By 2050, 16 million Americans will suffer from Alzheimer's disease. Researchers have come up with an algorithm that can process brain scans to detect early signs of the disease.
Over 5 million Americans currently live with Alzheimer’s, and that number could rise to 16 million by 2050, according to the Alzheimer’s Association. Among older citizens, 1 in 3 eventually die with Alzheimer’s or some form of dementia. Despite the prevalence of the disease, however, current methods of detection are expensive and not accessible to most people.
But recent work in artificial intelligence could help bring detection to the mainstream. While there is still no known cure for Alzheimer’s, early detection could help patients plan for the disease and even stave off some its worst effects through brain strengthening exercises.
At the University of Bari in Italy, a team of researchers has developed an algorithm that uses artificial intelligence to diagnose early signs of Alzheimer’s disease. The team trained the AI with MRI scans of 67 different brains, 38 of which were patients diagnosed with Alzheimer’s and 29 that were healthy. After the training, the algorithm accurately picked out other cases of Alzheimer’s 86% of the time.
The team’s test also included scans of brains with mild cognitive impairment (MCI)… [which] is often linked to the later development of Alzheimer’s disease… The algorithm correctly diagnosed these brains 84% of the time.
The AI was trained by observing changes in how regions of the brain were connected, using two signs that are most often used to diagnose Alzheimer’s disease.
The first is the buildup of plaques of beta-amyloid 42, which are fragments of protein that develop between the synapses of nerve cells. These plaques are sticky, making it difficult for nerve cells to pass electrical and chemical signals to each other to communicate.
The second are neurofibrillary tangles. Cells operate protein-based transport systems to move important materials, like food molecules and cell parts. A healthy transport system is neatly organized into parallel strands of protein conveyor belts — microtubules — stabilized by a protein known as tau. In an unhealthy brain, tau collapses, separating from the microtubules, and attaching to other tau molecules. These tau molecules eventually form threads which twist together to create tangles. The tangles block the transport system and the microtubules fall apart. Without access to those key materials, cells eventually die.
It’s still unclear whether plaques and tangles form because of Alzheimer’s or if they cause the disease. Many people with plaques have no symptoms of cognitive decline or Alzheimer’s disease — but the two phenomenon were the first notable differences that Dr. Alois Alzheimer detected between brains with the disease and brains without the disease.
The team at the University of Bari’s test also included scans of brains with mild cognitive impairment (MCI) that had eventually developed Alzheimer’s. MCI causes a slight but noticeable decline in cognitive abilities like memory, attention, and language and is often linked to the later development of Alzheimer’s disease. Of the 148 scans that were presented to the AI, 48 had MCI. The algorithm correctly diagnosed these brains 84% of the time.
Current methods of detecting Alzheimer’s include cerebrospinal fluid analysis, which measures and analyzes the chemicals that are present in the brain and spinal cord’s surrounding fluid. Physicians conduct spinal taps to sample the fluid and check for levels of tau and beta-amyloid. The challenge with this procedure is that analyses differ between institutions. Because there is no standard procedure of analysis, the same sample has been known to produce different results.
A newer method of detection is neuroimaging with radioactive tracers. In the case of Alzheimer’s disease, the tracer is a radioactive isotope that attaches itself to beta-amyloid in the brain. A positron emission tomography (PET) scan of the brain then picks up on radioactive activity and reveals the presence of plaques. The FDA has also approved two other similar radioactive tracers. Unfortunately, these procedures are expensive.
The AI algorithm promises to be cheaper and potentially more accurate than existing treatment because it can learn from MRI scans and grow more powerful the more information that it’s fed. The tech also doesn’t require expensive hardware or invasive procedures.
The original study “Brain structural connectivity atrophy in Alzheimer’s Disease” was published in September 2017. Full information is available here.
This report was created with data from CB Insights’ emerging technology insights platform, which offers clarity into emerging tech and new business strategies through tools like:
- Earnings Transcripts Search Engine & Analytics to get an information edge on competitors’ and incumbents’ strategies
- Patent Analytics to see where innovation is happening next
- Company Mosaic Scores to evaluate startup health, based on our National Science Foundation-backed algorithm
- Business Relationships to quickly see a company’s competitors, partners, and more
- Market Sizing Tools to visualize market growth and spot the next big opportunity