GTO-U-net synergy: Pioneering hyperparameter optimization for enhanced medical image segmentation

Abstract

The combined use of a Modified Gorilla Troops Optimization (MGTO) algorithm and a U-Net framework for medical imaging was evaluated in this research. The MGTO has been modified to find hyperparameters necessary to stabilize training (learning rate, batch size, and filter size) as well as to improve convergence time. In that sense, the MGTO's ability to increase the degree of balance between exploration and exploitation during hyperparameter search can improve efficiency while navigating through the complex, non-convex optimization surfaces of deep learning models. Results based on the BraTS20 Dataset support that the MGTO optimized learning configuration provided significantly better performance than some standard optimization technique methods (99.37% accuracy; IoU = 0.86; loss = 0.012 etc.). Additionally, the MGTO-generated improvements in convergence characteristics reduced the total training duration approximately 25% compared to more traditional optimization techniques. Overall, the results indicate that systematically optimized hyperparameters contribute directly to improved learning efficiency, segmentation accuracy and computational performance, further demonstrating how effective the proposed MGTO framework is for use in medical imaging applications.