Data Harmonization: An Imaging-driven omics database / repository on COVID survivors for retrospective understanding of COVID disease and planning for future care.
Consistent imaging protocols will require normalization / harmonization of data sourced from multiple platforms, hospitals and vendors. AI has shown a remarkable ability to generalize and group / tease out patterns from high-dimensional data. Machine / deep learning algorithms should rely on mix-omics integration of imaging and physiological measures. There is an urgent need for new models of multi-modal transfer learning (e.g., understanding lung and heart functional interactions), and incremental learning as cohorts grow at an ever-faster face, combining data from states/countries.
We plan that imaging harmonization methods be applied on all discharged COVID patient imaging data. This phenotyping could lead to a better retrospective understanding of COVID disease pathways and prepare for future management of COVID-derived chronic pathologies. In addition, there are significant new chronic pathologies expected in COVID survivors (cardiomyopathy, pulmonary aspergillosis, hemoglobin / iron deficiencies) in the longer term, which will be challenging to treat and / or recognize. The harmonized baseline data during acute phase (US/CXR) and at discharge time (CT/CMR) would help tremendously in our ability to understand the implications of these pathologies.
The proposed harmonization platform would include normalization across vendors, sites, possible variations in protocols and patient size. We describe AI based harmonization methods to leverage a large number of baseline scans from existing and ongoing studies for density measures, texture and later airway topology. During this initial phase, the Columbia cohort would harmonize 2,500 subjects in total, sampling in proportion five distinct cohorts. In the long term we aspire to develop data sharing tools, with possible partnerships for long term / global infrastructure and computing, integrate expertise in multiple imaging modalities, lead an open AI approach to model, predict and understand stages of COVID-19.
Andrew F. Laine received his D.Sc. degree from Washington University (St. Louis) School of Engineering and Applied Science in Computer Science, in 1989 and BS degree from Cornell University (Ithaca, NY). He was a Professor in the Department of Computer and Information Sciences and Engineering at the University of Florida (Gainesville, FL) from 1990-1997. He joined Columbia University in 1997 and served as Vice Chair of the Department of Biomedical Engineering 2003 – 2011, and Chaired of the Department of Biomedical Engineering 2012 – 2017. He is currently the Percy K. and Vida L. W. Hudson Professor of Biomedical Engineering and Professor of Radiology (Physics).
His contributions and impact of research includes being among the first to use multiscale wavelet representations to enhance subtle details in mammograms and improve the detection of breast cancer. Today, the multi-scale algorithm he developed is used in almost all commercial digital mammography systems. His more recent / current research focuses on image analysis to classify pulmonary emphysema subtypes (COPD) in lung images and cardiac disease with the aim to identifying pathways of disease processes amenable to intervention and cure. He is also collaborating on research projects on applications of dual energy lung CT including generating virtual contrast images from non-contrast CT using deep learning techniques.
Professor Laine served on the IEEE ISBI (International Symposium on Biomedical Imaging) steering committee, 2006-2009 and 2009 – 2012. He was the Program Chair for the IEEE EMBS (Engineering in Biology and Medicine Society) annual conference in 2006 held in New York City and served as Program Co-Chair for IEEE ISBI in 2008 (Paris, France). He served as Area Editor for IEEE Reviews in BME in Biomedical Imaging since 2007-2013. He was Program Chair for the EMBS annual conference for 2011 (Boston, MA). Professor Laine Chaired the Steering committee for IEEE ISBI, 2011-2013, and Chaired the Council of Societies for AIMBE (American Institute for Medical and Biological Engineers) in 2012-2013. Finally, he served as IEEE EMBS Vice President of Publications 2008 – 2012, and was the President of IEEE EMBS 2015 and 2016. He is currently past-chair of the IEEE EMBS Technical Committee on Biomedical Health Informatics. He is a Fellow of IEEE, AIMBE and IFMBE.