Automated Identification and Localization of Brain Tumor in MRI Using U-Net Segmentation and CNN-LSTM Classification

Abstract

Nowadays, the use of computers to evaluate medical images automatically is critical part of the life. Today's treatment method relies heavily on early diagnosis and accurate disease identification, which were formerly difficult for medical research to achieve. Brain Magnetic Resonance Imaging (MRI) is essential to the detection and treatment of brain tumor (BT). Tumor of the brain are the result of brain cell division that has gone awry or is otherwise out of control. The manual MRI segmentation of BT is a difficult and time-consuming process. The most critical factor in the effective treatment and identification of BT is the ability to accurately locate the tumor. The detection of BT is regarded as a difficult task in medical image processing. For analysing and interpreting MRI, there are semi-automatic and fully automated systems that require large-scale professional input and evaluation, with varying degrees of effectiveness. Automated identification and extraction of the tumor's localization from brain MRI will be proposed in this paper. To achieve this goal, the data collected from Kaggle and the collected data are processed. Then the U-Net is employed to segment the tumor region from the MRI. Next, the MRI is classified using DL models like Convolutional Neural Network (CNN), and the hybrid Convolutional Neural Network and Long Short-Term Memory (CNN-LSTM). Both process segmentation and classification are evaluated using the metrics. From the evaluation, it is identified that CNN-LSTM outperforms the CNN model.