An Interactive Deep Learning Approach for Brain Tumor Detection Through 3D-Magnetic Resonance Images
Sahar Gull, Shahzad Akbar, Khadija Safdar
Abstract
In medical field, brain tumor detection is an arduous task. Early segmentation of brain tumor is a crucial challenge to save the patient’s life by deep learning (DL). For efficient and reliable detection and classification of brain tumor, a convolutional neural network(CNN) based unified framework is proposed in this research study. The purpose of this research is to present a fully automated and effective solution to segment brain tumor than the existing researches. Additionally, it produces the performance with less error rate using 3D-MR images retrieved from BRATS 2018 dataset. The proposed approach includes pre-processing, data augmentation, segmentation and binary classification of brain tumor. In this regard, two different classifiers (Dense-Net, Dark-Net) are used for classification. The proposed framework achieved the dice similarity coefficient (DSC) of 97.91%, and accuracy of 98.67% for segmentation on 3D-MR images BRATS 2018 dataset. Similarly, the proposed framework achieved DSC of 98.14 %, accuracy of 98.26% on Dense-Net classifier and DSC of 96.4 %, accuracy of 96.52% on Dark-Net classifier for brain tumor classification on 3D-MR images BRATS 2018 dataset. The results indicate that the Dense-Net classifier accomplished a high accuracy than the Dark-Net classifier. Furthermore, we have also compared our framework with previous studies and outcomes demonstrate that our developed method achieved better segmentation and classification accuracies. Our proposed approach provides competitive performance and can be successfully applied in clinical medical applications.