Image-based dosimetry for the pptimization of liver radioembolization with 90Y microspheres

• Κύριος συγγραφέας: Πλαχούρης, Δημήτρης
• Άλλοι συγγραφείς: Plachouris, Dimitris
• Γλώσσα: English
• Έκδοση: 2022
• Θέματα: Radioembolization; Yttrium-90; Deep learning; Biodistribution prediction; Treatment planning; Monte Carlo simulations; Ραδιοεμβολισμός; Ύττριο-90; Βαθειά μηχανική εκμάθηση; Πρόβλεψη βιοκατανομής; Σχεδιασμός θεραπείας; Προσομοιώσεις Μόντε Κάρλο

Liver cancer is the sixth most common form of cancer worldwide and remains the third leading cause of death. As the most effective method of treatment is surgery, portal vein thrombosis, which accompanies liver cancer in most cases, is a contraindication to surgery and other possible treatments (transplantation, chemoembolization). In contrast, there is no contraindication to the choice of 90Y microspheres radioembolization as a method of treatment. 90Y radioembolization is a modern form of treatment of liver cancer which is in the focus of medical research due to its therapeutic efficacy. It is a hybrid method of treatment that combines elements of embolization and brachytherapy for the isolation and topical treatment of liver cancer. This local treatment is achieved due to the differentiation of the blood supply of liver cancer (80% - 100% of the blood supply is of arterial origin) from the common supply of healthy liver tissue (venous origin). Taking advantage of this differentiation, 90Y microspheres are selectively delivered through the arterial network to the cancerous tissue. These microspheres inoculate the cancer-supplying microarticles (restriction of perspiration) and deposit their therapeutic dose locally (with 90Y electron emission). Nevertheless, the method’s therapeutic effect is limited by empirical dosimetric models that are currently available in clinical routine. The most modern dose model "partition model" offers a more personalized dosimetry based on an estimate of the dose deposited in the individual lobes (partitions) of the liver. However, the distribution of the microspheres is considered to be incorrectly homogeneous. Using the biodistribution of the 99mTc-MAA substituent imaging (available from the radioembolization imaging protocol), the 90Y biodistribution can be forecasted and this information can be used into the partition model aiming at a more personalized treatment. However, the use of 99mTc-MAA to forecast the biodistribution of 90Y microspheres is controversial, due to significant deviations from clinical indications, in the biodistribution of the two radioisotopes. These discrepancies are due to morphological differences (density, size, number of particles), deviations in catheter position between the two procedures and Imaging modality limitations (image noise, reconstruction method). The present dissertation entitled "Image-based dosimetry for the optimization of liver radioembolization with 90Y microspheres" combines medical information processing, modeling, simulations and machine learning technologies. It was primarily intended to optimize the assessment of the therapeutic dose deposited by the 90Y microspheres during liver radiotherapy with the aim of developing patient specific dosimetry based on Monte Carlo simulations. The developed 3D DVK-based dosimetry model was tested on 25 patients’ datasets (14 90Y post-treatment PET/CT scans and 11 99mTc-MAA pre-treatment SPECT/CT scans) and consequently validated against direct MC simulations. The comparison results of the measured absorbed dose using tissue-specific DVKs and direct MC simulation revealed a mean difference of 1.07 ± 1.43% for the liver and 1.03 ± 1.21% for the tumor tissue, respectively. The largest difference between DVK-based model and full MC dosimetry was observed for the lung tissue (10.16 ± 1.20%). Consequently, SPECT/CT and PET/CT clinical imaging data of 99mTc-MAA and 90Y microspheres were utilized, to predict a the biodistribution of the two radiopharmaceuticals using modern artificial intelligence algorithms. To this end, a Deep Learning model was developed and trained using the aforementioned data of 19 patients. The comparison between the real and predicted PET/CT scans showed an average absorbed dose difference of 0.44% ± 1.64% and 5.42% ± 19.31% for the liver and the tumor area, respectively. The average absorbed dose differences were 0.03 ± 0.25 Gy and 7.98 ± 31.39 Gy for the non-tumor liver parenchyma and the tumor, respectively.