IRNetv: A deep learning framework for automated brain tumor diagnosis

Abstract

A brain tumor is a severe disease that typically shortens life expectancy in most cases. Thus, early diagnosis of tumors is of paramount importance to reduce the mortality rate and revamp the quality of life. The emergence of intelligent learning algorithms can facilitate the automated diagnosis of tumors with high accuracy and efficiency. To this end, this study proposes a fully automated, interpretable deep learning-based model for brain tumor diagnosis. Convolutional Neural Network (CNN) model combining the advantages of inception modules and residual connections is designed and implemented. As CNN is employed for feature extraction and classification by analysis of medical images, the need for laborious feature extraction using statistical methods is mitigated. Also, the proposed architecture possesses the ability to extract diversified features by using various convolutional filters or kernel sizes. The experimental evaluations of the proposed architecture using two public datasets consisting of 4600 and 253 brain images achieved an overall classification accuracy of more than 99%, which is better than the previous studies.