Evaluation of Effect of Optimizers and Loss Functions on Prediction Accuracy of Brain Tumor Type Using a Light Neural Network

Background: Accurate identification of brain tumor categories is essential for the effective therapeutic management of brain tumors. The complexity and diversity of the information involved necessitate the use of advanced and efficient techniques to enhance diagnostic accuracy. Materials and Methods: The main objective of this study was to evaluate the effectiveness of various types of loss functions, including Binary Cross-Entropy (BCE), Categorical Cross-Entropy (CCE), and Mean Squared Error (MSE), along with different optimizers such as Stochastic Gradient Descent (SGD), Adam, RMSprop, Nadam, and Adagrad. To identify the best scenarios between optimizers and loss functions, a dataset of ۷,۱۲۵ MRI images was divided into four categories: glioma, meningioma, pituitary, and no-tumor. The model was trained and tested over ۱۲ courses, utilizing appropriate training parameters and maintaining an equal distribution of data. Results: The performance of the model was evaluated using multiple metrics, including accuracy, precision, recall, F۱ score, confusion matrices, and ROC curve analysis. The results indicated that the Adam optimizer, when paired with either the CCE or BCE loss function, performed better than other combinations, achieving test validation accuracies of ۹۷.۸۹% and ۹۷.۴۰%, respectively. Additionally, Nadam and RMSprop were found to outperform the other optimizers. Conclusion: In conclusion, the results of this study suggest that integrating the most efficient optimizer and loss function into the convolutional neural network (CNN) architecture is a crucial step toward improving the accuracy of tumor classification. This advancement is significant for enhancing therapeutic management and outcomes for patients with brain tumors.

کلیدواژه ها:

Convolutional Neural Network ، Optimizer ، Loss Function ، Deep Learning ، Tumor Type Classification