Brain tumor detection in combined 3D MRI and CT images using Dictionary learning based Segmentation and Spearman Regression

3D CT and MRI brain images are used in brain tumor detection due to their tendency to compare tissue density. Hence various research has been presented previously to detect brain tumors from the CT and MRI image but they faced issues in both segmentation and classification processes. During the detection of brain tumors, the existing segmentation techniques require higher decomposition levels and were unable to accurately segment the mutually exclusive and exhausted regions and during the classification of various tumor types, the linear inseparability occurs due to tumor regions' significant similarity and lack of co-occurrence matrix for principal distinctive features. Hence, to accurately detect brain tumors, combined 3D CT and MRI brain images have been used in the novel model named Dictionary learning based Segmentation and Spearman Regression in which Sparse Mahalanobis Dictionary based MMRF Model has been proposed that utilized sparse coding with Mahalanobis Dictionary learning enables the creation of a dictionary matrix that discriminates between healthy and tumor tissues, achieving isolation without complex decomposition levels. When combined with the probabilistic weighted segmentation process in a Multimodal Markov Random Forest, it effectively delineates mutually exclusive and exhaustive regions in multimodal images, preventing tissue loss. Moreover, to solve the issues in the classification of various types of brain tumors, Nested PatchNet Spearman Regression is utilized in which principal distinctive features were extracted by forming nested 3 × 3 patches, and their co-occurrence matrix was generated to find the correlation between various tumor regions, thereby effectively eliminating linear inseparability and classifying brain tumors as the pituitary, meningioma, and glioma using Coyote Optimization-driven Lagrangian neural networks. The result obtained showed that the proposed model outperforms existing techniques with a high detection rate and low loss.