It is difficult to discern the clinical use of neuroimaging techniques in predicting outcome in these studies for the following reasons. First, it is unclear whether neuroimaging can predict outcome significantly better than other existing measures, such as behavioral testing (currently the most straightforward and inexpensive way of predicting reading outcome). In order to make the results clinically useful, it is important that neuroimaging techniques exceed currently available methods or have an add-on effect. Second, no validation or reliability analyses have been performed (except by Apostolova and colleagues,33 who used permutation tests). Finally, prospective analyses have not been performed; rather, all studies thus far have relied on retrospective correlational analyses of behavioral outcome and initial neuroim-aging measures.
However, one intriguing study on predicting memory decline in patients with postoperative temporal lobe epilepsy (TLE) has overcome many of these limitations.34 The authors examined preoperative behavior, brain volume, and functional MRI (fMRI) in a small sample of 10 patients with TLE to predict postoperative memory. They found that left-right hippocampal encoding activity difference showed reasonable sensitivity, specificity, and positive predictive value (20% to 100%) for predicting the amount of preoperative to postoperative memory decline.
Building upon these earlier studies, we addressed many of the limitations by comparing neuroimaging with existing assessment methods, performing validation analyses, and performing prospective analyses in a large sample of subjects. Our primary goals were to test whether neuroimaging can be used to predict reading success and to test whether we can achieve greater predictability by integrating neuroimaging measures into the currently available method of behavioral measures to predict reading success. We considered prediction to be an important goal because improved prediction of reading skills can facilitate identification of children who may benefit most from intensified or alternative reading instruction so that reading failure is minimized or even prevented.
Our present study focused primarily on one reading skill as an outcome measure that is thought to be essential for effective reading: word decoding. Decoding refers to the ability to determine the sound of a word from letters and syllables. Decoding ability is fundamental to reading because learning to read involves learning to relate the sounds of known auditory language (phonology) to letters (orthography). It is known that early and systemat- ic emphasis on decoding leads to superior reading skills,35,36 that decoding accounts for most of the variance in reading comprehension,37-39 and that the development of language-specific phonology is essential for reading success.40 Therefore, better methods for early identification of young children at risk for impaired decoding abilities hold promise for improving the specificity and effectiveness of early intervention and later achievement of reading skills.
A relatively pure test of word decoding involves reading pronounceable nonsense words aloud, because their proper pronunciation can only be derived from decoding skills (as opposed to words memorized by sight and context). Such a test also measures phonemic awareness, which is the awareness that separable sounds (phonemes) are blended to produce words. Phonemic awareness has been found to be one of the best predictors of reading success.41-45
Findings from a retrospective study
In our first study, we examined 53 normal readers aged 8 through 11 years, to find how behavioral and brain measures taken in the autumn of an academic year (Time 1) predicted decoding skills in the spring of the same academic year (Time 2) by measuring performance on the Woodcock Reading Mastery Test's (WRMT) Word Attack (WA) subtest.46 WA requires the child to attempt to read pronounceable nonwords of successive difficulty aloud. Initial assessment was performed using a standard set of reading and other behavioral tests, fMRI during a real-word rhyme judgment task (a standard fMRI task that taps into phonemic awareness, as does WA), and VBM measures of gray- and white-matter densities.
Using whole-brain regression analyses, we identified regions that showed significant correlation with brain activation (P < .001, uncorrected) or gray- or white-matter volume (P < .001, family-wise-error corrected) at Time 1 that correlated with Time 2 WA standard scores (WA-ss) adjusted for age. We then extracted contrast estimates or volume measures from these regions and submitted them to multiple regression analysis.
We found that specific patterns of brain activation during phonologic processing and gray- and white-matter morphology in regions critical for normal reading were correlated with later reading skills (Figure 1A) and that the neuroimaging model explained 66% of the variance (Figure 1B). We also identified behavioral tests at Time 1 that correlated with Time 2 WA-ss and performed similar multiple regression analyses. In contrast, the behavioral model, which consisted primarily of measures related to phonemic awareness, explained only 49% of the variance (Figure 1C).