IMG-18. Deep Learning Method for Region Prediction and Recognition of Brain Tumors: An MRI Semantic Segmentation Mode

Based on statistics from the Ministry of Health and Welfare in 2021, brain tumor cases account for 0.62% of all malignant tumors, while deaths caused by these tumors make up 1.14% of all malignant tumor deaths in the same year. This indicates that although the incidence rate of brain tumors is relatively low, their impact on patient survival rates is significant. Due to the unique nature of brain function, the diagnosis and treatment of brain tumors differ from those of other tumors. Brain tumors vary widely in type and have complex causes, with clinical manifestations including blurred vision, hearing loss, headaches, and seizures. Therefore, researching diagnostic and treatment methods for brain tumors is of great importance. With the rise of medical imaging technology and intelligent healthcare, the application of artificial intelligence in brain tumor diagnosis is crucial for improving accuracy and reducing diagnostic time, particularly for common brain tumors like pituitary tumors and glioblastomas. In light of this, this project aims to utilize magnetic resonance imaging (MRI) sequences to create a dataset and apply various deep learning models, such as YOLOv7, YOLOv9, U-Net, Dense U-Net, and Swin U-Net, for semantic segmentation of brain tumors to enhance the diagnostic accuracy of pituitary tumors and glioblastomas. The proposal aims to use deep learning algorithms for tumor segmentation and identification in brain MRI, incorporating attention mechanisms to improve the model’s generalization ability. Through automated MRI analysis methods, the project seeks to rapidly and accurately identify brain tumor locations and compare various semantic segmentation models to identify the most suitable method for brain tumor segmentation. This research will assist physicians in making more accurate diagnoses and more effective treatments, while also improving patient quality of life and reducing healthcare costs.