Pediatric Brain Tumor Classification and Segmentation Using Transfer Learning from Adult Datasets

When treating patients with brain cancer, one of the first steps is to classify the tumor type. Classifying tumors involves assessing the appearance of tumor cells under a microscope. Cancer cells that have an appearance similar to normal cells are described as low-grade (grade 1) and cells that appear distinctly abnormal are considered high grade (grade 3). The classification of a tumor is critical to make treatment decisions. Another important task is tumor image segmentation. This involves using medical imagery to differentiate the tumor from normal brain tissue. Researchers are beginning to train computers (using algorithms, instructions that the computer uses to assess information) to complete tumor classification and segmentation. There are many structural differences between adult and pediatric brain tumors, but researchers hope that these analysis methods will be transferable. The goal of this project is to train a deep learning/CNN algorithm using adult brain MRI data and applying transfer learning techniques that will then be used with pediatric brain tumor data. Researchers will access the Pediatric Brain Tumor Atlas for invaluable information across tumor types that will help them train newly developed algorithms.