21 jun Research into predicting dementia through language and speech
The first symptoms of dementia can sometimes remain hidden from bystanders for a long time. Only when the disease is in a later stage do the symptoms become more visible to the environment. Research is currently being conducted at the University of Groningen into the use of software to diagnose dementia in younger people at an early stage. By analyzing the language use and speaking skills of test subjects, the researchers hope to detect abnormalities that are a possible indication of dementia.
Dementia comes in many different variants and affects not only people of old age, but also relatively young people under the age of 65. Alzheimer’s and frontotemporal dementia (FTD) are two forms that are regularly seen at a younger age. This latter variant is characterized, among other things, by the development of problems with language and speech. PhD student Roelant Ossewaarde therefore focuses in his research on (subtle) differences in language use between dementia patients and subjects without dementia in order to be able to identify the brain disease at an early stage.
“When people have advanced dementia, you notice it immediately. You don’t need any specific knowledge for that. For example, they will use fewer words, or more words with less content. But we are looking for the features in language that occur first,” said Ossewaarde in a statement article from NEMO Kennislink. “To hear that subtlety, you need certain software. We are now developing this, because there is no speech analysis software for this specific group of people for Dutch yet.”
Compare recordings
The patients in the study were diagnosed with a specific variant of frontotemporal dementia called primary progressive aphasia (PPA). The complaints with language and speech develop gradually in this variant and slowly increase over time. To analyze the language use of the test subjects, Ossewaarde conducts various conversations with them, after which the audio recordings of the conversations are examined with the help of software.
“The question is, of course: when is the first moment that you can hear it and you do need the subtlety of the software for that. You only see the patterns when you listen to a very long recording. Only then do the statistical connections between all those linguistic aspects begin to be noticed. But that is very difficult to analyze, because transcribing half an hour of speech in detail is very time consuming. That is why we try to use speech analysis software for this, which we adapt especially for this, because it is actually not yet available for this group.”
The speech of the test subjects is examined, among other things, for word use, how fluently people are spoken and for the complexity of the sentences. “The algorithm can combine all those variables in all kinds of possible ways and then arrives at much stronger predictions,” says the PhD student. “And because we ultimately know the diagnoses of these people, because they participate in a hospital in a clinical process, we can compare the results with what the neurologist says. And there we see that machine learning can predict quite well who should be in the control group and who should be in the patient group. My job as a scientist now is to investigate why the algorithm made that prediction. […] Many people I speak to say it took a long time for my neurologist to recognize that I have this instead of that. It makes their lot more bearable if they know what they have, if the label is right.”
Possible applications
It is also useful for the use of therapies to be able to quickly make the correct diagnosis. Professor of Neurolinguistics Roel Jonkers, promoter of the research, explains: “With therapy you help people with dementia and their loved ones to deal with their problems. The problems are very personal: whether the brain damage is just a bit forward or backwards affects your use of language. There is a flowing and a non-flowing variant. In the fluent variant, patients mainly have problems with language meaning, in the non-fluent variant they have more grammatical problems. The speech therapist can then offer tailor-made therapy.”
“Maybe there will come a time when there will be methods to treat dementia – that it will be limited or maybe even stopped,” he continues. “Even then you want to know as early as possible that it is there. So we don’t have to wait for that drug to come out and then see how we’re going to diagnose, we have to do that at the same time. That is why we are already working on it.”
Lees here the full article from NEMO Kennislink.
–
