Early Detection of Brain Tumor and Survival Prediction Using Deep Learning and An Ensemble Learning from Radiomics Images

The brain is the most vital organ present in the human body & no doubt that if abnormal growth of brain cells has been detected, it leads to the development of Brain Tumor (BT) which is the most dangerous disease & if not diagnosed & treated timely in such circumstances the results are the fatal & leading cause of death. Many techniques are available for radiographic image segmentation & for fast clinical Pixels (PXs) segmentation used for accurate classification strategies for early detection & diagnosis of disease. Various radiological images (RI) like MRI, CT scan, X-Rays, etc have become a crucial tool for the diagnosis of such critical diseases. The individual effect of Deep Learning (DL) & an Ensemble Learning (EL) algorithm has shown an augmentative impact on accurate BT or Brain Cancer (BC) classification. Automated segmentation of BC or BT is an essential step for the early diagnosis of various stages of tumor or cancer, generally BT can be classified into 03 types they are as follows: Meningioma (MEG), Pituitary (PITT), Glioma (GBM), & No Tumor (NT). The monitoring of disease progression & its survival rate can be determined using the CNN, VGG16, ResNet50, Inception V3 modules. The performance of the proposed multi modal for the current work using different architecture for categorization of various types of Tumor & survival rate will be evaluated in terms of Accuracy, Precision, Dice Score (F1 Score), Recall, Processing time, etc. The Accuracy detection & type of tumor levels performed via radiologists is dependent on their experience & it is too time-consuming which delayed the immediate treatment resulting in the excessive progress in the disease, so computer-aided technology could be very critical to aid with the diagnosis accuracy. On evaluating the results of the work done the over all best results was observed in ResNet50 (Fine Tunning) with 98.63 %, 97% in Glioma tumor (GBM) detection, 98% in Meningioma tumor (MEG) detection, 98 % in Pituitary tumor (PITT) detection, & 99% in No Tumor (NT) detection, the obtained results were tested on a dataset of 7022 MRI images. Experimental results display the robustness of the proposed ResNet50 (fine tunning) architecture.