The overall purpose of the ML4MI initiative is to foster interdisciplinary collaboration between machine learning (ML) experts and medical imaging researchers at the University of Wisconsin, in order to develop and apply state-of-the-art ML solutions to challenging problems in medical imaging. This initiative responds to rapidly growing interest in ML techniques within medical imaging research, due to the unprecedented potential to solve challenging problems in areas such as image reconstruction, image processing, and computer-aided diagnosis.
A regular seminar series began in February 2018, and includes 1) seminars describing technical developments in ML with potential biomedical applications, 2) seminars by local or external Radiology researchers, describing problems that may benefit from ML approaches and ongoing projects involving ML techniques, and 3) seminars by biomedical researchers (not in Radiology), describing pioneering experiences applying ML in their fields of study. The seminar location will alternate between ECB/WID and SMPH/WIMR. These seminars will also provide an opportunity for UW researchers to become familiar with researchers “on the other side of campus.”
This workshop will be held on October 5, 2018 at the UW Fluno Center. The one-day workshop will further facilitate face-to-face communication and discussion between researchers from various backgrounds and encourage collaboration. The event will include major presentations by leaders in the fields of ML and Radiology, discussion panels, and poster presentations by junior researchers and students.
Register for the ML4MI 2018 Workshop
Call for poster presentations: We are soliciting poster submissions for the ML4MI 2018 workshop. Interested presenters should send an abstract of at most one page to Diego Hernando (email@example.com) by September 5. Abstracts should include a title and names/affiliations of the presenting author and co-authors. Abstracts may include figures if desired, and overlap with submissions to other conferences is OK. We especially encourage submissions from students and junior researchers. Invitations to present a poster for selected submissions will be sent out by September 19.
Pilot grant proposals – 2018 RFA now closed
The purpose of this pilot grant program is to foster interdisciplinary collaboration between ML experts and medical imaging clinicians and researchers at the University of Wisconsin’s Departments of Radiology and Medical Physics, and College of Engineering. Specific topics of interest include the development and characterization of novel ML methods with significant medical imaging applications, and the development and validation of new imaging applications for state-of-the-art ML methods.
First Annual UW Deep Learning for Medical Imaging Bootcamp
On July 23rd and July 25th, the Department of Radiology hosted Deep Learning Bootcamps as part of the Machine Learning for Medical Imaging initiative. The two bootcamps served nearly 50 interested students, post docs, scientists, and faculty from several UW departments including Radiology, Medical Physics, Electrical and Computer Engineering, Biomedical Engineering, and others to provide hands-on experience in working with deep learning in the context of medical imaging.
The principal organizers and instructors of the event were Tyler Bradshaw, PhD, Jacob Johnson, MS, and Alan McMillan, PhD, from the Department of Radiology and Kevin Johnson, PhD, from Departments of Medical Physics and Radiology.
While the course included discussion of the underpinnings of neural and deep learning networks, the focus was to provide hands-on exercises for participants to gain experience in creating and working with deep learning networks. The hands-on exercises demonstrated the capabilities of deep learning in areas such as detection of disease from chest radiographs, determination of MRI modality, segmentation of lung CT images, conversion of T1-weighted MR images into T2-weighted images, and reconstruction of MR k-space data using a deep learning network. The source code for the exercises has been posted on GitHub.