Brain-Tumor Segmentation Using U-Net Neural Network Architecture

Brain tumor segmentation, Delineating tumor areas from a fin-n-healthy brain tissue in medical pictures is critical for proper diagnosis, planning of treatment, and observing alignment progression. This task is critical for ensuring sustainable healthcare by enabling personalized treatment strategies, enhancing recourse distribution, and improving patient results. Traditional physical techniques for segmentation are lengthy, subjective, and highly susceptible to inter-observer variability, making healthcare delivery less efficient and cost-effective. To clarify these limits, this research investigates the use of the U-Net neural-network-architecture for the purpose of the segmentation automatically of brain tumor using Magnetic Resonance Imaging inputs. U-Net, through its encoder-decoder arrangement along with its skip connections, revealed exceptional results in image segmentation typically in the medical domain. The proposed technique comprises pre-processing MRI images to improve tumor visibility, modeling a U-Net on the BraTS2020 dataset, which is a collection of annotated brain tumor pictures with 4 different classes. The model is trained to fragment separate tumor sub-regions, with tumor core, edema, and enhancing tumor. The suggested model demonstrates excellent performance across various evaluation criteria. It achieves a Mean of IoU (Intersection over Union) of 0.85, indicating strong alignment amongst actual segmentations and predictions. With Sensitivity and Specificity score of 0.99, the model effectively detects true positives and true negatives. Its Accuracy is 0.99, Dice coefficient is 0.75. Results show how effective of the U-Net like architectures are in accurately segmenting brain tumors, outperforming traditional segmentation techniques.

Ҳали таржима қилинмаган