Abstract

Aims and Background: A tumour is a kind of cell that grows uncontrollably and uniformly. Brain tumors (BT) are one of the leading causes of mortality in humans. In the United States, almost half of all patients die from primary BT each year. Brain tumors are diagnosed using electronic modalities. Magnetic resonance imaging (MRI) is one of the most widely used and popular electronic modalities for tumor detection.

Objectives and Methods: In this research article, we included classification and optimization, with categorization using a deep convolution neural network (DCNN) and optimization using the AdaBound Optimizer (AO). When filtering input pictures, the aim of utilizing DCNN for classification is to maintain spatial connections. The spatial connection is critical for detecting the tumor- non-tumor area interface and assessing edge tissues. As a result, it categorizes the input with a higher likelihood, and for ease of implementation, the AO is utilized. It uses less memory, is more computationally efficient, and achieves superior nonstationary goals. AO also performs effectively with large datasets/parameters because of the noisy or sparse gradient requirement.

Results: Finally, the results show improved accuracy, faster convergence, and better generalization compared to traditional optimizers. These findings highlight the potential of the model for more reliable and efficient automated medical diagnosis.

Conclusion: According to the simulated results, the training accuracy rate of the AO model is 91.17% and 96.87% for 5 and10 epochs, respectively. Relative to the training accuracy of the SGD optimizer models for 5 and10 epochs is 96.7% and 99%, respectively.

Keywords: Deep Learning, brain tumours, saliency, DCNN classifier, AdaBound optimizer, MRI images.