Friday, October 14, 2016

Developments at MIT: Automated Screening for Childhood Communication Disorders

Children with speech and language disorders, especially under the age of six, often do not  have their disabilities caught early due to lack of identification of the issue from parents and teachers. If the disorders are not caught early in the child’d development, it can lead to academic and social anxiety as the children become older. It is a fact that 60% of kids go undiagnosed until after kindergarten which is an unnecessarily high number. MIT’s researchers at the Computer Science and Artificial Intelligence Laboratory are trying to reduce that percentage by generating a computer system that can automatically screen young children for speech and language disorders. The team of computer scientists have made steady progress but have not  yet completed their work. 

The system works by first analyzing audio recordings of children’s performances on standardized storytelling tests. The scientists plan on making the screening of the children speech completely automated and possibly making it accessible through phones and tablets for low-cost screening for large amounts of children. Two graduate students in electrical engineering and computer science at MIT used machine learning (which you can read about in one of my previous blog posts) to search through large sets of training data for patterns that correspond to particular classifications. The graduate students identified 13 acoustic features of children speech that their machine learning system could search and correlate to a specific disorder. The machine learning was trained on three different tasks: identifying any impairment, identifying language impairments and identifying speech impairments. 

There was an issue with considering age and gender as those both can affect how a child speaks. One of the graduate students used a statistical analysis mechanism called residual analysis to identify correlations between subjects age and gender and the features of their speech. The student then altered the correlations before she fed the data to the machine learning algorithm. This advancement could lead to more children having their speech disorders corrected before it becomes a large negative part of their lives.

References:

2 comments:

  1. Hey Mariah, this is a really cool concept! I had no idea that so many disorders were going undiagnosed in adolescents. This could be a huge contribution to society if, once it was completed, the software was made free and therefore more accessible for families who maybe don't have the opportunity to consult with medical professionals.

    ReplyDelete
  2. the fact that this could one day be available on computers and tablets is huge. one day we might be able to just download an app that will screen for these types of disorders. Its crazy what we can do in medicine with computers and also makes me wonder how the role of normal doctors will change in the future

    ReplyDelete