A low parametric CNN based solution to efficiently detect brain tumor cells from ultrasound scans

Abstract

Implementation of deep neural networks in medical imaging of brain cells to identify tumors is shaping up to be a reliable approach in medical science. Early and accurate detection of brain tumor cells is critical for effective patient treatment. Initially, this paper demonstrates three widely used CNN models, VGG16, ResNet50, and MobileNet, to identify tumor cells in brain MRI images with an accuracy of 97%, 94.5%, and 99% respectively. The dataset used for this study contains brain MRI scans called Br35h. The purpose of this research is to develop a modified CNN model to attain similar performance statistics to widely accepted CNN models and extract all meaningful and precise information from images with the least amount of error possible while maintaining greater run-time efficiency. The proposed model achieved an overall classification accuracy of 98.5%. Finally, the model's performance is also compared with that of the previously stated models, and it is observed that it performs on par, if not better, than those models despite having a fraction of the total parameters. This study aims to contribute to the computer-aided diagnostic (CAD) system by implementing the proposed model with relatively fewer computational power requirements.