Improved Diagnostic Accuracy in Dependent Personality Disorders: A Comparative Study of Neural Architectures and Hybrid Approaches on Functional Magnetic Resonance Imaging Data

Recently, the advancements in neuroscience studies captured the attention to find the activation pattern and functional connectivity of the brain regions through functional magnetic resonance imaging (fMRI) data. This study explores the cognitive state classification for decision-making and patterns of activation from the prefrontal cortex of human brain. The goal of our study is to help patients in making rewardless-related decisions by performing the respective task in a particular state. Towards this goal, we compared different variants of artificial neural networks (ANNs) to the adaptive neuro-fuzzy inference system (ANFIS) for useful analysis of brain activities. We used fMRI data among two Brodmann areas (BAs) 10 and 47 with total of 1200 voxels to determine the voxels which played a dominant role in decision making. In recognizing two decisions, the sensitivity of judgment in the evaluation was recorded by the score of accuracy; true-positive and false-positive classification rates. Generalized regression neural network (GRNN) surpasses eight network architectures by showing the best results with the least variance and an average accuracy of 90%. Moreover, among ANFIS models, the performance of the Fuzzy C-Mean is overall best with an average accuracy of 95.45% along with subtractive clustering followed closely with an average accuracy of 94%. The results reveal the decisive victory of the ANFIS hybrid intelligence system as the best in classification. The results of ANNs and ANFIS algorithms have significantly confirmed the locations of spatiotemporal activation patterns associated with rewardless decisionmaking in a larger setting and with multiple statistical validation methods.