Voxel-wise Prediction of Post-therapy Dosimetry for 177Lu-PSMA I&T Therapy using Deep Learning

Abstract

1424 Introduction: PSMA-directed radioligand therapy (RLT) has become one of the effective treatment options for metastatic castration-resistant prostate cancer (mCRPC). However, individual treatment planning is still not feasible as it is for the external beam radiotherapy. Our group has presented an organ-based research in the prediction of post-therapy dosimetry in 2019. However, an organ-based approach is unable to reveal the heterogeneity of dose distribution and therefore is not sufficient for the realization of treatment planning. In this study, we propose an approach for voxel-wise prediction of post-therapy dosimetry from pre-therapy positron emission tomography (PET) using deep learning. Materials and Methods: 30 patients with mCRPC treated with 177Lu-PSMA I&T RLT were retrospectively included in this study. Totally 48 treatment cycles with 68Ga-PSMA-11 PET/CT directly before the treatment and at least 3 post-therapeutic SPECT/CT dosimetry imaging were considered for this proof-of-concept study. Post-therapy voxel-wise dosimetry was calculated using Hermes Voxel Dosimetry. 3D RLT Dose generative adversarial networks (GANs) were developed with a 3D U-net generator and a convolutional neural network (CNN) based discriminator. A dual-input-model was designed to incorporate both information from PET and CT, for the purpose of anatomical coregistration. Both voxel-wise content loss alongside image-wise loss were taken into account for better synthesis performance. K-fold cross validation was applied to verify the trained network. Results: The proposed 3D RLT Dose GANs achieved the voxel-wise mean absolute percentage error (MAPE) of 17.56%±5.42%. The dual-input-model was able to synthesize dose maps with comparable accuracy while preserving anatomical consistency, which achieved a MAPE of 18.94%±5.65%. Conclusions: Our preliminary results demonstrate the potential of artificial intelligence to estimate voxel-wise post-therapy dosimetry both qualitatively and quantitatively. This may provide a practical solution to improve the dosimetry-guided treatment planning for RLT.