Alzheimer’s disease is notoriously difficult to diagnose. Its symptoms are very similar to other conditions like Parkinson’s disease and vascular dementia (and may even occur simultaneously in some patients), making the task of diagnosing a patient’s condition challenging for physicians. An even more difficult task is predicting whether an individual with mild cognitive impairment will later progress to Alzheimer’s disease. Even advanced techniques have poor predictive power. Positron emission tomography (PET), a type of brain scan that measures the energy consumption of different brain regions, has only a 57.2% rate of accuracy.
To address this dilemma, a team of Canadian and South Korean scientists tested five different computer algorithms for their accuracy in diagnosing or predicting Alzheimer’s disease. Their results were published this week in Scientific Reports.
The researchers used a database of PET scans from the Alzheimer’s Disease Neuroimaging Initiative to train the computer models. Their study included 94 patients with Alzheimer’s disease and 111 age-matched healthy patients. They found that one model, called the Support Vector Machine with Iterative Single Data Algorithm (SVM-ISDA), could distinguish the Alzheimer’s patients from healthy controls with 80% accuracy.
The researchers then tested the performance of the computer models on three different PET scan databases. This time, rather than distinguishing Alzheimer’s disease from healthy patients, they wanted to see whether the models could predict whether individuals with mild cognitive impairment would develop Alzheimer’s disease within the next 3 years. Here the SVM-ISDA once again came out on top, though its predictive power was lower than for the previous task. The model predicted which patients would develop Alzheimer’s disease with an overall accuracy of around 51-59%. The other algorithms all had less than 50% accuracy.
They next wanted to see whether the computer models could distinguish Alzheimer’s disease from two other types of dementia: Lewy body disease and Parkinson’s disease. These conditions are both frequency misdiagnosed as Alzheimer’s disease. They found that in patients who had Parkinson’s disease but had not yet developed dementia, the computer models could distinguish between Alzheimer’s and Parkinson’s. However, for patients with Lewy body disease or Parkinson’s disease dementia, the models could not distinguish them from Alzheimer’s disease.
The conclusion for this study was that the SVM-ISDA is the most accurate computer model for diagnosing and predicting dementia based on PET scans. However, while the model performed fairly well in diagnosis, its ability to predict Alzheimer’s disease in patients with mild cognitive impairment was barely more than 50%, and it couldn’t distinguish Alzheimer’s from other forms of dementia.
This highlights how much research is still needed to be able to predict patients’ prognosis. Earlier diagnosis could mean earlier administration of treatments, which might make them more effective in slowing or preventing the onset of Alzheimer’s disease. It would also allow patients and their families more time to plan for the future.