Innovative deep learning and quantum entropy techniques for brain tumor MRI image edge detection and classification model

Brain Tumors (BT) are the foremost basis of cancer death. They are affected by the uncontrolled and abnormal growth of cells in the spinal canal or brain. The main issue with a BT is identifying its shape, location, and dimension. Despite numerous efforts and promising outcomes in tumour recognition, precise classification from benign to malignant type is still difficult. A frequently employed device in analyzing these conditions is a magnetic resource image (MRI); however, medical specialists' physical assessment of MRI images causes troubles owing to time restraints and variability. In the preceding few years, because of artificial intelligence (AI) and deep learning (DL), significant developments have been prepared in medical science, such as the Medical Image processing model, which aids doctors in analyzing disease timely and effortlessly; before that, it was time-consuming and tiresome. This study proposes an Innovative Deep Learning and Quantum Entropy Techniques for Brain Tumor Edge Detection and Classification (IDLQET-BTEDC) model in MRI imaging. The primary goal of the IDLQET-BTEDC model is to improve accuracy and efficiency in identifying BTs using multi-images such as detected and edge images. To accomplish this, the IDLQET-BTEDC approach involves pre-processing, which contains two processes: the wiener filter for noise removal and adaptive gamma correction for contrast enhancement. Furthermore, the segmentation process adopts dual approaches focusing on region and edge detections. The tumour region is segmented using enhanced UNet with NAdam optimization, while the quantum entropy (QE) edge detection is applied to delineate the tumour boundaries. In addition, the IDLQET-BTEDC model performs feature extraction by using Multi-head Attention fusion to combine EfficientNetV2 and Swin transformer (ST). The graph convolutional recurrent neural network (GCRNN) classifier is utilized for BT detection and classification. Finally, the hyperparameter tuning of the GCRNN model is performed by employing the Siberian tiger optimization (STO) model to achieve superior accuracy. To demonstrate the good classification outcome of the IDLQET-BTEDC approach, an extensive range of simulations is performed under the Figshare BT dataset. The performance validation of the IDLQET-BTEDC technique portrayed a superior accuracy value of 98.00 % over existing methods. © 2025 The Authors