A robust tuned EfficientNet-B2 using dynamic learning for predicting different grades of brain cancer

The brain is the central nervous system component, consisting of nerve cells called neurons. Neurons play a vital role in transmitting signals throughout the body. When cells within the brain undergo abnormal and uncontrolled growth, it can lead to the formation of brain tumors. Prompt detection and treatment of these tumors are crucial to prevent serious consequences, including death. The location and size of a brain tumor can impact a person’s facial and head symmetry. Accurately diagnosing brain tumors requires the expertise of doctors who can manually assess and categorize them using magnetic resonance imaging (MRI) scans. This precise diagnosis is crucial for planning effective treatment options and enhancing the patient’s quality of life. Deep learning (DL) has the potential to greatly assist in the diagnosis and treatment of brain cancers. We can ensure more accurate clinical diagnoses and improve treatment outcomes by utilizing DL techniques to analyze MRI images and classify brain tumors. In this paper, we propose a DL technique utilizing the EfficientNet-B2, which leverages the power of deep neural networks (DNN) for brain tumor detection. Our model incorporates the adaptive learning rate (ALR) technique, where the learning rate (LR) adjusts automatically at the beginning of each step based on the training accuracy and loss value from the previous step. Moreover, our proposed model utilizes the gradient-weighted class activation mapping (Grad-CAM) algorithm to identify specific areas impacted by ground glass opacities (GGO) that are linked to brain tumors. The proposed DL model aids in the early detection of brain tumors, allowing for timely intervention and improved patient outcomes, and it streamlines the diagnostic process, resulting in reduced time and cost for patients. We conducted tests using two public datasets, Br35H and brain tumor (BT), for binary and multiclass classification tasks, respectively. They were preprocessed using resizing and normalization techniques to ensure consistent input. Our proposed DL model was then compared with traditional classifiers. For the multiclass classification task, our proposed model achieved accuracy, specificity, precision, recall, and F1-score of 99.81 %, 99.87 %, 99.62 %, 99.62 %, and 99.62 %, respectively. Additionally, the model achieved a perfect accuracy score of 100 % for the binary classification task. The assessment of our proposed classifier revealed that the proposed DL model based on EfficientNet-B2, incorporating the ADL rate technique, shows promising potential in accurately and efficiently diagnosing brain tumors. Its high performance and ability to reduce diagnosis time and cost make it a valuable instrument for clinicians in the realm of neurology.