A momentum-based stochastic fractional gradient optimizer with U-net model for brain tumor segmentation in MRI

Abstract

Brain tumor segmentation from magnetic resonance imaging (MRI) is a critical task in medical image analysis and is essential for diagnosis, treatment planning, and monitoring. This study presents a new approach leveraging the U-Net architecture combined with a newly proposed Stochastic Fractional Moment Gradient Descent (SFM) optimizer. This proposed new hybris combination addresses the issues of convergence speed and segmentation precision. The proposed SFM optimizer introduces a fractional gradient component that provides a more refined update mechanism compared to traditional gradient descent, incorporating momentum to accelerate convergence and avoid local minima. The model was trained and validated on the multiple brain tumor segmentation datasets. The experimental results demonstrate that the U-Net with SFM optimizer outperforms conventional U-Net models using standard optimizers such as Adam and SGD. Specifically, our approach achieved a Dice Similarity Coefficient (DSC) of 0.88, surpassing the baseline U-Net model's DSC of 0.84. Additionally, our method showed a 20% improvement in convergence speed, reducing training time significantly while maintaining high accuracy. Qualitative analysis of segmentation outputs also confirmed that our model effectively delineates tumor boundaries with higher precision, particularly in challenging cases with heterogeneous tumor appearances. These results suggest that the integration of the SFM optimizer with the U-Net architecture provides a robust framework for accurate and efficient brain tumor segmentation in MRI, with potential applications in clinical practice.