Identification markers by machine learning to predict Alzheimer's disease

Nearly 50 million people in the world suffer from Alzheimer's disease or another form of dementia. These irreversible brain disorders slowly cause memory loss and destroy thinking skills, so much so that personal care becomes very difficult or impossible.

Although no treatment currently exists, some medications can delay the progression of symptoms for several years, thus extending the quality of life of patients. However, for these drugs to be effective, the disease must be diagnosed early, before symptoms of cognitive decline become apparent. Current research suggests that brain damage associated with Alzheimer's disease probably begins a decade or more before the onset of symptoms. Reliable screening tools to predict people at risk of developing Alzheimer's disease are urgently needed.

Recently, a team from the Brookhaven National Laboratory of the US Department of Energy (DOE), the Columbia University Medical Center, Stony Brook University, and Ilsan Hospital in South Korea have shown that Combining two modes of magnetic resonance imaging Image-based image analysis and image classification using machine learning models can be a promising approach to accurately predict Alzheimer's risk .

"Such a multimodal imaging analysis can improve predictive power by identifying key diagnostic markers of the disease," said Shinjae Yoo, a staff scientist at Brookhaven Lab's Computational Science Initiative.

Images of the brain reveal changes

Previous MRI studies in which scientists have analyzed the shape and dimensions (morphometry) of the brain have revealed that Alzheimer's disease involves characteristic changes in the anatomy of the brain. For example, thinning of the cortex and decrease of the hippocampus – both brain structures playing an important role in memory – are diagnostic markers of Alzheimer's disease. Although structural MRI is routinely used in the context of clinical evaluation when a person is suspected of having the disease, it lacks precision and generalizability among patient populations for constitute a reliable predictive tool.

Over the last decade, scientists have begun to investigate whether a different MRI mode, called diffusion MRI, could provide additional information to physicians to improve their predictive power. The diffusion MRI captures how water molecules move in biological tissues and the mapping of this diffusion process can reveal subtle changes in tissue microstructure.

Using diffusion MRI, scientists have found abnormalities in the white matter – a type of brain tissue – in patients with Alzheimer's disease. The white matter, located under the cortex, consists of millions of bundled nerve fibers (axons) that connect neurons in different regions of the brain's gray matter, a structured network that functions as bundles of communication cables in the brain. The white color comes from the fatty layer of electrical insulation (myelin) that covers the axons, allowing them to send nerve impulses faster into the brain. The gray matter has relatively few myelinated axons, so it takes on the color of the nerve cell bodies that make it up.

A new research direction

To date, the majority of research on Alzheimer's disease has focused on degeneration of gray matter, but due to recent advances in computational modeling tools, scientists are becoming more and more interested in more to the white matter. For example, improvements to tractography algorithms – a computational method for reconstructing white matter extents from biophysical models of nerve fiber orientations – allow for more accurate estimates of the microstructure of the white matter.

Preliminary research suggests that white matter integrity decreases in people at risk for Alzheimer's disease. At MRI, degeneration appears as bright white patches called hyperintensities. However, scientists do not know to what extent the "structural connectome" of the white matter – the brain wiring system or the unique pattern of connections between the billions of neurons in the brain – contains additional information about the risk of neuromuscular disorders. ; Alzheimer's. on the structural MRI.

Prediction powered by automatic learning

Yoo and his team undertook to answer this question using structural and diffusion MRI images collected from a group of more than 200 elderly patients who visited a dementia clinic at Ilsan Hospital. Neurologists had previously diagnosed in these patients Alzheimer's disease, a mild cognitive impairment (stage between dementia and cognitive decline expected from normal aging) or subjective cognitive decline (considered the first sign of dementia in patients ). or other cognitive functions but normally perform standard screening tests).

To process and analyze the raw images, the team members designed a rigorous pipeline of several existing algorithms. Next, they applied machine learning to form image-derived classification models on brain "phenotypes" resulting from their analysis: estimates of brain shapes and volumes (morphometric data) and structural connectivity of the brain. the white matter (tractographic data) in each patient. . They then used the models to make diagnostic predictions.

"In a study using data from a dementia clinic, we achieved 98% accuracy in the detection of Alzheimer's disease and 84% prediction of mild cognitive impairment, precursors of disease. Alzheimer's, "said Jiook Cha, researcher and assistant professor. of Neurobiology at the Department of Psychiatry of the Medical Center of Columbia University. "The accuracy of our machine learning models on cerebral connectome estimates were 10% and 29% higher, respectively, than imaging markers used in clinical settings (eg, white matter hypersignals). we have reproduced these results. "

By comparing the performance of their different models of machine learning, team members determined that the structural connectome could be a clinically useful imaging marker for Alzheimer's disease.

"The model formed on both morphometric data and connectomic data has more accurately classified Alzheimer's disease and mild cognitive impairment than the model formed only with morphometric data," Yoo explained. "In addition, the connectome model classified mild cognitive impairment and subjective cognitive decline as accurately as the combined model, in contrast to the morphometric model, which did not rank well."

These results suggest that diffusion MRI could be a valuable tool in early detection of Alzheimer's disease. Neuroscientists believe that mild cognitive impairment and subjective cognitive decline are precursors of Alzheimer's disease. Abnormal changes in white matter detected at these preclinical stages could indicate patients at increased risk of developing Alzheimer's disease. The ability to identify such microscopic changes years before more severe macroscopic changes could lead to better treatments and perhaps even healing.

"This study strongly points to the possibility of using multimodal MRI – particularly the structural connectome diffusion MRI – to accurately predict the risk of Alzheimer's," Cha said. .

Follow-up studies based on retrospective patient data will determine if this approach could be implemented in clinical settings.

Explore more:
The imaging shows a rupture of the cerebral connection at the beginning of the Alzheimer's disease

More information:
Yun Wang et al. Diagnosis and prognosis with the help of machine learning, formed on the morphometry of the brain and the connectomes of the white matter, biorxiv (2018). DOI: 10.1101 / 255141, https://www.biorxiv.org/content/early/2018/01/30/255141