Modeling prosopagnosia using dynamic artificial neural networks

Prosopagnosia is a brain disorder causing the inability to recognize faces. Previous studies have shown that the lesions producing the disorder can occur in diverse areas of the brain. However, the most common region is the “fusiform face area” (FFA). In order to model the basic properties of prosopagnosia two networks have been used concurrently: the Feature Extracting Bidirectional Associative Memory (FEBAM-SOM) and the Bidirectional Associative Memory (BAM). The FEBAM-SOM creates a 2D topological map from correlated inputs through the categorization of various exemplars (faces and various objects). This model has the advantage of using a sparse representation which encompass both localist and distributed encoding. This process simulates the FFA in the brain by exhibiting attractor-like behavior for the categorization of all faces. Once the faces have been learned, the BAM model associates specific faces (and objects) to their corresponding semantic labels. Simulations were performed to study the recall performance in function of the size of the lesions. Results show that the recall performance of the names associated with faces decrease with the size of lesion without affecting the performance of the objects.