Designing an Automatic Brain Tumor Detection and Patient Survival Prediction with Atrous Spatial Pyramid Pooling

Nowadays, most people have been affected by brain tumor disease, which is the deadliest disorder among adults and children. It poses a significant health concern and has serious consequences for the patients. Brain tumors grow when the proliferation of brain's abnormal cells. The timely diagnosis of brain tumors can result in proper treatment planning that enhances the survival chances of the patients. 'Computed Tomography (CT) and Magnetic Resonance Imaging (MRI)' are considered as imaging models to monitor brain tumors. Nevertheless, because of the unavailability and complexity of the resources to train the models of deep learning, it is difficult to recognize the malignant and benign regions. Hence, this work presented an effective detection method using deep learning. The novelty of the developed model is to automatically detect brain tumors and forecast the survival rate of patients based on the detection output. Thus, the developed model helps to improve the life quality of the patients and also aids clinicians in making proper and accurate treatment plans. Initially, the required images for identifying brain tumors are taken from the standard data resources. The collected images are directly forwarded to the classification module. In this module, the classification is performed using the Region Attention-based Dilated Densenet with Atrous Spatial Pyramid Pooling (RADD-ASPP). The generated outcome classifies the brain tumor into normal and abnormal. For predicting the patient survival rates, the abnormal features are then subjected to Long Short-Term Memory (LSTM). This technique efficiently predicts the patient's survival rates into mild, moderate, and severe. The implemented deep learning framework is then validated by including several existing techniques. From the achieved result, better detection as well as prediction rates are achieved by the proposed model. Therefore, the designed approach is more robust to assist the physician in brain tumor detection and survival prediction.