Medical Imaging Informatics and AI

Bridging Data Science and Clinical Biomedical Imaging

Medical Imaging Informatics and Artificial Intelligence at UCSF is headed by Dr. Dugyu Tosun-Torgut and brings together world-class researchers from multiple disciplines in order to find new, innovative ways to use artificial intelligence and imaging for medical diagnosis. 

Medical Imaging Informatics and AI

By uniting neurologists, engineers, and data scientists Medical Imaging Informatics and Artificial Intelligence will be extremely impatctful in increasing the scope of our current imaging systems when it comes to the brain. 


The Medical Imaging Informatics and Artificial Intelligence Lab at UCSF aims to foster a truly collaborative environment. All team members are expected to contribute and participate in meaningful ways as we seek to discover novel new ways to utilize technology to better diagnose and treat patients.

We value long term partnerships and create a trusting environment for all to succeed.  We are always on the look-out for great people (B.Eng./Sc., M.Sc., Ph.D., students, interns, post-doctoral fellows, project managers, science writers). Convince us about what makes you stand out from other potential applicants: 

  • Rigorous, innovative, and technologically-savvy
  • Eager to achieve lofty goals
  • Proof of academic success 
  • Expertise in coding
  • Well-versed in one or more aspects of neuroscience and neuroimaging
  • Resilent to hard, brutal, and sometimes soul-crushing nature of academia
  • Stand a great chance of getting funding
  • Have a clear plan for future​

Women and visible minorities are encouraged to apply.

​Should you wish to join our team of exceptional, compassionate, and dedicated scientists and engineers, please feel free to send your information to [email protected].

Currently open positions:

Prospective Partners

At UCSF, MedAI investigators partners with business, industry, private and public organizations, other acacademic institutions, as well as the community to most productively advance the research needed to combat illness worldwide.

Patnerships bring expertise, perpsective and resources to UCSF MedAI whilst also providing partners, such as biotech companies and Artificial Intelligence Agencies, access to intellectual capital and cutting edge research.

Should your agency, organization, or academic cohort wish to work with the MedAI team, please email us at [email protected].



  • Derek Flenniken
  • Erik Ellestad


  • Alison Myoraku
  • Pallavi Rane
  • Clara Sorensen


  • Ryan Ellis
  • Dhaval Kadia


  • Adam Lang

Med AI Research

Lower entorhinal perfusion is associated with higher tau PET independent of amyloid PET in Alzheimer's

Lower entorhinal perfusion is associated with higher tau PET independent of amyloid PET in Alzheimer's

Rubisnski et al., in our 2021 Neurobiol Aging paper, assessed the contribution of markers of small vessel disease (SVD) including white matter hyperintensities and cerebral microbleeds in addition to tau positron emission tomography (PET) and amyloid-PET as predictors of cerebral blood flow (CBF) abnormalities. 

Reduced CBF of the brain tissue is a common pathological alteration in Alzheimer’s disease (AD). Despite CBF alterations being common in AD, their link to primary pathologies including Aβ and tau remains unclear. Previously, a correlation between higher Aβ-PET and lower CBF was demonstrated across the spectrum of sporadic AD. Furthermore, brain autopsy studies report that antemortem CBF reduction in AD is related to higher postmortem Braak stages of tau pathology. Given that Aβ deposition is a strong predictor of tau accumulation, any association between tau and CBF may be related to Aβ deposition. Therefore, the main focus of this study was to test whether local tau-PET is associated with lower CBF in spatially corresponding brain regions, with and without controlling for the contribution of A β. 

Furthermore, SVD is common both in aging and AD. Since increased tau-PET levels are also observed in subjects with SVD, the study also assessed to what extent the associations between tau and CBF are mediated by SVD.


  • Tau-PET is associated with lower CBF in the entorhinal cortex across the AD continuum.
  • The associations between Tau-PET and CBF are independent of Aβ pathology.
  • Amyloid-PET is associated with lower CBF in temporo-parietal regions.

Rubinski A, Tosun D, Franzmeier N, Neitzel J, Frontzkowski L, Weiner M, Ewers M. Lower cerebral perfusion is associated with tau-PET in the entorhinal cortex across the Alzheimer's continuum. Neurobiol Aging. 2021 Jun;102:111-118. Epub 2021 Feb 10 PubMed.

Advanced Neuroimaging Research Improves Quality of Life for People with Parkinson's and Their Families

Translational Value of Neuroimaging for Alzheimer’s Disease

Clinical diagnosis of Alzheimer’s disease is through a comprehensive cognitive testing, assessment of medical and family history, and symptom tracking. The true diagnosis of Alzheimer’s disease is actually done at autopsy.

The classic signs of Alzheimer’s disease are amyloid plaques and neurofibrillary tangles. The amyloid plaques build up outside of the nerve cells in the brain and we think that interaction of the nerve cell with the plaque causes nerve cells to make tangles inside the cell. The tangles cause the cells to degenerate and die. We have evidence that these amyloid plaques start occurring in people’s brains one to two decades before any symptoms like memory loss begin to show.

From a clinical symptoms perspective, especially in the early disease stages, many diseases present clinically very similar to AD, like Frontal Lobe Dementia or Vascular Dementia, but without true AD pathology of amyloid and tangles. Before we could point out the above biomarkers, clinical diagnostic methods were not precise or accurate enough to give someone a solid AD diagnosis. After these biomarkers, we could diagnose someone more definitively with AD by either viewing these plaques with PET imaging or analyzing cerebrospinal fluid from a spinal tap to look for these disease-inducing beta amyloid proteins. It is a technological triumph that we can use these methods, but there is a huge problem with them in clinical practice as well as clinical trials.

First, PET Imaging gives us a fantastic look at the human brain in vivo, but requires exposure to x-ray radiation, is extremely expensive, and requires a health center to possess that technology to begin with. Most medical centers do not actually have access to PET Imaging systems or imaging tracers required to highlight these proteins in human brain, and this diagnostic test is not always covered by insurance. While PET is expensive and exposing individuals to radiation, the less expensive diagnostic method, an analysis of spinal fluid, requires a spinal tap. Spinal taps are painful, can cause uncomfortable side effects and are not always easy to complete on potentially non-compliant patients. Alzheimer's symptoms can cause confusion at the best of times, so needing to be still while under a great deal of pain may be impossible, if not very difficult.

At the moment, most of the AD clinical trials on potential treatments are being run using the participation of individuals who likely have not been confirmed to have biomarkers of AD, for the reasons stated above. As a result, success rate may be artificially low. The treatments, medications or otherwise, might work on AD but not for other forms of dementia, skewing the data to look more like a placebo effect. Researchers aren’t currently able to justify the costs and risks of the more accurate tests, resulting in a barrier to scientifically accurate clinical trials. However, [the VA/UCSF researchers] seem to have found the answer to this conundrum. Utilizing a large data set of potentially AD affected individuals together with healthy individuals and individuals with early AD-like clinical symptoms, we compiled MRI scans and were able to detect a way to recognize the same biomarkers through the much safer, much more financially and geographically accessible MRI technologies that already exist. Not only are MRI’s safer, with no radiation exposure, these brain scanning diagnostic tools are much more comfortable to endure by the AD study participants. This is extremely important, because this allows clinicians and researchers alike to truly diagnose AD etiology in dementia-affected or early symptomatic populations via an additional screening step.

Future clinical studies will be able to afford to utilize the more precise PET imaging and/or spinal tap diagnostic tools if they can limit the number of people they are testing via this initial checkpoint. This will allow for greater diversity and accuracy within clinical trials. Also, because this test is so low-risk to patients, the ability of healthy individuals to be preemptively screened will allow physicians to potentially be proactive and administer preventive medications or life-style modifications to slow the progress of the toxic effects of these proteins  before cell death begins and more importantly before large symptoms arise.

This is incredibly exciting because when the cause of the degeneration is discovered earlier in the progression, there is a much better prognosis for the individuals. Neurodegeneration can be permanent, so it is extremely crucial to not miss an active neuroprotective window where interventions are much more successful. These brain scans also have applications that would not be obvious at first glance. There are certain individuals with these biomarkers who are somehow free of any symptoms and otherwise quite healthy. If we begin to scan more otherwise healthy brains in order to discover these biomarkers before illness becomes apparent, we will also have the ability to find AD-marked yet symptomless individuals. Studying these brains will allow us to potentially find new ways to fight Alzheimer’s, once we know how the brain is being protected in these rare cases.

It is incredibly important that we educate the public about how amazing this breakthrough is. In fact, it could save lives. More public support behind cost-effective, accurate diagnostic scans of the Brain may bring about further funding for researchers working towards AD treatments. Also, public awareness is likely to lead towards a push to a different diagnostic standard, and will hopefully inspire more people to contribute to science by becoming research participants, even if they don’t currently have neurodegenerative symptoms. These discoveries were only possible through the application of AI and machine learning. These research tools require a great deal of data in order to be effective. The Brain, as you may know, is an extremely complicated organ, and there are so many questions that still exist. The more brain scans we have access to, the more we can compile data, and the more data we have, the more exciting discoveries we can make. We hope that with the participation of patients and further collaboration with researchers both in academia and biotechnology that we can uncover much more about the brain.

Medical Imaging Informatics and AI wins 2016 PPMI Data Challenge

2016 PPMI Data Challenge winners with Michael J. Fox

November 29, 2016, The Michael J Fox Foundation named the winners of the 2016 Parkinson's Progression Markers Initiative Data Challenge (PPMI). The foundation asked researchers to provide a model of Parkingson's disesae subtypes or baseline predictors using PPMI data, in order to better accelerate testing of new treatments for Parkinson's. These models would allow researchers to better design clinical trials and choose better patient volunteers in these studies.

The winners were Duygu Tosun-Turgut, PhD, Assistant Professor of Radiology and Biomedical Imaging at UC San Francisco and co-director of the Center for Imaging of Neurodegenerative Diseases at the San Francisco Veterans Affairs Health Care System; and Fei Wang, PhD, assistant professor of health care policy and research at Weill Cornell Medicine. Each received a $25,000 award furnished by MJFF and supported in part by GE Healthcare. Dr. Tosun-Turgut, PhD is the Director and Founder of the Medical Imaging INformatics and AI department at UCSF.

"The Parkinson's Progression Markers Initiative offers a rich pool of open-access data from which to make connections that advance our understanding of Parkinson's disease and impact how we approach drug development," explained Mark Frasier, PhD, MJFF senior vice president of research programs. "The award encouraged scientists from other disciplines to lend their expertise to our efforts to find a cure. Drs. Tosun-Turgut and Wang have provided a strong basis to build on."

Dr. Tosun-Turgut  was able to show that an MRI scan of brain structure and functionality (diffusion tensor imaging) and Unified Parkinson's Disease Rating Scale III (motor examination) total score at baseline were the best factors for determining an accurate Parkinson's diagnosis. "Parkinson's is a highly variable disease, which hinders clinicians' ability to give patients a clear prognosis and researchers' ability to efficiently measure the impact of treatments on the disease process," said Dr. Tosun-Turgut. "Identifying early clinical markers of rate of progression can benefit clinical care and testing of new therapies."

Learn more about this award at the Michael J Fox Foundation and PPMI

Medical Imaging Informatics & AI Faculty

Associate Professor
Assoc Professor in Residence
Associate Professor
Assoc Prof in Residence