Artificial Intelligence for Pediatric Cancer

Stanford University, Palo Alto, CA

Imaging tests are essential for diagnosing cancers in children and for monitoring tumor response to therapy. By combining magnetic resonance imaging (MRI) and positron emission tomography (PET), pediatric cancers can be imaged with 80% reduced radiation exposure compared to traditional imaging tests. However, the acquisition of combined MRI and PET scans takes about 3-4 times longer (60-90 minutes) compared to traditional combined PET and computed tomography (CT) scans (20-30 minutes). These long acquisition times are a major bottleneck for clinical translation of PET/MRI technologies. To achieve faster diagnoses and higher throughput, we propose to develop novel deep convolutional neural networks (Deep-CNN) that can accelerate PET/MR image data acquisition and interpretation. The goal of our project is to develop Deep-CNNs for accelerated PET and MR image data acquisition of children with cancer. We will accomplish this goal by using deep convolutional neural networks (Deep-CNN) to reconstruct high resolution MR and PET images from low resolution inputs. In addition, we will train our Deep-CNN to classify responders and non-responders based on the image data. To the best of our knowledge, this is the first attempt to apply Deep-CNN to pediatric oncology applications. Results will be readily translatable to the clinic and thereby, will have major and broad health care impact.