YoDenBi-NET: YOLO + DenseNet + Bi-LSTM-based hybrid deep learning model for brain tumor classification

Abstract

Brain tumor, which is the deadliest disease in adults, grows rapidly and disrupts the functioning of organs. Brain tumors can be of different types, depending on their shape, texture, and location. The correct detection of these types helps the field specialist to make the correct diagnosis and thus save the patient's life. In this study, a three-stage hybrid new classification framework based on YOLO + DenseNet + Bi-LSTM is proposed to classify glioma, meningioma, and pituitary brain tumor types. In this framework, the brain region is detected first through the YOLO detection algorithm. In the second stage, deep features are extracted from this region via a pre-trained deep learning architecture, and in the final stage, brain tumor classification is performed by way of the Bi-LSTM network which is another deep learning model. The proposed model offers high test accuracies of 99.77% and 99.67%, respectively, for three brain tumor types using hold-out and tenfold cross-validation techniques on a dataset containing 3064 MRI images. With its high performance in validation and test sets, the proposed hybrid model is better than other previous studies and so it can be used as a useful decision support system for field specialists.