The Limits of Diffusion Brain Imaging in Autism Research

What the scan records

Water in living tissue is in constant motion. Inside a bundle of nerve fibres it travels more easily along the bundle than across it, and a diffusion scan measures precisely that: how strongly the movement of water at each point in the brain favours one direction over the others. From thousands of such points, software infers the probable course of the large fibre bundles running between one region and another.

Two numbers summarise the result. Fractional anisotropy is a figure between zero and one describing how sharply the water at a given point prefers a single direction. Mean diffusivity describes how far the water travels in total, whichever way it goes.

Neither figure measures the strength of a connection, and neither can report what changed in the tissue to move it. Myelination, the packing density of axons, the diameter of those axons, the permeability of cell membranes, the multiplication of the cells that manufacture myelin, and the water content of the tissue all influence these numbers. Six properties feed one figure, and the figure cannot distinguish among them.

Thirty-three such studies, covering 831 autistic participants and 836 typically developing controls, were pooled in 2022 by Min Li and colleagues at Osaka University and the Hamamatsu University School of Medicine, writing in Autism Research. The tracts under examination were those associated with language: the arcuate and superior longitudinal fasciculi, which connect temporal and frontal regions, and the uncinate, inferior fronto-occipital and inferior longitudinal fasciculi, which run forward from the temporal and occipital lobes.

Across those tracts the autistic groups showed lower fractional anisotropy and higher mean diffusivity than the controls. Most standardised differences fell between 0.32 and 0.71, on a scale where 0.2 is conventionally small and 0.8 large; the widest was 0.93, for mean diffusivity in the left dorsal pathway. Several comparisons produced nothing at all. Fractional anisotropy in the right inferior fronto-occipital fasciculus differed by 0.08, and in the left inferior longitudinal fasciculus the difference reached only p = 0.064, a trend rather than a result.

Dividing the studies by age altered the picture. Reduced fractional anisotropy appeared in every language tract among the studies of children, and in none of the studies of adults, whose mean ages ran as high as 32.9 years. Elevated mean diffusivity persisted in both.

What it means for you and your child

Nothing in the pooled data connected these tract measurements to how well any participant used language. Verbal IQ was tested as a moderator across the tracts and moved none of the results. The authors could not combine the individual studies’ own correlations between diffusion measures and language performance either, because those studies had used different tests and reported different statistics. A meta-analysis assembled around language networks finished without evidence that the condition of the tracts tracked language.

That absence governs how the finding may be used. Differences of this magnitude describe distributions that overlap heavily, and many autistic participants will have had higher fractional anisotropy than many controls. No threshold divides the groups, no clinic offers the scan as a test, and no measurement here accounts for why a particular four-year-old speaks in single words.

The participants were also a particular sample, and the studies disagreed with one another to a degree the authors flagged, between 45 and 83 percent of the variance in most of the pooled comparisons. Thirteen enrolled only males. Thirty-one of the 33 examined autistic people with an IQ above 70, which leaves the children whose language difficulties are most severe largely absent from the evidence gathered to describe language.

The studies of children found reduced fractional anisotropy in every language tract, and the studies of adults found it in none. Those were different people, scanned once each, in laboratories working separately across two decades. The comparison is consistent with diffusion measures that approach typical values as the brain matures, and it establishes nothing about the course any single person’s took.

The authors read the age difference as abnormalities that may ease with development, and they state plainly that the longitudinal studies needed to check that reading, following the same children forward, have not been done in sufficient number. In typical development fractional anisotropy peaks somewhere between adolescence and young adulthood, and mean diffusivity reaches its lowest point three to six years after that. Whatever is happening with your child’s language, this literature sets no deadline on it.