Show simple item record

Federated Learning for Classification of COVID-19 Severity based on Chest X-Rays

dc.contributor.authorBeaubien, Matthew
dc.contributor.advisorHadjiyski, Lubomir
dc.date.accessioned2023-06-08T20:21:55Z
dc.date.available2023-06-08T20:21:55Z
dc.date.issued2023
dc.identifier.urihttps://hdl.handle.net/2027.42/176962
dc.description.abstractAI models require large datasets to be sufficiently accurate and generalizable for clinical use. Usually, the datasets of individual institutions are small, and due to the nature of privacy laws concerning medical data, it’s very difficult to combine institutional datasets into a larger dataset to reliably train these models. Federated learning is an approach that helps mitigate this problem. Federated learning refers to how developers can utilize training data from various institutions for building diagnosis models without each developer having access to every other dataset. In particular, there are various “sites” each of which has access to its own set of data, and the sites cannot share data. There are different ways to apply the federated learning paradigm for training models including various collaborative learning approaches where the individual institutions train the model on their local data and then exchange only the model with the other institutions. The other institutions will subsequently train this model on their local data. This way, a model is trained on all available data, but no institution is required to send data to another institution, preserving the privacy of the data. During the COVID pandemic, multi-institutional datasets of chest x-rays were compiled from patients at varying levels of COVID severity. This data is publicly available and categorized into various time periods and institutions of origin. This project utilizes this multi-institutional data and employs various collaborative federated learning techniques for training machine learning methods for COVID severity classification. Some methods we investigate include transferring model weights between sites, and combining them after concurrent training. The results will be analyzed to determine which methods yield the best results. It is expected that these federated learning techniques will efficiently utilize the multi-institutional dataset and help obtain accurate models.
dc.subjectmachine learning
dc.subjectAI
dc.subjectradiology
dc.titleFederated Learning for Classification of COVID-19 Severity based on Chest X-Rays
dc.typeProject
dc.subject.hlbtoplevelEngineering
dc.contributor.affiliationumComputer Science
dc.contributor.affiliationumcampusAnn Arbor
dc.description.bitstreamurlhttp://deepblue.lib.umich.edu/bitstream/2027.42/176962/1/CapstoneReport_COVID19CXR_-_Matthew_Beaubien.docx
dc.description.bitstreamurlhttp://deepblue.lib.umich.edu/bitstream/2027.42/176962/2/CapstonePresentation_-_Matthew_Beaubien.pptx
dc.identifier.doihttps://dx.doi.org/10.7302/7698
dc.working.doi10.7302/7698en
dc.owningcollnameHonors Program, The College of Engineering


Files in this item

Show simple item record

Remediation of Harmful Language

The University of Michigan Library aims to describe library materials in a way that respects the people and communities who create, use, and are represented in our collections. Report harmful or offensive language in catalog records, finding aids, or elsewhere in our collections anonymously through our metadata feedback form. More information at Remediation of Harmful Language.

Accessibility

If you are unable to use this file in its current format, please select the Contact Us link and we can modify it to make it more accessible to you.