Comparison of models across metrics.

Early diagnosis of brain tumors is important for successful treatment and better patient consequences in industrial information systems. This research employs Mask Region-based Convolutional Neural Networks (R-CNN), radiomics integration, and the Gray Level Co-occurrence Matrix (GLCM) to progress brain tumor detection and segmentation using Magnetic Resonance Imaging (MRI) images. The Mask R-CNN is employed to precisely label tumor regions of interest (ROIs) by precisely segmenting pixels. Radiomic features in the form of texture, shape, and intensity measures are calculated on the segmented ROIs to provide a numerical estimate of tumor heterogeneity. Meanwhile, GLCM-based texture features are computed to derive the fine-scale spatial classifications of pixel intensities in tumor regions, which may provide more information regarding tumor structure. Such handcrafted features are merged with deep features obtained with the help of the Mask R-CNN architecture to form a strong feature set that takes the benefits of both classical radiomics and deep learning (DL). The combination of the feature set is classified using a detection head multilayer perceptron (MLP) to predict the presence and the nature of a tumor. The experimental results on an experimental brain tumor MRI dataset proved that the proposed approach is better than independent Mask R-CNN and radiomics-based approaches. The model is very sensitive, specific and precise in differentiating between glioma, meningioma, pituitary adenoma and non-tumor cases. The feasibility of this hybrid solution is the combination of handcrafted and deep features to further the brain tumor detection that provides medical practitioners with a handy tool of detecting brain tumors early and accurately.