<p>Brain tumors are characterized by the uncontrolled growth of abnormal cells within the brain tissue. Their size and location can significantly impact cognitive and physical functions such as coordination, memory, and overall neurological health. These tumors can either originate within the brain or metastasize from other areas of the body. Traditional diagnostic approaches often struggle with limited accuracy, high false-positive rates, and difficulty detecting complex tumor patterns. In order to address these issues, this research proposes a novel approach that integrates the Dynamic Path-Controllable Deep Unfolding Network (DPCDUN) with the Duck Swarm Algorithm (DSA) approaches to improve detection accuracy of brain tumor. Initially, MRI images are collected BraTs 2018 and Figshare dataset. Subsequently, the Structure-Preserving Diffusion Model for 3D Mesh Denoising (SPDM-3DMD) for pre-processing improves the quality of 3D MRI data for brain tumor imaging. After that, Prompt-Tuned Multi-Task Taxonomic Transformer (PTM3T) uses shared taxonomic information to manage several related tasks at once. Then, Polymorphic Graph Attention Network (PGAN) is used to extract the feature and detect brain MRI images using Dynamic Path-Controllable Deep Unfolding Network (DPCDUN) model for complex spatial relationships between different parts of the tumor. Finally, improve the brain tumor detection process, Duck Swarm Algorithm (DSA) is ensuring optimal performance. The proposed DPCDUN-DSA approach achieves exceptional accuracy (99.8% on BraTs 2018), F1 Score (98.9% on BraTs 2018), accuracy (99.8% on Figshare), F1 score (98.7% on Figshare), and error rate 0.1% significantly improving brain tumor detection, while minimizing false positives, making it highly accuracy for medical diagnosis.</p>