High School Student Biomedical Engineering Society Irving, Texas, United States
Introduction: A phantom used for radiation therapy was created alongside a deep learning model used in fluoroscopic radiation therapy to detect tumors. The phantom was created using polyvinyl chloride, elastic resin, and plastic filament. After creating molds of organs such the liver, lungs, and heart, an arduino and hydralic pumps were programmed to simulate breathing with various patterns controlled by a remote. Finally, a deep learning model was used to localize tumor clusters and determine how radiation therapy would affect the patient.
Materials and
Methods: A synthetic anthropomorphic phantom was created in 1:4 ratio to simulate the movement of a patient’s liver motion and deformation during normal breathing. The phantom was consisted with spines, rib cages, lungs, liver, liver lesion, and muscles. The breathing motion was created by controlling the air flow pumping in and removing from the expandible lungs using a python-controlled Arduino board. These different organs and body were fabricated with different materials. The shapes were controlled by 3D printing technology. Two fabricating approaches were included in the study a) SLA printing, and b) mold casting. The skeletons and lungs were 3D-printed directly using SLA printer. The skeletons were printed with rigid acrylic-based resin. The lungs were created with elastic resin with 50 Shore A hardness which could be held up 2-to-3-times air volume changes with 2mm thickness of lung “tissue”. The liver lesion and body were casted with silicone (Shore 00-30 hardness) in 3D printed mold. The liver was casted with super soft Polyvinyl chloride (PVC) to have similar hardness as human liver. Silicone tubes were used to connect the lungs and air pumps for volume control. The DeepDRR model was tested on 2000+ images of chest, liver, and lung CT and MR scans, most of which contained other phantoms. Datasets were taken from kaggle.
Results, Conclusions, and Discussions: The motion phantom was scanned under 4DCT, MR simulator, and MR-LINAC for feasibility study. While the phantom was positioned in the machines, the air pump and Arduino control board was set at the console. The connecting air tubes were routed through the conduit to establish lung volume control. The intensity response between organs under different imaging modalities were as follows: Lung (CT/MR): 7-9 Hz Liver (CT/MR): 4-7 Hz Heart (CT/MR): 12-13 Hz
The developing motion-enabled, dual-modal (MR/CT) anthropomorphic phantom is aimed to establish a research platform for imaging technique and motion management solution testing under both MR and CT imaging. By mimicking human breathing motion could provide better understanding and better solution to represent the intrafraction lung and liver motions during radiotherapy. Many thoracic anthropomorphic phantoms are available in the market for CT/CBCT motion experiments; however, there is a great gap bringing the motion study to MR environment due to strong magnetic fields. A pneumatic-driven breathing system based anthropomorphic phantom is an approach without having any metallic parts in the motion platform. By pumping the air in and out of 3D-printed lungs, the expansion of the lung translates the motion/deformation to other body parts such as liver and chest to create realistic breathing motion. The controlling unit could stay outside of the vault without interfering with the magnetic fields.
Furthermore, the deep learning model had anexceptional ability to detect where the tumors (clumps of silicone) were located in correlation to vital organs such as the lungs, heart, and liver. The overall accuracy of the DeepDRR model came out to be over 90% for the validation accuracy.
Acknowledgements (Optional): UT Southwestern Medical Center & University - MAIA Lab (Radiation Oncology).
