< Back to previous page

Project

Distributed processing of concrete entities in temporal neocortex.

The focus of my doctoral research project is to elucidate how concrete entities are being processed in the temporal lobe. Univariate analysis techniques on functional magnetic resonance imaging (fMRI) data have beenthe gold standard in neuroimaging analysis for over a decade. These studies have provided a lot of valuable insights on the functioning of the temporal cortex in humans. For instance, the Lateral Occipital Complex (LOC), overlapping with the posterior fusiform gyrus, has been identifiedas a region that responds to the shape of objects. More anteriorly, evidence from patients with semantic dementia suggests that lesions of the anterior temporal pole result in a pan-modal semantic impairment. 
Using univariate analysis, it has not been possible until now to demonstrate how the representations of concrete entities are organized within the temporal lobe. In recent years, multi-voxel pattern analysis (MVPA) has emerged as a sensitive approach that allows for a better characterization of the language network. Studies using MVPA have demonstrated that information about different stimuli (houses, faces, artifacts, animals, ...) can be derived from the fMRI response patterns in the inferotemporal cortex. However, in contrast to univariate studies, most MVPA studies did not match for perceptual and lexical differences between picture stimuli. When effects are found for word stimuli, perceptual confounds are no longer a concern because the meaning and orthography of a word areunrelated. Representational similarity analysis (RSA) is a recent addition to MVPA. RSA can be used to compare relationships between entities derived from fMRI response patterns to the relationships between the sameentities calculated by other means, e.g. behavioral judgments.
During my doctoral research project, I aimed to integrate prior knowledge about patient lesion evidence and neurophysiology with fMRI data on healthycontrols using MVPA. To this end, I developed a new MVPA approach basedon the cosine similarity measure. This approach first calculates the average cosine similarity between all fMRI response patterns of trials belonging to a condition of interest (e.g. trials representing a same entity). Second, this value is compared to 10,000 random permutation labellings. This means that the labels of the trials are randomly reassigned andthe cosine similarity values are recalculated with this new label. Finally, the significance of the result is determined by comparing the result using the true labels to the distribution of all random permutation labellings. If the cosine similarity values are increased above chance, the fMRI response patterns contain information about the condition of interest. I also implemented an RSA using the cosine similarity measure. Forthis RSA, cosine similarity is calculated between the similarity between all entities based on the fMRI data and the similarity between all entities based on behavioral data. Once more, the significance of the results was determined by random permutation labelling.
Two main MVPA studies were undertaken during this doctoral research project. A first project (Chapter 3) focused on right mid-posterior fusiform gyrus, which has been implicated in shape processing. Prior work reported a failure of the structural description system in patient JA, who suffered a stroke in right mid-posterior fusiform gyrus. This patient could not distuingish real from chimera objects and could not produce as many visual features from memory as a control group. The lesion delineated on the scan of thispatient was used as a volume of interest in 46 healthy young controls in the MVPA study. I used 3 well-matched semantic clusters of artifacts in this project (musical instruments, vehicles and tools). The participants either had to respond if the stimulus was a word or picture or had towatch the stimuli passively. When stimuli were presented foveally, I found a significant increase in cosine similarity if the same object was shown, or if pictures belonging to the same semantic cluster were shown. However, in 3 control experiments, I manipulated the size, location, color, orientation of the presentation and used different exemplars of the same picture stimuli. As soon as the location was manipulated, the apparent effect of semantic similarity disappeared. I found that object identity was reflected in the fMRI response patterns in right mid-posterior fusiform gyrus, even if location, size, color, orientation or exemplar were varied. These findings fit with the behavioral data of patient JA andconfirm that right mid-posterior cortex is a part of the structural description system.
In a second MVPA project (Chapter 4), I investigatedthe role of left perirhinal cortex. In this study, a well-matched sample of 24 animals was presented either as a word or a picture. A total of 35 subjects performed a feature verification task during the fMRI scan. For every animal, they had to indicate whether 8 features were applicable or not. By means of univariate analysis, I contrasted the activation caused by this task with a low-level control task in which the participants had to make a word-picture decision. This contrast showed that activation was increased in the left occipitotemporal cortex during the feature verification task. I examined different regions within this volume of interest. Representational similarity analysis on left perirhinal cortexdemonstrated that the relationships between entities derived from behavioral data were reflected in the relationships between the fMRI responsepatterns to words. These findings are in accordance with the growing evidence that left perirhinal cortex plays an important role in binding semantic information. The reflection of semantic similarity within left perirhinal cortex can be interpreted as supporting evidence for the "Similarity In Topography" hypothesis, which postulates that the neurons in this area are organized according to the similarity of the concepts their combined activation patterns represent. 
Evidence from patient lesion studies remains crucial in the study of the semantic processing pathway. During my PhD project, I performed extensive neuropsychological testing in 7 patients that suffered a lesion in the right occipitotemporal cortex (Chapter 5). The patients participated in a self-paced picture naming task as well as a tachistoscopic picture and word identification task. Furthermore, they took part in a feature generation task, in whichthey had to generate as much as possible different features for each of55 entities, coming from very different categories (biological and nonbiological). The same tests were administered to 16 age-matched controls.Until now, category-specific impairment following lesions of the right occipitotemporal lobe were considered rare, but only very few consecutive series have been reported. Three of the 7 patients tested, displayed category-specific impairment. Two patients with large lesions suffered from an impairment for naming biological entities, which is linked to the higher processing demands required by biological entities. A further case, with a small lesion confined to right medial fusiform gyrus, showed disproportionate naming impairment for nonbiological versus biological entities, specifically for tools. This tool-specific deficit following a right medial fusiform lesion conforms to what one would predict based on functional imaging in the intact brain. Two of the stroke patients, who suffered very small lesions, demonstrated a deficit of visual feature generation for animate entities.
To conclude, during my doctoral thesisI have developed a new approach for MVPA which has proven its worth in dissecting the functional anatomy of object and language processing to agreater detail. Using MVPA, I have provided additional evidence that semantic processing is a gradual process which takes place in an anteroposterior gradient within the occipitotemporal cortex. The first stages of processing are heavily depended on the input modality the information isbeing presented in, e.g. visually, auditory or by means of written words. Gradually, this input is transformed into a neural code which allows interaction with the elaborate semantic-associative network. Future challenges lie in the implementation of this technique for patient work. 
Date:1 Oct 2009 →  6 Feb 2014
Keywords:Temporal neocortex
Disciplines:Neurosciences, Biological and physiological psychology, Cognitive science and intelligent systems, Developmental psychology and ageing
Project type:PhD project