This paper proposes a hybrid deep learning framework for classifying brain tumors from MRI scans, combining Bayesian optimization and a quantum theory-based marine predator algorithm (QTbMPA) for feature selection. The framework addresses the challenge of class imbalance in the dataset by using a sparse autoencoder to generate new images, followed by the fine-tuning of two pretrained neural networks using Bayesian optimization. Deep features are extracted from the global average pooling layer, and the QTbMPA algorithm is used to select the best features for final classification. The proposed framework was tested on an augmented Figshare dataset and achieved an accuracy of 99.80%, a sensitivity rate of 99.83%, a false negative rate of 17%, and a precision rate of 99.83%. The results show significant improvement in classification accuracy compared to existing methods. The framework includes contrast enhancement, data augmentation, hyperparameter optimization, and feature selection to improve the performance of deep learning models in classifying brain tumors. The QTbMPA algorithm is designed to select the best features from the two pretrained models and fuse them using a serial-based approach for final classification. The framework demonstrates the effectiveness of combining Bayesian optimization and quantum theory-based algorithms in improving the accuracy of brain tumor classification from MRI scans.This paper proposes a hybrid deep learning framework for classifying brain tumors from MRI scans, combining Bayesian optimization and a quantum theory-based marine predator algorithm (QTbMPA) for feature selection. The framework addresses the challenge of class imbalance in the dataset by using a sparse autoencoder to generate new images, followed by the fine-tuning of two pretrained neural networks using Bayesian optimization. Deep features are extracted from the global average pooling layer, and the QTbMPA algorithm is used to select the best features for final classification. The proposed framework was tested on an augmented Figshare dataset and achieved an accuracy of 99.80%, a sensitivity rate of 99.83%, a false negative rate of 17%, and a precision rate of 99.83%. The results show significant improvement in classification accuracy compared to existing methods. The framework includes contrast enhancement, data augmentation, hyperparameter optimization, and feature selection to improve the performance of deep learning models in classifying brain tumors. The QTbMPA algorithm is designed to select the best features from the two pretrained models and fuse them using a serial-based approach for final classification. The framework demonstrates the effectiveness of combining Bayesian optimization and quantum theory-based algorithms in improving the accuracy of brain tumor classification from MRI scans.