This project has been developing and testing radiomic tools to predict pneumonia outcomes using data from MIDRC. The project builds on an automated pipeline developed by University Pennsylvania used previously to characterize lung parenchymal characteristics in chest CT and X-ray scans, which has performed well in a preliminary evaluation in pneumonia classification in chest CT scans. The investigators seek to develop and validate image analysis and machine learning tools to 1) predict clinical outcome (such as mortality and requirement of mechanical ventilation) using baseline chest radiographs and 2) predict progression of imaging infiltrates in temporal chest radiographs.

Current plans include

• Expansion of work to chronic diseases in which medical imaging plays a role: Data from MIDRC will be utilized to develop and test a tool that will leverage radiomics and deep learning to predict outcome utilizing transfer learnings from our current models to predict more immediate model outcomes, and investigating whether these are related or independent to further development of disease symptoms