This paper presents a novel approach for brain tumor segmentation in MRI images, combining handcrafted features with deep learning. The authors propose a cascaded strategy that integrates handcrafted features, which capture domain knowledge, with a Global Convolutional Neural Network (GCNN). The GCNN architecture consists of two parallel CNNs: the Confidence Surface (CSPathways CNN) and the MRI Pathways CNN (MRIPCNN). The CSPathways CNN processes the handcrafted features, while the MRIPCNN processes the four MRI modalities (T1, T1c, T2, and FLAIR) along with the ground truth. This method aims to leverage the strengths of both feature-based and data-driven approaches to improve the accuracy of brain tumor segmentation. The proposed model was evaluated using the BraTS dataset, which includes MRI volumes and annotations for different tumor sub-regions. The authors compared their method with state-of-the-art techniques and achieved a Dice score of 87%, outperforming existing methods. The study highlights the potential of combining traditional feature-based approaches with deep learning to enhance the accuracy of brain tumor segmentation, which could aid clinicians in diagnosing and treating patients more effectively. Keywords: Brain tumor, Health Risks, Handcrafted features, Global-pathway CNN, Local-pathway CNN.This paper presents a novel approach for brain tumor segmentation in MRI images, combining handcrafted features with deep learning. The authors propose a cascaded strategy that integrates handcrafted features, which capture domain knowledge, with a Global Convolutional Neural Network (GCNN). The GCNN architecture consists of two parallel CNNs: the Confidence Surface (CSPathways CNN) and the MRI Pathways CNN (MRIPCNN). The CSPathways CNN processes the handcrafted features, while the MRIPCNN processes the four MRI modalities (T1, T1c, T2, and FLAIR) along with the ground truth. This method aims to leverage the strengths of both feature-based and data-driven approaches to improve the accuracy of brain tumor segmentation. The proposed model was evaluated using the BraTS dataset, which includes MRI volumes and annotations for different tumor sub-regions. The authors compared their method with state-of-the-art techniques and achieved a Dice score of 87%, outperforming existing methods. The study highlights the potential of combining traditional feature-based approaches with deep learning to enhance the accuracy of brain tumor segmentation, which could aid clinicians in diagnosing and treating patients more effectively. Keywords: Brain tumor, Health Risks, Handcrafted features, Global-pathway CNN, Local-pathway CNN.