Medical imaging is a common diagnostic and research tool for application to diseases in the human body. The advent of advanced machine learning (ML) and neural network techniques has opened the possibility of learning more from imaging that has been previously possible. Beyond standard classification and segmentation applications of neural networks in imaging, there are questions about how sure we can be of the output of a neural network which is viewed as a black box. This talk will highlight work with three areas within the SOM all focused on medical imaging research. Neural network segmentation of images remains a primary application of AI algorithms, but the information returned by neural networks may be in question by the medical community (and AI researchers, too). A method of quantifying the uncertainty in segmented images, will be shown, and then discuss applying a modification to a sequential learning algorithm. Further, preliminary results in image artifact detection will be shown on OCT angiography images for large scale processing of OCTA images. The application of the AI/ML techniques are still at an infancy in medical imaging and new questions are being asked about how to apply simple and advanced AI in medical imaging for understanding disease progression.
Speaker Biography
Dr. Craig Jones earned a BSc (Hon) in Computer Science and Mathematics, an MSc in Medical Biophysics, and a PhD in Physics all with focus in image processing and numerical optimization algorithms on medical images. Dr. Jones completed a postdoctoral fellowship at KKI in the Kirby Center for Functional Brain Imaging doing numerical optimization and image processing work on advanced MRI acquisitions in collaboration with neurologists and neuroradiologists. He returned to Canada for a few years and worked at the Robarts Research Institute (UWO) in London, Ontario and looked at the quantification of uptake of agents in animal images based on numerical fitting of curves. He moved to a data science company, Spry, in Baltimore and applied machine learning / data science techniques for business applications. Imaging was always his interest and so he moved to the Space Telescope Science Institute in Baltimore and worked on a team that created imaging algorithms for creation, fixing and interpreting images from the James Webb Space Telescope. Then several years ago he returned to Johns Hopkins and has been working in the Malone Center for Healthcare in Engineering collaborating with numerous medical doctors in multiple departments within SOM. His interest is in AI/ML numerical optimization and image processing techniques applied to medical images.