March/April 2026 Issue
Cooperative Imaging
By Jessica Zimmer
Radiology Today
Vol. 27 No. 2 P. 22
Collaborative Program Aims to Improve Medical Imaging
In January 2025, the University of California, San Francisco’s (UCSF) department of radiology and biomedical imaging and GE HealthCare initiated a joint research program called Care Innovation Hub. The program will improve patient care and technologies in three areas: medical imaging, neurological and neurodegenerative disease, and precision oncology. Care Innovation Hub is a collaboration through which physicians, researchers, engineers, and managers from the two organizations work together on research studies to address meaningful clinical challenges.
UCSF’s research is occurring at three of its San Francisco campuses: Parnassus, Mission Bay, and China Basin. GE HealthCare staff are participating in offices around the world, including Milwaukee; Haifa, Israel; and the United Kingdom.
“Each of the three pillars has a list of projects. For example, one project under ‘Imaging Service of the Future,’ led by Peder Larson, PhD, a professor in the department of radiology and biomedical imaging at UCSF, is ‘Maximizing the Utility of Accelerated MRI Image Reconstruction.’ The projects will refine technologies and change clinical practices to improve patient care. Each of the pillars is driven by the need to improve the technologies we use today,” says Sharmila Majumdar, PhD. Majumdar is a professor of radiology and Margaret Hart Surbeck Distinguished Professor in Advanced Imaging at the UCSF School of Medicine.
The Care Innovation Hub is the latest program between UCSF and GE HealthCare. The two organizations have been working together since the 1980s.
“The examination of clinical work in medical imaging at UCSF is part of what led to the emergence of MRI as an important tool in radiology,” Majumdar says.
Support for Care Innovation Hub includes equal contributions from UCSF and GE HealthCare. The program will run for three years. Due to the long timeline for project signings, some projects just began this winter. Others started last January.
“The point is to bring academia and industry together, to accomplish things we cannot do alone” says Christina Rombola, senior director of research and scientific affairs for the United States and Canada for GE HealthCare. “We want to make sure we [GE HealthCare] are developing, creating, and translating innovation that will make a difference in clinical settings.”
Advancing Imaging
On the edge of San Francisco in UCSF’s Mission Bay campus, Andreas Rauschecker, MD, PhD, studies how AI and advanced neuroimaging techniques can map the brain. The data he collects will determine what images can relay about the diagnosis and progression of neurodegenerative diseases like dementia. Rauschecker serves as the head of the brain health and neurodegenerative disease pillar of Care Innovation Hub.
“Our studies will indicate which patients are likely to respond to treatment, whether they’ll experience side effects, and how serious those side effects may be,” Rauschecker says. He is also assistant professor of radiology, codirector of the Center for Intelligent Imaging, and associate director of the neuroradiology fellowship program at the UCSF School of Medicine.
Rauschecker had a number of conversations with GE HealthCare researchers before Care Innovation Hub launched. The talks focused on the questions physicians want answered, which technologies they are currently using, and what changes in those technologies could answer their questions.
“Right now, we use state-of-the-art high-field 3 T magnets at UCSF’s Weill Institute for Neurosciences,” Rauschecker says. “We’re really focused on developing advanced imaging protocols using these tools.”
One area of UCSF’s work is in protocols for diffusion MRI, a technology that is widely used to trace white matter pathways in the brain. White matter creates a network of myelinated axons in tissues deep within the brain. These axons enable signaling between regions of the brain.
“We know from earlier studies on animals that alterations in the white matter are early markers of neurodegenerative disease,” Rauschecker says. “We are now using a multishell diffusion imaging sequence (a specific MRI technique that involves acquiring diffusion-weighted data at different values) to gain further insights into the white matter. Over time, this practice will become a standard part of our clinical protocol.” He says GE HealthCare has devised devices that acquire and analyze diffusion-weighted data to operate more rapidly and make higher-quality measurements.
Quantitative Values
Another objective of UCSF researchers is to encourage GE HealthCare engineers to find ways to obtain state-of-the-art quantitative T 1 maps in patients. T 1 maps use magnetic signal to quantitatively assess damage, such as myelin loss, lesions, edema, and microstructural changes. Such types of damage can indicate the progression of diseases like multiple sclerosis.
“Most MRI is not quantitative. It gives arbitrary values of units,” Rauschecker says.
A Magnetization Prepared 2 Rapid Acquisition Gradient Echoes (MP2RAGE) sequence is a dual-acquisition approach that acquires two distinct gradient echo volumes at different times. It allows for the creation of a uniform image that has less bias than an image generated by a standard MPRAGE sequence.
“The MP2RAGE sequence helps us get quantitative values to compare the brain images between individuals,” Rauschecker says. “Then we can try to understand what’s different about the neurobiology of patients who progress faster, in terms of degeneration. That’s what is going to lead to more tailored treatments.”
Rauschecker’s current project for Care Innovation Hub is titled “Advanced Imaging of Clinical Patients With Suspected Neurodegeneration to Predict Amyloid-Related Imaging Abnormalities (ARIA).” Amyloid refers to abnormal, insoluble fibrous protein aggregates that are found in tissues and organs. Amyloid can disrupt body structures and functions.
“This is where AI comes in,” he says. “We use the advanced imaging data and AI algorithms together in the clinic to understand patterns of disease that are more likely to progress faster or slower in specific patients. It’s a wonderful tool to understand what is going on in these data sets.”
Rauschecker adds that none of the data will affect patients’ treatment at this stage. “It’s not clear when that would change. That’s hard to predict,” he says.
Enabling AI
One issue that arose at the beginning of the collaboration was debugging connections between UCSF’s research and clinical units. UCSF also had to develop methods to encrypt and secure patient data. Such action ensured the data would be safe from hackers during data exchanges in the cloud as well as phone and video conferences.
“[UCSF] still needs to protect patient information in accordance with the HIPAA Act,” Majumdar says.
Dynamic and evolving regulations relating to AI and data sharing presented another concern. Teams at UCSF engaged in a fair amount of reading and discussion of obligations to understand respective and shared requirements. Still, AI innovations are worth the time spent.
“The rising number of patients, the burnout among radiologists, the aging population, and the increasing burden of neurodegenerative diseases … we’re seeing this every day at every institution, which is why we have a focus on AI-driven technologies that enable more efficient workflow from end to end,” says Suchandrima Banerjee, PhD, senior global director for neuro MR research for GE Healthcare.
Specifically, AI enables data silos that do not communicate with one another to share data easily. “That is helping with the shortage of scan operators and assisting radiologists with reading images. And intelligent devices can make that process seamless,” Banerjee says.
Data Challenges
One of the challenges in making imaging faster, safer, and more robust is parts of the body function in different ways. A heart beats and is in motion. A head does not move as an image is taken of it. As a result, AI algorithms are organ dependent. In other words, the algorithms have to be written to tailor the use of the tool to capture and analyze data that relate specifically to a body part.
“What works with MR will not work in spectral CT imaging,” Majumdar says. “Add to this the fact that different care centers have different generations of machines, and each hospital may have their own way of reporting on their images. Now you can see how complex the equations (relating to changing tools and practices) can become.”
One of the concerns about refining AI algorithms is there is not enough usable data to improve the nuances of the algorithms. The developers who create the software need specific data to fine-tune the algorithm. Fortunately, data generation is one objective of Care Innovation Hub.
“The images of the past may not be totally recoverable or in usable formats to help us improve medical imaging. One of the biggest tasks regarding images is data curation. You want to make sure the data are clean and usable,” Majumdar says.
Right now, AI algorithms can review images that date back to the 1990s. This is the decade when images became digitized, as hospitals began to use PACS to store and retrieve medical images.
“Say, however, that you wanted to integrate data stored in EHRs with images in PACS or DICOM. This is very difficult,” Majumdar says. “It is a long-term goal [but not specifically one of the Care Innovation Hub] to see if we can bring different types of data together.”
Parallel and Prior Programs
UCSF and GE HealthCare hope to make progress through the Care Innovation Hub because many causes of neurodegenerative diseases and other serious diseases like cancer are poorly characterized and understood. “The biological mechanisms can start 20 years before memory loss,” Banerjee says. “Just a decade ago, when a physician detected decline, all they could do was manage the symptoms. There is still no cure for Alzheimer’s. But two to three years ago, we got a ray of hope about disease-modifying therapies for patients in the early and middle stages of Alzheimer’s. That was a big breakthrough.”
Administering such therapies requires a great deal of monitoring with MR images. More accurate images will allow physicians to make more personalized treatment decisions.
GE HealthCare has gained experience in collaborative partnerships outside of its prior work with UCSF, including a program with Cincinnati Children’s. This program began in April 2025 and established the first pediatric Care Innovation Hub in the United States. The Cincinnati Children’s program also involves examining and gathering data from clinical and research work to advance medical imaging. These include MRIs and CT scans.
Another example of GE HealthCare’s work with UCSF is the Head Health Initiative (HHI), a multiyear project launched in 2013 by the NFL. The partner list for the HHI later expanded to Under Armour and the US National Institute of Standards and Technology. It centered on advancing research, diagnosis, and treatment for mild traumatic brain injury and concussion. The HHI furthered GE HealthCare’s development of advanced MRI technologies and high-resolution images, among other achievements.
For GE HealthCare, having a window into a hospital’s day-to-day interaction with patients and its methods of care delivery complements what its engineers work to accomplish. “We develop technology solutions for delivering improved care. Strategic research and development like the UCSF Care Innovation Hub make it possible for us to see firsthand what’s needed,” Banerjee says.
As the UCSF Care Innovation Hub has only been underway for a year and some projects within the program have just begun, GE HealthCare sees it as too early to share specific examples of the data and feedback that can translate into clinical care. “However, all of this is going to be part of what informs our strategy going forward,” Banerjee says.
As an example, GE HealthCare is collaborating with Duan Xu, PhD, a professor in residence at the department of radiology and biomedical imaging and lead of the imaging research for neurodevelopment laboratory at UCSF. Xu’s project is titled “Improved Motion Correction for MRI.” Such work could set a series of goals to keep pace with specific expected hardware developments.
Rombola says the biggest missed opportunity is to create innovation that does not improve health care. “One of our goals in the pillar ‘Imaging Service of the Future’ is to see how we can reimagine the whole imaging service line chain, from ordering an exam to a patient undergoing an imaging scan to reporting results,” she says. “AI is a way to innovate in that area. There’s such a demand for imaging that we need to rethink how we’re meeting that need.”
— Jessica Zimmer is a freelance writer living in northern California. She specializes in covering AI and legal matters.
