. A cohort of patients was extracted from the database of a clinical trial in PRRT. They were administered with a single administration of 177Lu-DOTATOC. All patients underwent 5 SPECT/CT sequential scans at 1 h, 4 h, 24 h, 40 h, 70 h post-injection that were subsequently registered using rigid and deformable algorithms. A similarity index was calculated to compare rigid and deformable registration algorithms. 3D absorbed dose calculation was carried out with the Raydose Monte Carlo code. The similarity analysis demonstrated the superiority of the deformable registrations (p < .001). Average absorbed dose to the kidneys calculated using rigid image registration was consistently lower than the average absorbed dose calculated using the deformable algorithm (90% of cases), with percentage differences in the range [-19; +4]%. Absorbed dose to lesions were also consistently lower (90% of cases) when calculated with rigid image registration with absorbed dose differences in the range [-67.2; 100.7]%. Deformable image registration had a significant role in calculating 3D absorbed dose to organs or lesions with volumes smaller than 100 mL. Image based 3D dosimetry for 177Lu-DOTATOC PRRT is significantly affected by the type of algorithm used to register sequential SPECT/CT scans.

Influence of voxel S factors on three-dimensional internal dosimetry calculations.

Internal dosimetry is a fundamental instrument for the personalization of nuclear medicine therapies, to maximize the therapeutic effect while minimizing the radiation burden to other organs. Three-dimensional (3D) dosimetry can quantify the impact of heterogeneous radiopharmaceutical distributions in organs, lesions and tissues. We analysed the influence of radionuclide voxel S factors in 3D dosimetry of 111In, 177Lu and 90Y, the most used radionuclides in Peptide Receptor Radionuclide Therapy (PRRT). Calculations were carried out for kidneys on a workstation equipped with a software for 3D dosimetry (Imalytics STRATOS, Philips AG), adopting a computational anthropomorphic phantom and, retrospectively, the SPECT-CT image series of a clinical case of PRRT. Two sets of voxel S factors were adopted: the pre-loaded Philips kernels, calculated by direct Monte Carlo simulation, and the ones calculated through a previously proposed analytical approach. Philips 111In kernel did not account for mono-energetic Auger or Conversion electrons. Results indicate a difference of about -32% in voxel S factors for 111In in 4.42mm voxel size and around -35% in 4.80mm voxel size, particularly self-dose values; this lead to significant shift in dose histograms and average doses. For 177Lu and 90Y, differences are about 2% and 12% for 4.42mm voxels and about -8% and 9% for 4.80mm voxels, respectively, attributable to the different calculation methods of the voxel S factors; this does not lead to significant discrepancies between the two dose histograms. Consequently, voxel S factors must account accurately for all radiations emitted by the nuclide.