Detecting Brain Tumors in Mri Images Using a Hybrid Approach of Machine Learning and Bayesian Optimization

One of the most important neurological diseases that demands accurate and timely diagnosis by magnetic resonance imaging (MRI) is brain tumors. But the manual analysis of MRI images is time-consuming and may be influenced by image characteristics and diagnostic interpretation. In this study, a systematic pixel-based feature extraction method is proposed in combination with a Deep Multi-Layer Perceptron (DMLP), based on Bayesian Optimization (BO), for optimized brain tumor detection without the need for human intervention. The proposed framework involves processing the MRI images to obtain the desired region of interest (ROI) and deriving the required numerical features based on each extracted pixel of the ROI. Bayesian Optimization is used to automate the decision of the number of neurons in the two hidden layers, minimizing the need to design the structure of the network manually. The optimized model in the second phase of the optimization process had 126 neurons in the first hidden layer and 59 neurons in the second hidden layer, with a best model validation performance of 98.54%. With the aid of this optimized architecture, the final model achieved an accuracy of 98.84%, and the precision, recall, and F1-score were 98.84%, 98.82%, and 98.83% on the evaluated test data, respectively, while the specificity was 98.82%. The results have shown that systematic MRI pixel extraction, CAD, DMLP classification, and Bayesian hyperparameter optimization can offer a viable computational framework to automate brain tumor detection.