Imaging Morphology of Pancreas in Diabetic Patients following Acute Pancreatitis (IMMINENT)

Machine and Hybrid Intelligence Lab will act as the AI Core lab for the IMMINENT study. Dr. Bagci (PI) joins the study as the director of the AI Core Lab and will coordinate the project’s technical and clinical aspects with Dr. Yazici (University of Illinoi at Chicago) and other PIs of the IMMINENT study. Dr. Bagci will be a bridge between clinical sciences and artificial intelligence for imaging and imaging analysis aspects of the IMMINENT study. As an experienced AI scientist in the field and holding a relevant NIH R01 project for pancreatic cyst detection, pre-cancer diagnosis and analysis with MRI, Dr. Bagci will provide oversight for the entire project and will ensure that scientific agenda and research plan are being followed. He will maintain communication within the IMMINENT study group through Dr. Yazici and other investigators via biweekly (with Dr. Yazici) and monthly meetings (T1DAPC IMMINENT study group). Dr. Bagci will supervise a postdoc researcher at the AI core lab towards all specific aims of the study for the AI core. Mainly, Dr. Bagci and his team will develop novel machine learning algorithms for MRI analysis of pancreas in health and diseases. More specifically, the following analysis will be included (but not limited to): • Deep Learning based segmentation of pancreas and measurement of pancreas volume, • Deep Learning segmentation of pancreatic necrosis and measurement of necrotic volume • Automatic segmentation of psoas muscle and measurement of psoas muscle volume • Assessment of pancreas shape metrics including surface area to volume ratio, bounding box and convex hull volume and solidity, bounding ellipsoid volume and principal axes lengths, and exploring other morphometric and geometric parameters • Classification of euglycemic, pre-DM and DM cohorts using clustering and more advanced prediction algorithms AI Core Lab will develop AI prediction systems for pancreas MRI analysis. Particularly, these algorithms will be explainable in decision space and interpretable in model construction. Additionally, the AI Core Lab will conduct throughout analysis with external data to test the trustworthiness of AI algorithms being developed. An expected number of MRI data set is 1000 MRIs. MRIs will be collected by the core image analysis lab (CIAL) at the Indiana University and quality control will be ensured prior to image transfer to ACL. Dr. Bagci, Dr. Yazici and the participating postdoc will work together to confirm the quality of the studies with MRI pre-processing methods such as intensity standardization, inhomogeneity correction, and noise removal. Robustness, generalization, and trustworthiness of the algorithms will be tested in comparisons with several baselines from the deep learning field. Training from scratch as well as transfer learning and domain adaptation methods will be tried too. Dr. Bagci has developed the state-of-the-art pancreas segmentation algorithms for both CT and MRI and those methods will be used as competitive baseline methods first, and will be compared with potentially similar methods from U-Net, Dense-Res-U-Net, and Capsule based segmentation algorithms such as SegCaps. Once the algorithms are optimized for the MRI segmentation, calculated quantification parameters (from segmentation, prediction, and morphologic measurements) will be uploaded for each MRI scan into the IMMINENT study database. Segmentation masks for the pancreases, pancreatic necrosis, and psoas muscles into IMMINENT study database.