Although such tests lack accuracy, they do help in early detection or even nudging people to get themselves tested.
But such technological breakthroughs are not only limited to finding solutions to tackle Sars-Cov-2
Although one aspect of technology, during this pandemic, has been to reach more and more people and aid in fast-tracking of vaccines, researchers have also been increasingly relying on technology to detect coronavirus. One possible way has been to analyse cough patterns of symptomatic patients and make assessment faster.
Such tests can be done using the app too. Although such tests lack accuracy, they do help in early detection or even nudging people to get themselves tested. But such technological breakthroughs are not only limited to finding solutions to tackle Sars-Cov-2, Last week, but researchers from IBM and Pfizer also released a study which may help detect the onset of Alzheimer’s and may help delay the onset of the disease.
So, what is the approach adopted by IBM and Pfizer researchers in the study? The study uses language data to detect Alzheimer’s onset. This technique is used by doctors to detect signs of speech impairment in written descriptions of an image. The technique is used to test the cognitive abilities of the patient. IBM and Pfizer, in this case, collaborated to study 703 samples of 270 participants Framingham Heart Study of 1948. As the study was done on people showing symptoms of Alzheimer’s, the model was used to predict if AI could detect the onset and how the results matched with people who actually developed Alzheimer’s.
So, what are the results? IBM used the cookie theft picture (see picture) to determine language samples and found that “cookie-theft picture description task performed better than predictive models” which incorporated Apolipoprotein E, which is linked to Alzheimer’s, demographic variables like age, gender, education and neuropsychological test results.
What was the accuracy of the IBM model? The study used two methods CV and hold-out to run logistic regression and determine the accuracy using area under the curvey approach. The accuracy in the case of a linguistic model was 0.73 and 0.74 under the CV and hold-out approach, much higher than the accuracy of 0.64 and 0.60 for non-linguistic models. Even when both the linguistic and non-linguistic techniques were combined, the predictive power was lower at 0.72 and 0.67. More important, in the case of females, the prediction was more accurate than males. The predictive power in females was 0.84 as compared to 0.63 for males. College graduates were more difficult to predict than participant without college degrees. The predictive power was 0.70 and 0.76.
So, how early can the model detect Alzheimer’s and what are the implications? The study says the meantime to the diagnosis of mild AD was 7.59 years. Although there is no cure for Alzheimer’s, early detection can help delay the onset.