Tag Archives: algorithm

Artificial Intelligence Could Help Us Predict Alzheimer’s Disease

Many experts agree that preventing the progression of Alzheimer’s disease is much more effective than trying to reverse it once the damage is done. This makes early diagnosis critical. Unfortunately, most Alzheimer’s patients are not diagnosed until relatively late in disease progression, when toxic amyloid plaques have already accumulated in their brains to potentially irreversable levels. However, this might soon be changing with the recent surge in artificial intelligence technology.

figure1.jpg

This diagram shows the stages of Alzheimer’s disease, which can begin up to 20 years before diagnosis. Most patients are not diagnosed until the mild or moderate stage, since this is when cognitive impairments become more noticeable. Image Source

This research was described in a paper published in Neurobiology of Aging by scientists from McGill University in Canada. Their goal was to create an algorithm to predict whether people with mild cognitive impairment would progress to dementia. They utilized a noninvasive technology called positron emission tomography (PET). PET involves the patient lying inside a donut-shaped machine, similar to a CAT scanner. The machine can measure which areas of the brain have higher or lower activity levels based on how much glucose each brain region is consuming.

20_a

A patient lying inside of a PET machine. Image Source

The researchers used PET scans from 273 patients with mild cognitive impairment. 43 of these patients were diagnosed with probable Alzheimer’s disease at a follow-up appointment two years later. Then the scientists trained an artificial intelligence algorithm to predict which patients would develop Alzheimer’s based on their PET scans.

They used the data to generate the map of the brain that’s shown below. The red-colored areas indicate a higher odds ratio (OR). This means that unusual activity levels in those brain areas are associated with an increased risk of Alzheimer’s disease. For example, an odds ratio of 3 means that a person with unusual activity levels in that brain area is 3 times as likely to develop Alzheimer’s compared to someone with normal activity levels.

gr2.jpg

Figure 2 of the paper shows which brain regions are the most important for predicting the risk of Alzheimer’s disease.

The algorithm was able to predict which patients would progress to Alzheimer’s disease with an accuracy of 84%. This is better than any previously-developed PET algorithms, and comparable to more invasive diagnostic techniques such as spinal taps. This is exciting news because it suggests that a painless, noninvasive technology can be used to predict Alzheimer’s disease with a fairly high degree of accuracy.

As always, we have to point out a few problems with this study. For one thing, it’s impossible to know for sure whether the patients’ Alzheimer’s disease diagnoses performed by doctors at the follow-up appointment were actually correct. This is because many forms of dementia have similar cognitive symptoms, and can be easily confused during diagnosis. Parkinson’s disease, vascular dementia, or even a urinary tract infection can be misdiagnosed as Alzheimer’s disease. (For more info see Is it really Alzheimer’s? 10 common misdiagnoses you should know about). Only a postmortem brain analysis can reveal for sure whether the patients truly had Alzheimer’s. This muddles our ability to judge how accurate the algorithm really was.

Another problem is that PET scans can be quite expensive, costing upwards of $7,000. If a patient is incorrectly diagnosed with Alzheimer’s, this could lead to futher costs for medication to treat a disease they don’t actually have. Finally, while the mild cognitive impairment stage is earlier than most Alzheimer’s patients are diagnosed, it can still be up to ten years after the true beginning of the disease. We still have no reliable way to make diagnoses that early. Nonetheless, this study is at least a step in the right direction. With future advances in artificial intelligence, we might be able to improve our diagnostic accuracy at earlier stages of the disease.

 

Enjoy this post? Help it to grow by sharing on social media!
Want more? Follow AlzScience via email, Facebook, or Twitter!
Advertisements