Ziel/Aim:
The emerging PSMA-targeted radionuclide therapy (RLT) is an effective treatment for metastatic castration-resistant prostate cancer (mCRPC). The European council mandates that treatments should be planned according to the radiation doses delivered to individual patients. However, there is no method to predict the dosimetry before RLT, which hampers the realization of treatment planning. Therefore, we aimed to prove the concept to employ artificial neural networks (ANNs) to predict the post-therapy dosimetry.
Methodik/Methods:
A cohort of 43 patients with mCRPC received 177Lu-PSMA I&T including baseline 68Ga-PSMA-11 PET/CT. After first RLT cycle, the patients underwent 3 – 5 planar whole-body scans for purpose of dosimetry. Organ-based and whole-body average SUV uptake were obtained from pretherapy PET/CT scans. Blood test values (PSA, Albumin etc) were also included. Dosimetry was calculated for kidney, liver, spleen and salivary glands using Hermes Olinda 2. A 3-layer fully connected neural network was built up in Keras. 10-folder cross validation was applied to verify the trained network. Our results were compared with population-based dosimetry from literature. The influence of blood sample information on prediction was also assessed.
Ergebnisse/Results:
The proposed ANN achieved the dosimetry prediction error of 14.0 ± 12.4% for kidney, 15.7 ± 10.3% for liver, 77.5 ± 12.8% for salivary glands and 24.3 ± 16.1% for spleen. The inclusion of blood test didn't reduce the prediction error (p > 0.9), 15.8 ± 13.2% for kidney, 18.2 ± 9.4% for liver, 72.2 ± 15.8% for salivary glands and 28.1 ± 19.9% for spleen. In contrast, the prediction based on literature population mean has significantly larger error (p < 0.01), 46.2 ± 50.4% for kidney, 99.5 ± 238.7% for liver, 705.7 ± 377.7% for salivary glands.
Schlussfolgerungen/Conclusions:
The proof of concept study shows that ANN can significantly reduce the prediction error compared to generally population-based estimation. Artificial intelligence may provide a practical solution to improve the dosimetry-guided treatment planning for RLT.
