MRI based schizophrenia patient classification: A meta-cognitive approach

In this paper we investigate the problem of automatic classification of structural MRI images, to distinguish between schizophrenia patients and healthy controls. Our methodology involves usage of a meta-cognitive neural network architecture that addresses classification issues inspired by learning strategies of cognition in the human brain. Due to heterogeneity in schizophrenia patient population it is important to consider the novelty of the training samples. The high dimensionality of the dataset requires the algorithm to be computationally efficient and efficient use of past knowledge for generalisation. The above problems are tackled by the Projection Based Learning (PBL) Meta-cognitive Radial Basis Function (McRBF). We demonstrate the problem of schizophrenia imaging data classification with 40 patients and 40 control images that have been age and gender matched. For comparison the performance of PBL-McRBF has been compared with a quadratic kernel support vector machine. This paper demonstrates the performance of meta-cognitive classification algorithms to tackle ill posed problems in high dimensional neuro-imaging data sets.