AI may soon be able to aid screening for Alzheimer's disease by analyzing what one writes. A team from IBM and Pfizer announced that it had trained artificial intelligence models to detect early signs of Alzheimer's disease, which is known to be difficult to detect, by analyzing language patterns that govern the use of words.
Other researchers have previously trained various models to look for signs of cognitive decline - including signs of Alzheimer's disease - by using various types of data, such as: brain scans and clinical test results. However, what distinguishes the latest study from others is its use of previous information gleaned from the multigenerational "Framingham Heart Study", which has been tracking the health status of more than 14,000 people for three generations since 1948. Researchers say they will be able to predict the incidence of Alzheimer's disease years before Symptoms intensify to the extent that clinicians can observe them with regular diagnostic methods, if the new model's ability to monitor trends that govern these data continues in future studies, which will include larger and more diverse groups. In addition, this screening tool does not require x-rays or invasive tests. The results of the study, funded by the company "Pfizer" and conducted by "IBM," were published on Thursday in the journal Eclinical Medicine.
The new AI models are providing "a nudge to expert medical practitioners on how to spot some subtle changes early on, before a clinical diagnosis can be made," says Ajay Ruyuru, vice president of healthcare research and life sciences at IBM. The form alerts you that there are some changes [indicating] that you need to do a more thorough examination. ”
To train these models, researchers used digital copies of testimonies from the participants in the "Framingham Heart Study," in which they were asked to describe a picture of a woman who appeared to be washing dishes while her two children sneaked into the biscuit bowl without her knowledge. These descriptions did not preserve the handwriting of the original statements, says Roda O, director of neuropsychology at the Framingham study and a professor at Boston University. (Her team was responsible for backing up the data for the new research, but was only involved in that.) Even without the handwriting, the main artificial intelligence model was able to reveal linguistic features that are sometimes linked to early evidence of cognitive decline, according to IBM. These features include specific misspellings, repeated words, and the use of simple expressions instead of syntactic sentences. This evidence is consistent with clinicians' view of how Alzheimer's disease affects language, says Ruyuru.
The main model succeeded in predicting with up to 70% accuracy in participants in the Framingham study who would eventually develop dementia related to Alzheimer's disease before the age of 85. However, the results were based on previous data, and did not actually predict future events, and there are other reservations about the new research.
Artificial intelligence focused on the oldest groups of participants in the Framingham study, who often represent white non-Hispanics. Ou argues that this limits the possibility of generalizing the results to include more diverse societies in the United States and the rest of the world. In addition, it is not yet clear how the AI will perform when studying larger groups. The study data set by the journal e Clinical Medicine included only 40 people who had dementia at the end and another 40 people representing the control group who did not have it, said Gekaterina Novikova, director of machine learning at Winterlight Labs in Toronto. Novikova did not participate in the new study, and wonders whether the performance of artificial intelligence trained by the company «IBM» will change when predicting the emergence of Alzheimer's disease at different points in time before diagnosis.
Nevertheless, Novikova and Roda O / 2 praise the study as a great contribution to the field that may draw more attention to the role of AI in detecting Alzheimer's and thus, attract more resources to the field. “What I personally like about this [study] is that it is one of the sparse endeavors to analyze large-scale, factual data collected over a very long period of time,” says Novikova.
Roda argues that the new models might be more accurate if they could incorporate handwritten font. This ability provides additional clues, such as: evidence of slight twitches, transitions between lettering methods, and minimal letter sizes. "There are a lot of ... traits that [the researchers] did not take into account, which, if put together with linguistic features, will likely increase the model's predictability," explains Ou. In addition, IBM's models did not include oral language data. The ability of artificial intelligence to analyze speech in diagnosing Alzheimer's disease is one of the growing areas of research, and other systems have focused on detecting changes in vocal samples. These samples include clues not found in written samples, such as pauses while speaking.
Whether written or spoken, the language samples provide a source of information that does not require any relatively invasive intervention that enables tracking of people's cognitive health, compared to brain scans and other laboratory tests. In addition, the collection of linguistic data can be done remotely and in an inexpensive way, although this still requires the consent of the people who own the samples with their knowledge of the matter, as well as guarantees to preserve privacy - according to Ruyoro - especially that some people may not want to Learn about their chances of developing Alzheimer's, a disease that is currently incurable.
It may turn out that training models on spoken samples rather than written samples is more practical to reach the largest number of people in the long run, because writing requires literacy as opposed to speech. Novikova and her colleagues at the Winterlight Labs have been focused on teaching AI how to analyze the auditory and linguistic properties of spoken words. Roda used to record speech and handwriting using digital pens that could transmit the handwriting. This is for research purposes. It seems that the company «IBM» adopts these same ideas, and will apply them in future research.
“We are in the process of using this technology to gain a better understanding of diseases, such as schizophrenia [and amyotrophic lateral sclerosis] and Parkinson's disease,” says Guillermo Cecchi, who co-authored the new study and works as a principal researcher in the fields of computational psychiatry and neuroimaging at IBM. We do this in prospective studies that analyze samples of spoken speech taken with consent from similar verbal cognitive tests.