Study combines neuroimaging with machine learning to predict, with 96% accuracy, whether high-risk 6‑month-old babies will develop autism spectrum disorder (ASD) by age 2

Researchers use brain imag­ing and machine learn­ing to pre­dict which high-risk infants will devel­op autism. Cred­it: Car­oli­na Insti­tute for Devel­op­men­tal Disabilities.

—

A Sin­gle Brain Scan Has Been Used to Accu­rate­ly Pre­dict Autism at Just 6 Months Old (Sci­ence alert)

“Researchers have used brain scans and arti­fi­cial intel­li­gence to spot dif­fer­ences in how key areas of infant brains syn­chro­nise, allow­ing them to accu­rate­ly pre­dict which babies would devel­op autism spec­trum dis­or­der (ASD) as a toddler…The research, led by sci­en­tists from the Uni­ver­si­ty of North Car­oli­na at  Chapel Hill and Wash­ing­ton Uni­ver­si­ty, comes hot on the heels of an ear­li­er study that used two scans tak­en at 6 and 12 months to make a sim­i­lar prediction.

Not only has this new method reduced the num­ber of scans required to make the judge­ment, they were able to pre­dict with more than 96 per­cent accu­ra­cy which 6 month old infants would be diag­nosed with autism by age 2, com­pared to 81 per­cent previously…The researchers used mag­net­ic res­o­nance imag­ing to analyse the neur­al activ­i­ty of 230 regions across the brain in 59 infants who had at least one old­er sib­ling with a diag­no­sis of ASD…

A larg­er sam­ple size and more data will no doubt help deter­mine just how accu­rate this method of diag­no­sis could be in the long term, which is what the researchers are plan­ning next.

It’s also unlike­ly that a sin­gle test will form the basis of any future diag­noses – more like­ly, it would form one piece in a risk pro­file informed by research involv­ing a vari­ety of evaluations.”

The Study

Func­tion­al neu­roimag­ing of high-risk 6‑month-old infants pre­dicts a diag­no­sis of autism at 24 months of age (Sci­ence Trans­la­tion­al Medicine)

• Sum­ma­ry: In a new study, Emer­son et al. show that brain func­tion in infan­cy can be used to accu­rate­ly pre­dict which high-risk infants will lat­er receive an autism diag­no­sis. Using machine learn­ing tech­niques that iden­ti­fy pat­terns in the brain’s func­tion­al con­nec­tions, Emer­son and col­leagues were able to pre­dict with greater than 96% accu­ra­cy whether a 6‑month-old infant would devel­op autism at 24 months of age. These find­ings must be repli­cat­ed, but they rep­re­sent an impor­tant step toward the ear­ly iden­ti­fi­ca­tion of indi­vid­u­als with autism before its char­ac­ter­is­tic symp­toms develop.
• Abstract: Autism spec­trum dis­or­der (ASD) is a neu­rode­vel­op­men­tal dis­or­der char­ac­ter­ized by social deficits and repet­i­tive behav­iors that typ­i­cal­ly emerge by 24 months of age. To devel­op effec­tive ear­ly inter­ven­tions that can poten­tial­ly ame­lio­rate the defin­ing deficits of ASD and improve long-term out­comes, ear­ly detec­tion is essen­tial. Using prospec­tive neu­roimag­ing of 59 6‑month-old infants with a high famil­ial risk for ASD, we show that func­tion­al con­nec­tiv­i­ty mag­net­ic res­o­nance imag­ing cor­rect­ly iden­ti­fied which indi­vid­ual chil­dren would receive a research clin­i­cal best-esti­mate diag­no­sis of ASD at 24 months of age. Func­tion­al brain con­nec­tions were defined in 6‑month-old infants that cor­re­lat­ed with 24-month scores on mea­sures of social behav­ior, lan­guage, motor devel­op­ment, and repet­i­tive behav­ior, which are all fea­tures com­mon to the diag­no­sis of ASD. A ful­ly cross-val­i­dat­ed machine learn­ing algo­rithm applied at age 6 months had a pos­i­tive pre­dic­tive val­ue of 100% [95% con­fi­dence inter­val (CI), 62.9 to 100], cor­rect­ly pre­dict­ing 9 of 11 infants who received a diag­no­sis of ASD at 24 months (sen­si­tiv­i­ty, 81.8%; 95% CI, 47.8 to 96.8). All 48 6‑month-old infants who were not diag­nosed with ASD were cor­rect­ly clas­si­fied [speci­fici­ty, 100% (95% CI, 90.8 to 100); neg­a­tive pre­dic­tive val­ue, 96.0% (95% CI, 85.1 to 99.3)]. These find­ings have clin­i­cal impli­ca­tions for ear­ly risk assess­ment and the fea­si­bil­i­ty of devel­op­ing ear­ly pre­ven­ta­tive inter­ven­tions for ASD.

The Study and Innovation in Context

• Five rea­sons the future of brain enhance­ment is dig­i­tal, per­va­sive and (hope­ful­ly) bright
• Next: Har­ness­ing brain scans to per­son­al­ize autism-relat­ed behav­ioral interventions
• 2017 Sharp­Brains Vir­tu­al Sum­mit: Brain Enhance­ment in the Dig­i­tal Age (Decem­ber 5–7th, 2017)