Developing a Hybrid Framework for Brain Tumor Multi Classification from MRI Images Using LSTM

Globally, brain cancers create a serious threaten to public health. The class of tumor and early prognosis have a major impact on the patients' survival. Traditional Machine Learning (ML) models, rely on handcrafted feature extraction but often struggle with complex spatial patterns. Deep learning-based approaches, such as VGG16, excel at extracting hierarchical spatial features, while Long Short-Term Memory (LSTM) approaches can model sequential dependencies within feature representations. In this research, a hybrid approach integrating VGG16, LSTM, and ML classifiers for improved brain tumor classification is proposed. VGG16 serves as a feature extractor, generating deep spatial representations, which are then passed to an LSTM network to capture sequential dependencies. Additionally, ML techniques such as SVM, DT, XGB, NB and KNN are deployed for classification. The efficacy of the hybrid deep learning approach is compared with related approaches, demonstrating superior accuracy and robustness. This study highlights the advantages of combining CNNs, LSTMs, and ML classifiers to enhance brain tumor classification, providing an efficient and interpretable framework for medical diagnosis. The most notable enhancement occurred when SVM was combined with the hybrid VGG16-LSTM approach achieving an accuracy of 98.76%. This approach improves both categorization performance and computational efficiency, making it an efficient approach for forecasting multiclass brain tumors.