What if we are able to cure a disease 6 years before it will occur…
Currently in the USA approximately 5.7 Million people are living with Alzheimer’s disease (AD) and it is a 6th leading cause of death in this country. Based on Alzheimer’s association report number of affected by disease can raise to 14 Million by 2050.
The team from University of California, Berkeley and UCSF led by Benjamin Franc took data about 1,002 people living with AD from the Alzheimer’s Disease Neuroimaging Initiative (ADNI). A deep learning algorithm used given brain images to learn how the brain looks 6 years before the AD activates.
The learning algorithm processed more than 2,100 FDG-PET brain images. On 90 percent of dataset it was trained by researchers to recognise the differences between the healthy brain and the one affected by AD. Then algorithm tested “his knowledge” on the remaining 10 percent. Furthermore the researchers made an independent test in which it was given 40 FDG-PET of 40 people which were unknown to him before.
For those uninitiated FDG-PET is “18-F-fluorodeoxyglucose positron emission tomography”. In practice it is a test which shows metabolic activity. The radioactive glucose compound is put into the blood, then the PET scan shows the uptake of FDG in brain cells. The disability of metabolic activity of a brain is one of the first damages caused by AD.
The results of the research showed that the algorithm had 95% confidence on predicting the final diagnosis also on average it could predict AD 75.8 months before the final diagnosis. The algorithm focused on known areas of the brain but it also did pay attention to the rest of a brain.
In the future researchers would like to teach algorithm to recognise outcomes of other tests using different biomarkers specific to AD. Also the development of early diagnosis can help create preventative drugs.