NeuroTumorXpert: An Advanced Hybrid Deep Learning Model for Multi-Classification of Brain Tumor via Magnetic Resonance Imaging

Abstract

Introduction: Brain tumors must be detected early in order to improve the patient's prognosis and course of therapy.  To create specialized treatment approaches, brain tumors must be properly classified, segmented, and graded.  Despite the growing use of MRI in brain exams and improvements in AI-based detection methods, it can be difficult to create a reliable model for tumor diagnosis and classification from MR images.

Objectives: Thus, there is an urgent need for a more automated, effective, and precise method for detecting, classifying, and grading brain tumors. The need was for a solution that could process medical imaging data quickly and reliably, while ensuring accessibility and reducing the dependency on human expertise, ultimately aiding in timely diagnosis and improved patient outcomes.

Methods: Current diagnostic techniques mainly rely on radiologists' manual interpretation, which can cause delays and human error, particularly when malignancies are still in the early stages. Traditional imaging technologies, while useful, often fail to provide precise localization and classification of tumors, especially when dealing with varying shapes, sizes, and textures. So, this paper introduces a multi-classification brain tumor diagnosis model. It combines hybrid methods and deep learning models to detect tumors, classify tumors and grade glioma tumors.

Results: The proposed model, NeuroTumorXpert, has high accuracy, 99.69% for detection via fine-tuned VGG16 model, 99.15% for tumor type classification via fine-tuned InceptionV3 model and 96.64% for glioma tumor grading via Inception-ResNet-v2.

Conclusions: Identifying gaps and comparison was done to find out the lacking areas in past research related to multi classification of brain tumor, advancement in deep learning algorithm on medical imaging to increase accuracy and able to do multi classification