Analyzing Babies’ Vocalizations May Help in Early Detection of Rett

Machine learning able to 'hear' differences that may ID Rett, fragile X

Marisa Wexler, MS avatar

by Marisa Wexler, MS |

Share this article:

Share article via email
Rett syndrome diagnosis | Rett Syndrome News | announcement illustration of woman with megaphone

Analyzing babies’ vocalizations using machine learning could aid in the earlier diagnosis of Rett syndrome, a small study illustrates.

Researchers say such vocal analysis among infants ages 6 to 11 months may help to identify signs of Rett, or fragile X syndrome, long before a child would typically be diagnosed with these disorders in toddlerhood.

“We … demonstrate that, even if the studied individuals with [fragile X] and [Rett] had not yet shown clinical signs in their second half year of life, the machine ‘hears’ that they had already vocalised differently from TD [typically developing] individuals,” the researchers wrote.

The team noted that these early findings were based on data from a small number of individuals, and stressed a need for further research to confirm and expand this approach.

The study, “Automatic vocalisation-based detection of fragile X syndrome and Rett syndrome,” was published in the journal Scientific Reports.

Recommended Reading
ORCA communications tool

Caregivers Asked to Help in Adapting Tool ‘Critical’ to Therapy Trials

Both Rett and fragile X are developmental disorders whose symptoms usually begin in early childhood. Speech abnormalities are commonly associated with both of these genetic conditions.

In Rett syndrome, as well as in fragile X syndrome, patients often experience delays in obtaining a correct diagnosis. Many of the symptoms are common to a range of disorders, and are not on their own sufficient to diagnose the conditions.

Investigating vocalizations for Rett syndrome diagnosis

Now, a team with scientists in Germany and Austria tested the idea that machine-based analyses of vocal recordings may be useful for the early identification of Rett and fragile X.

The study included three babies with a diagnosis of Rett syndrome, and another three with one of fragile X. The infants ranged in age from 6 to 11 months.

All of the infants with Rett were female, in keeping with the typical sex-based prevalence of the disorder, while all of the babies with fragile X were male. A set of six typically developing babies, matched for age and sex, were included as controls.

Very simply, the analysis involved taking audio data from home recordings of the children, then feeding the data into a computer alongside a set of mathematical rules. The computer would then use those rules to “learn” how to sort the children into prespecified groups.

In an initial set of tests, the researchers evaluated whether these analyses could differentiate children with Rett or fragile X from sex-matched controls, or if they could differentiate between abnormal development (Rett or fragile X) and typical development. In these early analyses, the computer performed with 100% accuracy.

These results “not only demonstrate basic feasibility, but point to the high potential of the approach for future practical application in paediatric healthcare,” the scientists wrote.

Moreover, closer inspection of the data showed that the vocal features that were important for making these differentiations in the computer models were distinct for Rett as compared with fragile X.

“This suggests that early verbal peculiarities of individuals with [fragile X] and individuals with [Rett] acoustically manifest in different ways as compared to typical early verbal behaviour of gender-matched controls,” the researchers wrote.

In further analyses, the investigators tried using the voice-based analysis to sort all of the children into the appropriate group — Rett, fragile X, or control.

“The present study was the very first attempt to combine early vocalisation data of individuals with different late detected genetic disorders within one classification model,” the team noted.

Of the 12 children, nine were correctly classified. One child with fragile X was incorrectly classified as having Rett, while another child with fragile X and one with Rett were incorrectly classified as typically developing.

The researchers outlined some potential changes to the algorithm that might be helpful in improving accuracy. They also stressed the importance of further research with larger datasets to validate and refine the approach.

“Even though our findings indicate that this approach is [worthwhile] to be further followed, they have to be interpreted carefully and can hardly be generalised. This is primarily due to the very small dataset,” the scientists wrote.

The fact that analyses were based on recordings from home videos was also noted as a limitation.

“It has to be kept in mind that the recordings were originally not made with the parents’ intention to collect data for later scientific analyses, but typically to create a memory of family routines and special moments of their children’s infancy,” the researchers wrote, noting that this may lead to an underrepresentation of abnormal behavior that parents elected not to record.

Despite these limitations, the researchers stated that using home video “provides the unique chance to study early development in a natural setting and currently represents the best available approach to objectively investigate prodromal [early] behavioural phenomena in rare late detected developmental disorders such as [fragile X] or [Rett].”