Being able to diagnose such terrible diseases as Alzheimer’s is very important because treatments and interventions tend to be more successful in the early stages of the disease. However, early diagnosis seems to challenge the doctors themselves. Researchers have now linked the disease process to changes in metabolism, showing that certain changes in the absorption of glucose in certain areas of the brain could be linked to the disease, although these changes are difficult to recognize.
“The differences in brain glucose patterns are very subtle and diffuse,” says study co-author Jae Sohn of the Department of Radiology and Biomedical Imaging at the University of California, San Francisco. “People are good at finding specific biomarkers, but metabolic changes represent much more global and subtle processes,” he said.
The main author of the study, Dr. Benjamin Franc, of the same University, helped by the student Yiming Ding via the idea of the research group of large data in Radiology (Big Data in Radiology – BDRAD), composed of doctors and engineers, They focused on the science of radiological data. Dr. Franc was interested in deep learning, a type of AI where machines learn by the example that they are fed, as human beings do so that this technique could be applied to find changes in the metabolism of animals. people to eventually predict Alzheimer’s disease.
The researchers trained the deep learning algorithm through a technology called FDG-PET: 18-F-fluorodeoxyglucose positron emission tomography, in which the radioactive compound of glucose is injected into the blood. Positronic scanning can measure the absorption of FDG in brain cells, which are an indicator of metabolic activity.
The work was based on data from ADNI – Alzheimer’s Disease Neuroimaging Initiative, a site that focuses on clinical trials to improve prevention and treatment of Alzheimer’s. The ADNI dataset included more than 2100 FDG-PET images from 1002 patients. The researchers then trained the deep learning neural network with 90% of the data set and tested it with the remaining 10% of the data. Through deep learning, the algorithm was able to teach itself the metabolic patterns that correspond to the terrible disease.
Finally, the researchers tested the algorithm in an independent set of 40 images of 40 patients that had never been studied before. The algorithm was 100% sensitive to detect the disease for more than six years before its final diagnosis.
“We are very happy with the performance of the algorithm,” Sohn said. “He was able to predict every case that progressed to Alzheimer’s disease.”
We must indicate that we must be careful because the data set with which we worked is not large enough. However, for Sohn, this algorithm could be useful as a tool that complements the radiological work, especially in conjunction with other imaging tests, thus being able to have an early therapeutic intervention.
“If we can diagnose Alzheimer’s just when all the symptoms have already manifested, it is difficult to intervene,” says Sohn. “On the other hand, if we can detect the disease early, there is an opportunity to treat the disease and even stop it.”