Corpus Callosum Segmentation Using Enhanced Hyperparameters in the U-Net Model

Abstract

This study presents a deep learning segmentation technique for effectively segmenting the brain anatomy, particularly the corpus callosum, using magnetic resonance imaging. The approach employed in this study involves utilizing U-Net with Hyperparameter Optimization. Following the training of the program, the model is evaluated using a test set sourced from the same dataset. The evaluation compares the Ground Truth and Predicted Images using the dice coefficient (DC). The method was tested four times with different hyperparameters, yielding the highest accuracy F1-Score of 92.16% with 150 epochs and the lowest accuracy of 89.4% with 50 epochs. These outcomes indicate the effectiveness of the method used in producing favorable results.