Enhancing voxel-based dosimetry accuracy with an unsupervised deep learning approach for hybrid medical image registration.

BACKGROUND: Deformable registration is required to generate a time-integrated activity (TIA) map which is essential for voxel-based dosimetry. The conventional iterative registration algorithm using anatomical images (e.g., computed tomography (CT)) could result in registration errors in functional images (e.g., single photon emission computed tomography (SPECT) or positron emission tomography (PET)). Various deep learning-based registration tools have been proposed, but studies specifically focused on the registration of serial hybrid images were not found. PURPOSE: In this study, we introduce CoRX-NET, a novel unsupervised deep learning network designed for deformable registration of hybrid medical images. The CoRX-NET structure is based on the Swin-transformer (ST), allowing for the representation of complex spatial connections in images. Its self-attention mechanism aids in the effective exchange and integration of information across diverse image regions. To augment the amalgamation of SPECT and CT features, cross-stitch layers have been integrated into the network. METHODS: Two different 177 Lu DOTATATE SPECT/CT datasets were acquired at different medical centers. 22 sets from Seoul National University and 14 sets from Sunway Medical Centre are used for training/internal validation and external validation respectively. The CoRX-NET architecture builds upon the ST, enabling the modeling of intricate spatial relationships within images. To further enhance the fusion of SPECT and CT features, cross-stitch layers have been incorporated within the network. The network takes a pair of SPECT/CT images (e.g., fixed and moving images) and generates a deformed SPECT/CT image. The performance of the network was compared with Elastix and TransMorph using L1 loss and structural similarity index measure (SSIM) of CT, SSIM of normalized SPECT, and local normalized cross correlation (LNCC) of SPECT as metrics. The voxel-wise root mean square errors (RMSE) of TIA were compared among the different methods. RESULTS: The ablation study revealed that cross-stitch layers improved SPECT/CT registration performance. The cross-stitch layers notably enhance SSIM (internal validation: 0.9614 vs. 0.9653, external validation: 0.9159 vs. 0.9189) and LNCC of normalized SPECT images (internal validation: 0.7512 vs. 0.7670, external validation: 0.8027 vs. 0.8027). CoRX-NET with the cross-stitch layer achieved superior performance metrics compared to Elastix and TransMorph, except for CT SSIM in the external dataset. When qualitatively analyzed for both internal and external validation cases, CoRX-NET consistently demonstrated superior SPECT registration results. In addition, CoRX-NET accomplished SPECT/CT image registration in less than 6 s, whereas Elastix required approximately 50 s using the same PC's CPU. When employing CoRX-NET, it was observed that the voxel-wise RMSE values for TIA were approximately 27% lower for the kidney and 33% lower for the tumor, compared to when Elastix was used. CONCLUSION: This study represents a major advancement in achieving precise SPECT/CT registration using an unsupervised deep learning network. It outperforms conventional methods like Elastix and TransMorph, reducing uncertainties in TIA maps for more accurate dose assessments.

Simplified image-based dosimetry using planar images and patient-specific S-values.

BACKGROUND: Single time point measurement approach and hybrid dosimetry were proposed to simplify the dosimetry process. It is anticipated that utilizing patient-specific S-value would enable more accurate dosimetry assessment based on imaging compared to using the conventional MIRD S-values. PURPOSE: We performed planar image-based dosimetry scaled with a single SPECT image for the entire treatment cycle using patient-specific S-values (PSS dosimetry) of organs. PSS dosimetry could further simplify the dosimetry procedure compared with a conventional 2D planar/3D SPECT hybrid dosimetry, as PSS dosimetry requires only one SPECT/CT image for the treatment of the entire cycle, whereas the conventional hybrid dosimetry requires a SPECT/CT image for each treatment cycle. METHODS: 177Lu-DOTATATE SPECT/CT and planar image datasets acquired from Seoul National University Hospital (SNUH, Seoul, Republic of Korea) were utilized for the evaluation. Images were acquired 4, 24, 48, and 120 h after patients' intravenous injection of 177Lu-DOTATATE. Dose estimations based on a Monte Carlo (MC) simulation using the Geant4 Application for Emission Tomography (GATE) (v.8.2) were considered as the reference. Planar image-based dosimetry scaled with a single SPECT image was performed using the patient-specific S-value (PSS). Briefly, the CT image was considered as the patient's anatomical reference and PSSs were quantified using the multiple voxel S-value (VSV) method. Then, PSS dosimetry was performed by obtaining activity information from sequential planar images and a scaling factor derived from a single SPECT/planar image pair. Hybrid dosimetry using sequential planar images and a single SPECT image was performed for comparison. The absorbed doses of the kidneys, bone marrow (BM) in the lumbar spine, liver, and spleen calculated using the PSS and hybrid dosimetries were compared with the reference MC results. RESULTS: The mean differences (MDs) of the self-absorption S-values between S-value of OLINDA/EXM and PSS for the kidneys, liver, and spleen were -0.04%, -2.39%, and -2.62%, respectively. However, the differences in the self-absorption S-values were significantly higher for the BM (84.99%) and the remainder of the body (ROB) (280.84%). The absorbed doses estimated by the PSS and hybrid dosimetries showed relatively high errors compared with MC simulation result, regardless of the organ. In contrast, the PSS and hybrid dosimetries produced similar dose estimates. For the entire cycles of the treatment, the MDs of absorbed doses between PSS and hybrid dosimetries were -3.31%, -6.04%, 3.37%, and -2.17% for the kidneys, BM, liver, and spleen, respectively. Through a correlation analysis and the Wilcoxon signed-rank test, we concluded that there was no significant difference between the results obtained by the two dosimetry methods. CONCLUSIONS: As the PSS was derived using CT images with actual anatomical information and organ-specific volume of interest (VOI), PSS dosimetry provided reliable results. PSS dosimetry was robust in estimating the absorbed dose for the later treatment cycles. Therefore, PSS dosimetry outperformed hybrid dosimetry in terms of dose estimation for a greater number of treatment cycles.

Voxel-Based Internal Dosimetry for 177Lu-Labeled Radiopharmaceutical Therapy Using Deep Residual Learning.

Purpose: In this study, we propose a deep learning (DL)-based voxel-based dosimetry method in which dose maps acquired using the multiple voxel S-value (VSV) approach were used for residual learning. Methods: Twenty-two SPECT/CT datasets from seven patients who underwent 177Lu-DOTATATE treatment were used in this study. The dose maps generated from Monte Carlo (MC) simulations were used as the reference approach and target images for network training. The multiple VSV approach was used for residual learning and compared with dose maps generated from deep learning. The conventional 3D U-Net network was modified for residual learning. The absorbed doses in the organs were calculated as the mass-weighted average of the volume of interest (VOI). Results: The DL approach provided a slightly more accurate estimation than the multiple-VSV approach, but the results were not statistically significant. The single-VSV approach yielded a relatively inaccurate estimation. No significant difference was noted between the multiple VSV and DL approach on the dose maps. However, this difference was prominent in the error maps. The multiple VSV and DL approach showed a similar correlation. In contrast, the multiple VSV approach underestimated doses in the low-dose range, but it accounted for the underestimation when the DL approach was applied. Conclusion: Dose estimation using the deep learning-based approach was approximately equal to that in the MC simulation. Accordingly, the proposed deep learning network is useful for accurate and fast dosimetry after radiation therapy using 177Lu-labeled radiopharmaceuticals.

Comparison of voxel S-value methods for personalized voxel-based dosimetry of 177 Lu-DOTATATE.

PURPOSE: Voxel-based dosimetry is potentially accurate than organ-based dosimetry because it considers the anatomical variations in each individual and the heterogeneous radioactivity distribution in each organ. Here, voxel-based dosimetry for 177 Lu-DOTATATE therapy was performed using single and multiple voxel S-value (VSV) methods and compared with Monte Carlo simulations. To verify these methods, we adopted sequential 177 Lu-DOTATATE single-photon emission computed tomography and X-ray computed tomography (SPECT/CT) dataset acquired from Sunway Medical Centre using the major vendor's SPECT/CT scanner (Siemens Symbia Intevo). METHODS: The administered activity of 177 Lu-DOTATATE was 7.99 ± 0.36 GBq. SPECT/CT images were acquired 0.5, 4, 24, and 48 h after injection in Sunway Medical Centre. For the multiple VSV method, VSV kernels of 177 Lu in media with various densities were generated by Geant4 Application for Emission Tomography (GATE) simulation first. The second step involved the convolution of the time-integrated activity map with each kernel to produce medium-specific dose maps. Third, each medium-specific dose map was masked using binary medium masks, which were generated from CT-based density maps. Finally, all masked dose maps were summed to generate the final dose map. VSV methods with four different VSV sets (1, 4, 10, and 20 VSVs) were compared. Voxel-wise density correction for the single VSV method was also performed. The absorbed doses in the kidneys, bone marrow, and tumors were analyzed, and the relative errors between the VSV and Monte Carlo simulation approaches were estimated. Organ-based dosimetry using Organ Level INternal Dose Assessment/EXponential Modeling (OLINDA/EXM) was also compared. RESULTS: The accuracy of the multiple VSV approach increased with the number of dose kernels. The average dose estimation errors of a single VSV with density correction and 20 VSVs were less than 6% in most cases, although organ-based dosimetry using OLINDA/EXM yielded an error of up to 123%. The advantages of the single VSV method with density correction and the 20 VSVs over organ-based dosimetry were most evident in bone marrow and bone-metastatic tumors with heterogeneous medium properties. CONCLUSION: The single VSV method with density correction and multiple VSV method with 20 dose kernels enabled fast and accurate radiation dose estimation. Accordingly, voxel-based dosimetry methods can be useful for managing administration activity and for investigating tumor dose responses to further increase the therapeutic efficacy of 177 Lu-DOTATATE.