RF-DETR with multi-scale feature fusion and explainable attention for accurate brain tumor detection in MRI

Nonetheless, the detection and localization of the tumor areas are still difficult due to differences in tumor shape, size, and pattern of intensity changes amongst scans of different MRIs. The proposed study is an improved transformer-based detection framework that can combine RF-DETR with multi-scale feature fusion and explainable attention mechanisms to detect brain tumors better. The proposed solution utilizes a hierarchical feature extraction model, which is based on a convolutional convolutional backbone and the multi-scale fusion to extract fine-grained and global contextual information.

A transformer encoder-decoder network is used to end-to-end approximate long-range spatial dependencies and predict locations and class labels of tumors. Moreover, explainable attention module is added to indicate the areas that play the most important part in model predictions and, therefore, enhance interpretability and clinical importance. The presented model is tested on the brain MRI data which includes four classes: brain, glioma, meningioma, and pituitary.

Experimental results show that it has high detection performance, with a precision of 0.9545, recall of 0.9494, F1-score of 0.9520, mAP at 0.5 of 0.9432 and mAP at 0.5:0.95 of 0.7403 on the validation set. Comparative analysis has revealed that the proposed framework is better in detecting accuracy, and moderate computational complexity and outperforms the current CNN-based and transformer-based models. The ablation research also confirms that multi-scale feature fusion, as well as explainable attention, are helpful in providing impressive performance on models.

The findings show that the proposed framework can offer reliable, precise, and interpretable tumor detection and is thus a promising tool in the field of computer-assisted diagnosis and clinical decision support used in neuroimaging studies. The authors gratefully acknowledge the Deanship of Scientific Research (DSR) at King Abdulaziz University, Jeddah, Saudi Arabia, for funding this project under grant No. (IPP: 841-830-2026) and for providing technical support. This Project was funded by the Deanship of Scientific Research (DSR) at King Abdulaziz University, Jeddah, Saudi Arabia under Grant no. (IPP: 841-830-2026).

Department of Information Systems, Faculty of Computing and Information Technology, King Abdulaziz University, 21911, Rabigh, Saudi Arabia Arshad Hashmi, Anas W. Abulfarj, Mohammed Ameen & Hasan J. Alyamani Global College of Engineering and Technology, Muscat, Oman Department of Computer Science, Faculty of Computing and Information Technology, King Abdulaziz University, 21911, Rabigh, Saudi Arabia The authors declare that they have no conflicts of interest to report regarding the present study.

The dataset is openly accessible at given link: https://universe.roboflow.com/bnm/certificate-forgery-detection-avxar Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material.