Early Diagnosis of Brain Tumor Using Machine Learning Approaches

Abstract

Introduction: The brain tumor is formed by abnormal brain cells, leading to serious organ dysfunction and potentially death. These tumors vary widely in size, texture, and location. Diagnosing brain tumors is a lengthy process that requires the expertise of radiologists. Brain tumors are classified into types such as glioma, meningioma, pituitary tumors, and no tumor. With increasing patient numbers and data volumes, traditional diagnostic methods have become both costly and inefficient.

Objectives: Consequently, researchers have developed algorithms aimed at detecting and classifying brain tumors with an emphasis on accuracy and efficiency. Deep learning techniques are now commonly used to create automated systems capable of accurately diagnosing or segmenting brain tumors, particularly in classifying brain cancer.

Methods: This approach often incorporates transfer learning models in medical imaging. In this study, a modified Xception model is proposed, incorporating additional parameters to enhance the model’s efficiency.

Results: This modified Xception model was tested on the Sartaj brain-tumor-MRI dataset, achieving an accuracy of 99.02%, sensitivity of 99.0%, and specificity of 99.1%, with an F1 score of 99.1%. These performance metrics confirm that the proposed model is highly effective for diagnosing brain tumors.

Conclusions: Comparative analysis with other models indicates that this framework offers reliable and timely detection of various brain tumors. The results validate the proposed model as a valuable decision-making tool for experts in brain tumor diagnosis.