Suratgar, A. A. | bioRxiv Science

bioRxiv2023

Design of a Computational Intelligence System for Detection of Multiple Sclerosis with Visual Evoked Potentials

In this study, a new approach for modification of membership functions of a fuzzy inference system (FIS) is demonstrated, in order to serve as a pattern recognition tool for classification of patients diagnosed with multiple sclerosis (MS) from healthy controls (HC) using their visually evoked potential (VEP) recordings. The new approach utilizes Krill Herd (KH) optimization algorithm to modify parameters associated with membership functions of both inputs and outputs of an initial Sugeno-type FIS, while making sure that the error corresponding to training of the network is minimized. This novel pattern recognition system is applied for classification of VEP signals in 11 MS patients and 11 HCs. A feature extraction routine was performed on the VEP signals, and later substantial features were selected in an optimized feature subset selection scheme employing Ant Colony Optimization (ACO) and Simulated Annealing (SA) algorithms. This alone provided further information regarding clinical value of many previously unused VEP features as an aide for making the diagnosis. The newly designed computational intelligence system is shown to outperform popular classifiers (e.g., multilayer perceptron, support-vector machine, etc.) and was able to distinguish MS patients from HCs with an overall accuracy of 90%.

neuroscience↗

bioRxiv2022

Compression-enabled interpretability of voxel-wise encoding models

Voxelwise encoding models based on convolutional neural networks (CNNs) have emerged as state-of-the-art predictive models of brain activity evoked by natural movies. Despite their superior predictive performance, the huge number of parameters in CNN-based models have made them difficult to interpret. Here, we investigate whether model compression can build more interpretable and more stable CNN-based voxelwise models while maintaining accuracy. We used multiple compression techniques to prune less important CNN filters and connections, a receptive field compression method to select receptive fields with optimal center and size, and principal component analysis to reduce dimensionality. We demonstrate that the model compression improves the accuracy of identifying visual stimuli in a hold-out test set. Additionally, compressed models offer a more stable interpretation of voxelwise pattern selectivity than uncompressed models. Finally, the receptive field-compressed models reveal that the optimal model-based population receptive fields become larger and more centralized along the ventral visual pathway. Overall, our findings support using model compression to build more interpretable voxelwise models.

neuroscience↗