Design of Brain Tumor Detection System on MRI Image Using CNN

Brain tumor is an abnormal proliferation of brain cells, which may be benign or malignant in nature. Brain cancer, which is frequently diagnosed in individuals of all ages, is a malignant form of a brain tumor and one of the most severe forms of cancer. Each year, an estimated 300 cases of brain tumors, including those in children, are diagnosed in Indonesia. To detect brain tumors, imaging methods such as Magnetic Resonance Imaging (MRI) are utilized. However, radiologists' manual examination of MRI scans might lead to conclusions that differ from one doctor to the next (interobserver error). Research on brain tumor type classification on MRI images is also limited. To identify various types of brain tumors in MRI images, we will therefore construct a system utilizing Convolutional Neural Networks (CNN) and transfer-learning methods. In this study, the Flask framework was successfully used to develop a web-based application to identify distinct form of brain tumors in MRI scans. The model makes use of CNN architecture, a ResNet50V2 base model trained on the ImageNet dataset, a head model with 512 nodes and one entirely connected layer, and an output layer that forecasts the input into four classes of brain MRI images, including 'Normal','Glioma', 'Meningioma', and'Pituitary'. Appropriate parameter settings were used to achieve the highest accuracy. In this study, Adam optimization algorithm was used with 60 epochs and a batch size of 32. Additionally, a ten-fold cross-validation technique was implemented. 95% accuracy rate was achieved by implementing the proposed architecture.