An Effective Brain Tumor Classification Approach using Parallel Deep Convolutional Neural Networks

Brain tumors are a deadly disease, causing numerous deaths annually. Detecting and analyzing them quickly and automatically is vital to save countless lives worldwide. CNNs are widely employed for precise brain tumor classification, but their random feature extraction can lead to overfitting by disregarding local and global image features. To address this issue, this study introduces a parallel deep convolutional neural network (PDCNN) model. It utilizes two parallel networks to extract both global and local features from brain MRI. Additionally, dropout regularization and batch normalization techniques are employed to address the overfitting problem. This approach employs two deep CNNs with differing window sizes simultaneously to gain the benefits of simultaneous pathways, enabling it to collect both locally as well as globally characterize tumor features from MRI images. To assess the effectiveness of our proposed approach, we have used a publicly accessible brain MRI dataset known as Figshare Dataset. Our proposed approach has been successful in achieving a high accuracy of 97.39%, with a f1-score of 97%, a precision of 97%, and a recall of 97%. The proposed method achieves high accuracy and efficiency by extracting both high-level and low-level information from brain MRI images and outperforms other state-of-the-art approaches in brain tumor classification.