Tumor Localization And Classification From MRI Of Brain Using Deep Convolution Neural Network

Abstract

The medical field's most significant elements are early disease detection, appropriate treatment decisions, pre-and post-surgical patient behavior, and caretaker burdens. All such elements work towards improving the patient's quality of life. Hence, this study aims to localize and classify the grade of brain tumors using an interactive deep-learning technique, which assists us in measuring the impact of brain tumors on a patient's quality of life. The present study also proposes the physiological and psychological quality of life (PPQOL) model.

In this study, we collected data from a population 150, comprising brain tumor patients, caretakers, and medical professionals. We categorized the population as patients, caretakers, and medical expert team members. We also categorized patients based on pre and post-surgical status. For processing brain tumor MR images, an interactive automated deep learning technique is developed to localize and classify the various brain tumor grades.Further, a hybrid convolution neural network is developed by combining U-net and Z-net models. To measure patients' quality of life, we developed a matrix through questionnaire formulation to analyze physical, psychological, and social well-being. In summary, we classified brain tumor grades, identified the patient's surgical status, and further developed the quality of life matrix for a particular patient.

Our proposed methodology, evaluates the relationship between brain tumor grades, pre- and post-surgical analysis, and its impact on a patient's quality of life. Proposed method gives the highest accuracy for tumor grade classification. The model was trained for 100 epochs, and the benchmark results obtained outperformed those of existing models. The proposed Hybrid-CNN model achieved 0.9261 for dice, 0.9957 for pixel accuracy, and for F1-scores 0.8178. Based on the segmentation, the accurate identification of the tumor grade is categorized as T1, T2, T1ce, and FLAIR. The classification of tumor grades is further correlated with the developed quality of life matrix, which evaluates the effect on physiological, psychological, and social behavioral factors.

As neurological surgery needs a very precise location of brain tumors, our automated hybrid-CNN segmentation model provides clear visibility of brain tumor boundaries using accurate tumor segmentation, which can assist surgeons during surgical events. Furthermore, classification results are mapped with the quality of life matrix, which is beneficial to measure patients' progress in the pre-and post-surgery phases. The proposed research helps medical experts and caretakers deal with the various corrective measures for improving patients' lives.