FF-UNet: Feature fusion based deep learning-powered enhanced framework for accurate brain tumor segmentation in MRI images

Abstract

Medical imaging technology plays a crucial role in various medical sectors, aiding doctors in diagnosing patients. With brain tumors becoming a significant health concern due to their high morbidity and mortality rates, accurate and efficient tumor segmentation is essential. Manual segmentation methods are prone to errors and time-consuming. In this study, we investigate the potential of deep learning-based brain tumor MRI image segmentation techniques. We propose an enhanced approach called FF-UNet, which leverages feature fusion and combines the power of UNet and CNN models to improve segmentation accuracy. Preprocessing techniques are employed to enhance tumor visibility, followed by the utilization of a customized layered UNet model for segmentation. To mitigate overfitting, dropout layers are introduced after each convolution block stack. Additionally, a CNN process leverages the context of brain tumor MRI images to further enhance the model's segmentation performance. Experimental results demonstrate that our proposed framework outperforms state-of-the-art models in differentiating brain tissue. Across all datasets, our method achieves above 98% accuracy, with precision and Jaccard coefficient both exceeding 90%. Evaluation metrics such as the Jaccard index, sensitivity, and specificity validate the robust performance of our approach. The FF-UNet model holds great potential as a viable diagnostic tool, enabling radiologists to accurately segment brain tumor images and improve patient care.