Tuesday, 3 May 2016

Machine learning applied to autism screening going big time?

Machine learning, when machines, er.. learn, is of growing interest to the autism research field. The names Wall and Duda have filled quite a few posts on this blog (see here and see here for example) on this topic and their suggesting that applying machine learning algorithms to something like autism screening and detection could cut down on time taken and resources used.

As per the publication of the paper by Daniel Bone and colleagues [1] it appears that others working in autism research are also waking up to the idea that this might be a useful area to investigate. So: "In this work, we fastidiously utilize ML [machine learning] to derive autism spectrum disorder (ASD) instrument algorithms in an attempt to improve upon widely used ASD screening and diagnostic tools." Fastidiously is such a lovely word (particularly in the context of science).

The tools in question were the Autism Diagnostic Interview-Revised (ADI-R) and Social Responsiveness Scale (SRS) (both of which have already been machine learning 'applied') and their scores "for 1,264 verbal individuals with ASD [autism spectrum disorder] and 462 verbal individuals with non-ASD developmental or psychiatric disorders, split at age 10." And the results... well, let's just say that the authors were not disappointed - or at least less disappointed than on previous research occasions [2] - as they reported on created algorithms that "were more effective (higher performing) than the current algorithms, were tunable (sensitivity and specificity can be differentially weighted), and were more efficient (achieving near-peak performance with five or fewer codes)." Indeed: "We present a screener algorithm for below (above) age 10 that reached 89.2% (86.7%) sensitivity and 59.0% (53.4%) specificity with only five behavioral codes.Sensitivity and specificity are important concepts when it comes to something like screening instruments in terms of identifying 'all' those with a specific condition and making sure that no 'not-cases' aren't mistakenly identified as 'cases'. The nearly 90% sensitivity rate presented by Bone et al on the basis of 5 behavioural codes is not to be sniffed at.

The addition of one Cathy Lord to the authorship of the Bone paper also adds an air of inevitability that applying machine learning to autism research (and practice) is going to continue and increase. Not only because of her historical connection to the ADI-R [3] (which is a hefty document in anyone's book) but also given her very prominent role in autism research history. Who knows, I might one day be blogging about more big autism research names talking about Wall/Duda things including autism screening triage by YouTube? The final question is: outside of just behavioural variables, who would be brave enough to talk genetics/epigenetics/biology machine learning as the next step in autism screening and/or assessment?

----------

[1] Bone D. et al. Use of machine learning to improve autism screening and diagnostic instruments: effectiveness, efficiency, and multi-instrument fusion. J Child Psychol Psychiatry. 2016 Apr 19.

[2] Bone D. et al. Applying machine learning to facilitate autism diagnostics: pitfalls and promises. J Autism Dev Disord. 2015 May;45(5):1121-36.

[3] Lord C. et al. Autism Diagnostic Interview-Revised: a revised version of a diagnostic interview for caregivers of individuals with possible pervasive developmental disorders. J Autism Dev Disord. 1994 Oct;24(5):659-85.

----------

ResearchBlogging.org Bone D, Bishop S, Black MP, Goodwin MS, Lord C, & Narayanan SS (2016). Use of machine learning to improve autism screening and diagnostic instruments: effectiveness, efficiency, and multi-instrument fusion. Journal of child psychology and psychiatry, and allied disciplines PMID: 27090613