MRI of the human brain (.gif courtesy of iStock). In this collaboration between USC doctors and engineers, the AI algorithm was trained, in part, from tens of thousands of MRI scans from patients worldwide.
There are more than 55 million people globally with Alzheimer’s disease. Fifty-five million that are struggling to remember.
Enter AI, a technology that famously struggles to forget. What if it could remember hundreds of thousands of MRI scans and blood tests? And then, apply that knowledge to how we comprehend, diagnose and treat one of the leading killers in the United States?
That’s exactly the case with research being done in the Keck School of Medicine of USC and the USC Viterbi School of Engineering, led by Paul Thompson, the associate director of the USC Mark and Mary Stevens Neuroimaging and Informatics Institute and a professor in the Keck School of Medicine.
The Promise of AI in Early Detection
Paul Thompson
In 2019, Thompson and his team focused on identifying potential blood-based markers for early Alzheimer’s detection by using machine learning.
This approach revealed that factors such as cardiovascular health, hormone levels and immune response played significant roles in Alzheimer’s development, alongside the well-known amyloid plaques and tau proteins. Today, the research has advanced to using more sophisticated AI techniques like convolutional neural networks, vision transformers and federated learning.
These methods can analyze vast amounts of neuroimaging data, genetic information and other biomarkers to predict disease progression and identify early signs of Alzheimer’s with unprecedented accuracy. According to Thompson, one algorithm learned from reviewing more than 85,721 MRI scans from 50,876 patients, while another learned from poring over the 3 billion letters of the human genome to find signs of Alzheimer’s.
“Deep learning methods allow us to predict clinical decline and discover genomic markers associated with Alzheimer’s,” Thompson said. “These AI models are over 90% accurate in detecting Alzheimer’s from brain scans, a significant improvement from traditional methods.”
The Power of Collaboration: AI for AD
The Artificial Intelligence for Alzheimer’s Disease Consortium (AI for AD), spearheaded by Thompson, is leading this transformative journey. Funded by the National Institutes of Health (NIH) with an $18 million grant, the project spans 12 research sites across the U.S., with USC as the lead site.
“AI for AD is a major project trying to discover ways to treat people with Alzheimer’s disease, discover new medicines and find the best medications that exist today using AI in various exciting ways,” said Thompson, who also holds an appointment at the USC Viterbi School of Engineering.
Federated Learning: A Secure, Collaborative Approach
Data remains the key.
With hypothetical access to every relevant blood test and MRI, Thompson said, “AI would make it easier get the best treatments more quickly and not waste time on ones more likely to fail or have challenging side effects.”
But how to maintain patient privacy?
Jose-Luis Ambite
A key player in this research, Jose-Luis Ambite, a USC Viterbi associate research professor of computer science and a principal scientist at the USC Information Sciences Institute (ISI), has been instrumental in developing the AI technology. Ambite detailed how federated learning, a method allowing multiple sites to collaborate on training AI models without sharing sensitive patient data, is pivotal to their success. This approach not only maintains privacy but also enhances the model’s performance.
“Federated learning enables us to train a neural network on data from multiple hospitals without actually sharing the data,” Ambite said. “This is crucial for medical data due to privacy concerns and the difficulty of sharing such sensitive information.”
Ambite also discussed the challenge of ensuring the security and accuracy of the AI models.“ We encrypt the parameters of the neural network so that even if someone intercepts the data, they can’t access the sensitive information. This method has proven to be both secure and efficient, adding only a minimal overhead to the computational process.”
AI’s Role in Personalized Treatment and Drug Discovery
One of the most promising aspects of the current research is its potential to personalize treatment for Alzheimer’s patients. By integrating various types of data, AI can help determine the most effective treatments for individuals based on their unique genetic and clinical profiles.
“If we can predict how aggressive Alzheimer’s is and which treatments are best, we can personalize care,” Thompson said. “AI can help classify dementia, identify subtypes and stages, and even determine the likelihood of cognitive decline in specific domains.”
As an example, imagine “Mary,” age 62.
Mary shows signs of forgetfulness, is socially withdrawn and no longer expresses interest in activities she previously enjoyed. She gets a range of clinical tests, brain scans and blood tests, and an AI method detects vascular disease and elevated levels of brain amyloid, both of which can be treated if caught early. During the treatment for brain amyloid, an AI method checks Mary’s scans for side effects and shows her brain aging rate has slowed, motivating her to stick with the treatment plan.
Beyond diagnosis and treatment, the consortium’s work also extends to drug discovery. AI can identify new genomic markers that could become targets for future therapies. The consortium is now collaborating with pharmaceutical companies like Biogen to develop innovative imaging techniques that make it easier to track Alzheimer’s progression and test new drugs.
“Biogen funded us to develop an MRI-based test of amyloid plaques using a technique called neural style transfer,” said Thompson. “This AI-driven method could create affordable and widely accessible amyloid scans, making it possible to prescreen patients for clinical trials.”
A Glimpse Into the Future
Looking ahead, both Thompson and Ambite are optimistic about the future of AI in Alzheimer’s research. Ambite highlighted the potential of multimodal AI, which combines various types of medical data, such as brain imaging, genetic information and clinical records, to improve diagnosis and treatment predictions.
“Combining different types of data will enhance the AI’s ability to provide accurate diagnoses and effective treatment plans,” Ambite said. “We are already seeing the benefits of this approach, and it’s only going to get better as we continue to refine our models.”
Thompson echoed this sentiment while emphasizing the importance of making these AI tools both widely available and affordable. As these technologies continue to evolve, they hold the promise of not only improving diagnostics and treatments but also offering hope to millions affected by this debilitating disease.
“Our goal is to ensure that these advancements reach as many patients as possible, regardless of their location or resources,” he said. “We want to see these tools used globally to improve the lives of those affected by Alzheimer’s.”
Published on September 19th, 2024
Last updated on October 1st, 2024
