A New Machine Learning Algorithm for Alzheimer's Diagnosis

Researchers at Case Western Reserve University have developed a machine learning method of early diagnosis for Alzheimer’s, which they believe may be superior to previous methods.

Researchers at Case Western Reserve University have developed a machine learning method of early diagnosis for Alzheimer’s disease, which they believe may be superior to previous methods.

“No single marker has been proven to accurately categorize patients into their respective diagnostic groups,” the study authors wrote. Tests exist for certain Alzheimer’s biomarkers, and diagnosis can be done based on symptom onset, but the authors wrote that “Most do not simultaneously consider all diagnostic categories and provide suboptimal fused representations using the same set of modalities for prediction of all classes.”

In response, the team created what it calls a Cascaded Multi-view Canonical Correlation (CaMCCo) algorithm for diagnosis. Integrating a range of Alzheimer’s indicators, like mild cognitive impairment, hippocampal MRIs, glucose metabolism rates in the brain, and biomarkers.

"The algorithm assumes each parameter provides a different view of the disease, as if each were a different set of colored spectacles," Anant Madabhushi, one of the authors, said in a related press release. That would be the “multi-view” part of the name. The algorithm then assesses those variables in a cascade, separating the healthiest brains from the least healthy and then attempting to distinguish between those with mild cognitive impairment and those with verifiable Alzheimer’s.

The researchers tested the method on 149 patients. For determining healthy controls from impaired subjects, the CaMCCo algorithm beat out all other methods tested, whether single tests or modalities. For determining mild cognitive from Alzheimer’s, it was on pair with the other top-performing methods.

The researchers acknowledged a number of limitations in the work, which is early. “We only combine the modalities that independently provide the best accuracies on the training set, which may not be complementary. Nonetheless, we found that considering a subset of modalities provides improved performance over fusing all modalities,” they wrote. Another possible weakness is the potential for the propagation of error through the cascade, though that itself something that must be adjusted for in any cascade.

Despite certain limitations, the authors consider the work “a promising platform for fusion of multiscale, multimodal data for early diagnosis of Alzheimer’s Disease.” Preliminary as it may be, it is just one of many studies attempting to use machine learning to augment, or potentially improve, diagnostic accuracy for difficult medical conditions.

The study was published online today in Scientific Reports.