Predicting Cognitive Decay in Alzheimer’s Disease

CAMBRIDGE, MA—Over five million Americans are living with Alzheimer’s Disease, according to the Alzheimer’s Association, and as many as 16 million will have the disease in 2050. Diagnosis of the disease before symptoms appear has the potential to maximize treatment effects and significantly improve long-term prognosis. However, the variability in how Alzheimer’s presents and progresses over time can significantly complicate both identification of the disease and selection of appropriate treatments. Although techniques for computer-aided diagnosis of Alzheimer’s have been developed in recent years, these methods provide little insight into how the disease will manifest itself in a given patient.

To address these challenges, Draper scientists have developed a deep learning technique that can predict how specific aspects of an Alzheimer’s patient’s cognitive state will evolve in light of the patient’s current brain structure. This technique uses a novel Convolutional Neural Network (CNN) architecture and leverages the burgeoning quantity of structural MRI and cognitive score data collected by the Alzheimer’s Disease Neuroimaging Initiative to predict a patient’s cognitive exam subscores up to three years into the future given a current structural MRI scan.

“Our goal is to accurately predict specific aspects of an Alzheimer’s patient’s cognitive state such as learning, memory, and language production and comprehension,” said Lev E. Givon, a senior scientist at Draper’s Applied Machine Learning Group who co-developed the technology. “By doing so, our technique can empower clinicians to personalize treatment with respect to the patient’s future cognitive trajectory.” 

Givon noted that Draper’s approach provides details of a patient’s cognitive prognosis that might not be revealed by existing methods that can only predict a patient’s aggregate cognitive exam score. Givon will present the technology at the 2017 Conference on Machine Intelligence in Medical Imaging and the 2017 BioImage Informatics Conference this fall.

According to David O’Dowd, Draper’s associate director of biomedical solutions, the technology for cognitive state prediction has broad clinical implications beyond personalized treatment of Alzheimer’s Disease. “Our technology’s deep learning architecture can be adapted to predict specific aspects of disease progression in other ailments such as Parkinson’s Disease, cancer, and depression. With the ongoing growth in collected biomedical data, the same algorithm can be applied to many other problems within the biomedical field.”

Prediction of a patient’s future cognitive state from structural brain data is among the most recent innovations in Draper’s biomedical solutions portfolio. Draper has developed machine-learning diagnostics for identifying PTSD and mTBI faster and more accurately than ever before. Other biomedical tools developed at Draper include injectable brain implants to treat neurological and psychiatric disorders through targeted electric stimulation and organ-on-a-chip systems that replicate human biology to speed up the drug discovery process.