While there is no cure for Alzheimer’s disease, researchers at the University of California, San Francisco might have developed a way to detect the disease early.
Researchers have programmed a machine learning-algorithm that can detect early-stage Alzheimers about six years before an official diagnoses is made. This gives doctors a chance to detect and intervene with treatment earlier than current practices.
This solution is potentially groundbreaking in its ability for early detection; currently, Alzheimers is often detected too late, when a patient has a more advanced stage, and little can be done to help.
“One of the difficulties with Alzheimer’s disease is that by the time all the clinical symptoms manifest and we can make a definitive diagnosis, too many neurons have died, making it essentially irreversible,” said Jae Ho Sohn, of the UCSF’s department of radiology and biomedical imagery.
Here’s how the algorithm works:
In a recent study, Sohn combined neuroimaging with machine learning to predict whether or not a patient might develop Alzheimer’s at their first presentation of memory impairment, “the best time to intervene.” Specifically, he applied machine learning to positron emission tomography (PET) scans, which measure the levels of glucose in the brain (which helps in the diagnosis of the disease), and by training the algorithm to detect the disease by feeding it images from the Alzheimer’s Disease Neuroimaging Initiative, a dataset of PET scans from patients who were already diagnosed with Alzheimer’s. Over time, the algorithm learned which features are and are not important in predicting the onset of Alzheimer’s.
Once the algorithm was trained, it was tested further. The first test entailed 188 images that came from the Alzheimer’s Disease Neuroimaging Initiative, but had not been presented to the algorithm yet. The algorithm correctly identified 92 percent of patients who developed Alzheimer’s. The second test featured a novel set of scans from 40 patients with possible cognitive impairment; the algorithm identified correctly 98 percent. On top of that, these predictions were made on an average of about six years, before a patient receives an official diagnosis.
The next step for this algorithm is for it to be tested on a larger scale. If it can withstand that, researchers anticipate that it could be used at a memory clinic as a “predictive and diagnostic tool for Alzheimer’s disease, helping to get the patient the treatments they need sooner.”