Radiation therapy (RT) has a proven record of efficacy in treating many forms of pediatric brain tumors. However, it is associated with long-term side effects due to damage to healthy tissue. This is especially important in the developing brain, where long-term deficits can be seen in the areas of intelligence, attention, memory and psychomotor processing. To mediate these deficits, there has been a push away from whole brain irradiation to more targeted treatment by using dose painting intensity modulated radiation therapy (DP-IMRT). However, in order to use these techniques, more information about how dosing to organs-at-risk (OARs) affects outcomes, including volumetric changes in the brain. Voxel Healthcare LLC (formerly Advanced Medical Systems LLC) is the developer of ClickBrain ? an automatic pediatric MR brain segmentation tool that uses cloud-based deep learning (Google TensorFlow) technology for radiology clinical decision support.
In Aim 1 a, we extend ClickBrain to ClickBrain RT ? a system that will combine ClickBrain's pre-treatment brain structure segmentation outputs with radiation planning CTs and MRs to calculate dosing to OARs. ClickBrain RT will also segment longitudinal MRIs (1 month, 6 months, 1 year, 2 years) to track outcomes via volumetric changes. We will use OAR dosing, demographics, tumor type and grade, chemotherapy information, OAR and tumor volumetric measurements to predict tumor and OAR volumetric outcomes. We will adapt our existing version of a multi-time point machine learning technique to do this prediction task.
In Aim 1 b, a user interface for this cloud computing-based proof-of-concept system will be built to allow the RT planner to import patient information and see changes in predicted longitudinal post-RT OAR and tumor volumes, based on adjusting OAR dosages for a particular patient. Our initial validation (Aim 2) will focus on an existing database of 51 germ cell tumor patients acquired as part of standard of care and previous studies at Children's Hospital Los Angeles. Germ cell tumors have relative uniform size and location and provide an ideal dataset to validate our proof-of-concept system. Our long-term goal for ClickBrain RT is to train the machine learning algorithm to provide optimized recommended OAR dosage ranges based on patient history and tumor information. Our software will allow radiation oncologists to optimize treatment and vastly improve long-term quality of life in pediatric brain tumor survivors.
Mitigating radiation toxicity due to radiation therapy for the treatment of pediatric brain tumors is important to avoid long-term developmental side effects. Our ClickBrain RT software will use cloud computing-based deep learning technologies to automatically delineate key structures in the brain and to train a machine learning algorithm to predict volumetric changes due to treatment-related radiation dosing to these key structures. This will enable physicians to better avoid unnecessary radiation doses that may cause long-term deficits in children with brain tumors.
