Release date: 2016-06-17
There are no standard clinical tests for Alzheimer's disease (Alzheimer's disease), so doctors usually diagnose the disease by testing their patients for cognitive decline. But for patients with mild cognitive impairment (MCI), this diagnostic strategy is difficult to work with. The early symptoms of clinical dementia are not obvious, and when patients with mild cognitive impairment are predicted to develop Alzheimer's disease (not all diseases develop like this).
So it's no surprise that researchers want to apply deep learning to this challenge. Pamela Greenwood, a professor of psychology at George Mason University in Virginia, said, "Some tests like to tell you: 'This person will get Alzheimer's disease, that person won't.'"
Recently, a group of researchers at Harvard University, Massachusetts General Hospital, and Huazhong University of Science and Technology in China designed a comprehensive functional magnetic resonance imaging (fMRI) brain scan and many clinical data to predict Alzheimer's disease. program. In May, they publicly released the results of this research at the IEEE International Conference on Communications in Kuala Lumpur, Malaysia.
“We want to find it at a very early stage of the disease,†said Professor Li Quanzhen (Li), a researcher at the Clinical Data Management Center at Massachusetts General Hospital. “Many people want to use traditional machine learning methods to do this. Things, but the results are not good, because this problem is really difficult to solve."
After starting the test, they said that pairing the results of their deep learning program with a specific fMRI database data set would improve the accuracy by 20% over the results of traditional database and diagnostic methods. But when these traditional classification methods use this particular data set, their accuracy is similarly improved.
Javier Escudero, a biomedical engineer at the University of Edinburgh, said that the new diagnostics are not as big as the old ones, and may be just the result of better data.
If this is the case, then other experts who want to apply deep learning to the diagnosis of Alzheimer's disease may take a closer look at the data they use for analysis. The latest research: an fMRI scan that shows the relationship between various regions of the brain provides a finer view than previous scans that only recorded various values.
So far, the Harvard-led team is trying to predict how likely MCI patients will develop into Alzheimer's patients by combining fMRI scans and deep learning. fMRI is used because it displays the way the electrical activity of the brain of the subject is connected to different areas of the brain.
The term describing the connection of such different regions is called "functional connectivity", and the pattern of this connection changes after the patient has MCI because the signal transmission depends on the oxygen supply to the neuron, but Proteins accumulated in neurons of Alzheimer's patients almost sever the supply of oxygen, causing atrophy of the brain region.
These researchers want to know if they can predict the onset of Alzheimer's disease from this change in functional connectivity. They began with an analysis of neuroimaging data from 93 MCI patients and 101 general patients. Using time series-based 130 fMRI test data measured from 90 of these participants' brains, these researchers were able to discover how neural signals propagated over a certain period of time.
The next step is quite important, and the researchers processed these data sets to obtain a set of secondary data that describes the strength of the signal connections between these regions in the brain. That is, they created a "feature connection" map to show which areas and signals are most closely related.
Finally, the team established a deep learning program that can be used to explain these “maps†and combine these with clinical data such as age, gender, and genetic risk factors to predict whether a person is at risk for Alzheimer's disease. .
Li Quanzheng said that the results of the research have been almost accurate enough for clinical application. "When the accuracy of this method reaches 90%, it is very useful." She said, "Our accuracy is not that high, but it is already very close."
Even the previous laboratory technique to predict whether MCI patients would become Alzheimer's disease by detecting excess protein capacity in cerebrospinal fluid was only about 65% accurate. This means that some real patients are not diagnosed, while others are indifferent to worry that the disease does not actually worsen.
But after carefully reviewing the design briefs of the Harvard team's research, Dinggang Shen, a cognitive computing scientist working at the University of North Carolina at Chapel Hill, expressed skepticism.
“No one can achieve 80% or 90% accuracy in this area,†he said. “Based on such research, it is impossible to achieve this level†(the authors of the overview acknowledged that they gave Shen’s early drafts There are some typing errors, but insist that this accuracy is correct).
The results show that they only use fMRI data, and the accuracy of using functional connection maps is almost 20% higher than that of researchers in other related fields. However, those traditional prediction methods combine functional connections. The accuracy of the map's data has also increased by about 16%.
According to Escudero, this means that the use of functionally connected maps or interconnected signal strength to determine the probability of Alzheimer's disease is much more accurate than the previous methods of measuring brain signal readings. The biggest improvement is the use of functional connection data."
This latest experiment is part of a recent widespread attempt to apply deep learning or artificial intelligence to help doctors make complex decisions. Perhaps the most famous of these is IBM's Watson trying to help doctors face the mountain of medical records and research literature.
GreenWood of George Mason University put forward an important point that tells us that because there is no cure for Alzheimer's disease, any similar predictive measures for this disease have a limited effect. This technology also requires peer review and very, very much testing to truly achieve clinical application.
Source: Lei Feng Net
We're professional Otology Instruments manufacturers and suppliers in China, specialized in providing high quality medical instruments with reasonable price. We warmly welcome you to buy or wholesale bulk Otology Instruments for sale here and get quotation from our factory.
Ent Forceps,Ent Ear Instruments,Otology Instruments,Forceps Ent Instruments
Tonglu WANHE Medical Instrument Co., Ltd , https://www.tlvanhurhealth.com