Abstract
Introduction
Neuropsychological researchers have long been interested in relating the location of a patient's lesion to her cognitive impairment (e.g. Broca, 1861). Previous studies have primarily relied on small patient numbers, manual procedures for symptom-lesion mapping and binary data (Davidoff and De, 1994; DeLeon et al., 2007; Luzzatti et al., 2006; Tranel et al., 1997). Furthermore, the above studies did not control for potential confounds due to covarying factors, such as levels of attention; all of these factors may have contributed to the inconsistencies in the current literature. Here, we combined robust, automated lesion analysis procedures and standard CT imaging to investigate the neural networks involved in object naming in a clinically-relevant setting. Extensive prior research suggests that structures within the left ventral visual stream, anterior temporal and frontal cortex play a key role in mediating object naming (Humphreys et al., 1999; Martin, 2007; Price et al., 1996).
Methods
We tested 80 acute stroke patients using the object naming task from the Birmingham University Cognitive Screening (BUCS(Humphrey et al., 2007)). All patients underwent a CT scan as part of their clinical evaluation. Data were analysed using SPM5 SPM8b. The CT images were normalized to a CT template, the skull was removed. Subsequently the images were segmented and normalized again; finally the segmented grey matter (GM) images were smoothed using a 12mm Gaussian kernel (Fig. 1). The GM images were entered into a regression analysis that modeled subjects’ performance on the picture naming task; age, gender, mini-mental state and depression were included as covariates of no interest. (1) Using mass-univariate voxel-based morphometry (VBM), we tested at every single voxel for reduction in GM volume that was positively predicted by subjects’ picture naming performancen. (2) Combining Multivariate Bayesian Decoding (MVB(Friston et al., 2008) with smooth priors) and Bayesian model comparison, we compared the ability of GM voxels within (i) the entire brain, (ii) the left occipito-temporal, and (iii) left frontal cortices to predict subject's naming performance.
Results
On average patients named 9.1/14 ± 3.64std pictures correctly. When compared to age matched healthy controls, 42 patients were classified as impaired (<10, Fig. 2). VBM revealed that impaired picture naming predicted reduced GM in the left ventral visual stream, the anterior temporal pole and (bi-laterally) in the middle frontal gyrus (P <0.001, Fig 3). MVB and Bayesian model comparison showed that the left frontal gyrus did not predict behaviour better than the null model (log evidence (LG) <3). In contrast, the left occipito-temporal cortices were significantly better predictors than the null model (LG > 5, Fig 4).
Conclusions
In line with previous research we showed that lesions to the left ventral visual stream and frontal cortices impaired the ability to name objects. Using MVB we found that lesions within the left occipito-temporal cortices were better predictors of behaviour than lesions to the left frontal. We conclude that CT data of large number of patients can provide useful information for function-lesion mapping, here used to reveal the neural circuits involved in picture naming.
