The path to a world without dementia starts with a brain tissue sample. Researchers at the University of California, Davis are developing AI-driven tools to analyze vast digital archives of brain tissue scans — work that cannot be done at scale by humans alone — to better understand dementia and improve diagnosis and treatment. 

The multi-year initiative, called AggieBrain: AI for Next-Generation Neuropathology, is led by a collaboration between Brittany Dugger, leader of the UC Davis Neuropathology Core and associate professor at UC Davis Health and Chen-Nee Chuah, child family professor in Engineering and co-director of the UC Davis AI Center in Engineering. 

“We hope this research leads to new opportunities for precision medicine for dementia so that people can receive the right treatment at the right time,” said Dugger. “The goal is to make these tools freely available to researchers worldwide, ensuring no scientist is limited by computational resources or dataset constraints.”

The AggieBrain initiative is made possible by a gift of $420,500 from the Susan and Charles Berghoff Foundation with major support from Darrin Mollett and William “Bill” Ballhaus ’89. 

Formed in 2021, the Susan and Charles Berghoff Foundation was inspired by its co-founder, the late Sue Berghoff. She transformed her dementia diagnosis into something positive through advocacy and philanthropy.

Tackling a complex challenge

Dementia is a public health crisis: Over 7 million people in the U.S. are living with the affliction, and by 2050 this number is expected to rise to 15 million. 

One of the current challenges in dementia research is it can only be definitively diagnosed with an autopsy after death. Brain donation is the only means to confirm and type the disease.

Dugger points to cancer as an analogy: Decades ago, cancer was diagnosed and treated as a single disease, but today tumors can be deeply profiled and therapies tailored to specific types of malignancies. 

Similarly, dementia is a broad term for a brain disorder associated with different neurodegenerative diseases such as Alzheimer’s, Lewy body and vascular dementia; frontotemporal degeneration; and mixed-etiology dementia where individuals have more than one dementia type. 

In her work leading the Neuropathology Core at the UC Davis Alzheimer’s Disease Research Center, Dugger studies images of the human brain, analyzing an average of 44 slides per case — a labor-intensive process that involves identifying core pathological features like lesions and segmenting brain regions. However, current methods often miss microscopic details that distinguish different types of dementia. 

For example, Lewy body disease is characterized by abnormal aggregates of alpha synuclein protein (Lewy bodies and Lewy neurites), while Alzheimer’s disease is characterized by aggregates of amyloid-beta and tau proteins (amyloid-beta plaques and neurofibrillary tangles).

Training reliable and accurate models 

Reviewing glass slides containing brain tissues can be both time-consuming and demanding, but machine learning can automate this process to be completed in minutes. 

Chuah’s team is developing AI infrastructure and workflows to identify hallmarks of disease on a wide scale, referencing a large digital image archive with microscopic-level pathology annotations. 

“We are creating a one-stop research workflow, a centralized collection of carefully labeled brain tissue data, that serves as a trusted reference both scientists and AI users can easily access and analyze in one place,” said Chuah.

To ensure the AI models are accurate and reliable, the researchers are creating shared sets of benchmark data and standardized frameworks — a litmus test — to compare models and evaluate them for large-scale performance. 

Incorporating the knowledge of human experts is necessary to assess that models are interpreting and classifying data correctly. This close collaboration between Chuah’s and Dugger’s labs is integral, ensuring students and staff from engineering and medicine are deeply involved in AggieBrain from start to finish. 

Personal experience driving change 

The Susan and Charles Berghoff Foundation has also supported research initiatives and scholarships at Stanford University, San Jose State University and several community colleges.

“Dementia is a growing public health crisis and we’re simply not prepared for it,” said Charles “Chuck” Berghoff, foundation chairman and devoted caregiver for his wife during her journey with dementia. 

College of Engineering alum Ballhaus is the eldest son of Sue Berghoff and a respected aerospace engineer and technology leader. His wife, Darrin Mollett, shares a tragic family legacy: She also lost a parent, her father, to dementia. 

“Convening the knowledge and talent of neuropathology and AI/Machine Learning experts to tackle dementia makes good sense,” said Ballhaus. “Providing an accessible framework and tools for researchers to share deep knowledge and large data sets from brain banks is critical to solving the dementia challenge.”

Dugger and Chuah are also inspired by personal experiences with loved ones affected by dementia.

A path to accelerate and scale dementia research

AggieBrain builds on Chuah and Dugger’s ongoing collaborations that have received funding from the UC Noyce Initiative, the Chan Zuckerberg Initiative and the National Institutes of Health. 

For example, they are collaborating on the UC Davis segment of the Brain Digital Slide Archive (BDSA) — an NIH initiative involving more than 10 U.S. research institutions. Co-led by Dugger, the BDSA aims to provide infrastructure for sharing digital slide images of the human brain across participating universities to facilitate data analysis.

In addition to advancing dementia research, AggieBrain can unlock opportunities to train next-generation AI models to advance understanding of a range of neurodegenerative diseases — laying the groundwork for breakthroughs in diagnosis and treatment.

“AggieBrain's impact is poised to extend well beyond its initial scope. The computational methods at its core, including computer vision, large language vision models and pathology foundation models, translate naturally to medical image analysis challenges in radiology, neuroengineering and other disciplines,” said Chuah.