DB-FGA-Net: Dual Backbone Frequency Gated Attention Network for Multi-Class Brain Tumor Classification with Grad-CAM Interpretability

Saraf Anzum Shreya, MD. Abu Ismail Siddique, Sharaf Tasnim · 2025

Read DB-FGA-Net: Dual Backbone Frequency Gated Attention Network for Multi-Class Brain Tumor Classification with Grad-CAM Interpretability by Saraf Anzum Shreya, MD. Abu Ismail Siddique, Sharaf Tasnim (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2510-20299.txt.

Introduction

Saraf Anzum Shreya, MD. Abu Ismail Siddique, and Sharaf Tasnim propose DB-FGA-Net, a network for classifying brain tumors in medical images across several classes. Early, precise diagnosis matters for treatment, yet many deep learning systems lean on heavy data augmentation, which can hurt generalization and clinician trust. Their design joins two backbones, VGG16 and Xception, with a frequency-gated attention block that captures local and global features together.

Grad-CAM visualizations show the tumor regions behind each prediction, linking model output to clinical reading. Without augmentation, the model reaches accuracy above ninety-nine percent on one dataset in the four-class setting and generalizes at nearly ninety-six percent on an independent dataset. A graphical interface provides real-time classification with tumor localization for clinical use.

Catalog record

CREATORSaraf Anzum Shreya, MD. Abu Ismail Siddique, Sharaf Tasnim
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE62 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2510-20299

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

DB-FGA-Net: Dual Backbone Frequency Gated Attention Network for Multi-Class Brain Tumor Classification with Grad-CAM Interpretability Saraf Anzum Shreya, MD. Abu Ismail Siddique, Sharaf Tasnim Abstract Brain tumors are a challenging problem in neuro-oncology, where early and precise diagnosis is important for successful treatment. Deep learning-based brain tumor classification methods often rely on heavy data augmentation which can limit generalization and trust in clinical applications. In this paper, we propose a double-backbone network integrating VGG16 and Xception with a Frequency-Gated Attention (FGA) Block to capture complementary local and global features. Unlike previous studies, our model achieves state-of-the-art performance without augmentation which demonstrates robustness to variably sized and distributed datasets. For further transparency, Grad-CAM is integrated to visualize the tumor regions based on which the model is giving prediction, bridging the gap between model prediction and clinical interpretability. The proposed framework achieves 99.24% accuracy on the 7K-DS dataset for the 4-class setting, along with 98.68% and 99.85% in the 3-class and 2-class settings, respectively. On the independent 3K-DS dataset, the model generalizes with 95.77% accuracy, outperforming baseline and state-of-the-art methods. To further support clinical usability, we developed a graphical user interface (GUI) that provides real-time classification and Grad-CAM-based tumor localization. These findings suggest that augmentation-free, interpretable, and deployable deep learning models such as DB-FGA-Net hold strong potential for reliable clinical translation in brain tumor diagnosis. 1 Introduction In general, a tumor is a clump of cells that grows out of control. Tumors can be classified into two categories: malignant and benign. Malignant tumors are cancerous and affects other parts of the body. On the other hand benign tumors grow slowly and are non-cancerous. When tumors occur in the brain or near it, is called a brain tumor. Brain tumor is one of the worst tumors. It is ranked to be the 5th most common cancer by the ABTA organization [ 4 ] . It causes harm to the brain causing headaches to sensory loss, motor deficits even cognitive impairment [ 1 ] . There are three common types of brain tumor; such as meningioma, glioma and pituitary. Tumors that arise from the meninges are meningiomas. Meninges are the membranes covering the brain and spinal cord. Meningiomas are mostly benign but can be malignant sometime. Glioma tumor occurs in the glial cells which is usually a malignant type of tumor. Pituitary tumor develops in the pituitary gland which is situated in the center of the brain. Pituitary tumors are generally benign. [ 2 ] 1.1 Background and Motivation According to the American Cancer Society, in 2023 approximately 25 thousand malignant brain and spinal cord tumors will be diagnosed in the United States where 14,420 cases in males and 10,980 in females [ 3 ] . More than 12,000 cases of primary brain tumors are recorded annually, including 500 children and young people which equates to 33 people every day [ 5 ] . Early detection of brain tumors has significantly increased the survival rate as well as reduced the need for exploratory surgery to make a diagnosis. MRIs, CT scans, ultrasounds and PET scans are usually used to diagnose brain tumors [ 6 ] . Among them MRI provides better soft tissue contrast making MRI the most popular for brain tumor diagnosis, treatment planning and surgical guidance. [ 7 ] . Despite its advantages, manual diagnosis of brain tumor using MRI scans is labor-intensive. It is not immune to human error either. Widely available CT imaging is useful in emergencies but it suffers from lower soft tissue contrast and exposes patients to ionizing radiation [ 7 ] . Although PET is effective in measuring tumor metabolism it has limited accessibility and on top of that it is expensive. Furthermore, traditional diagnosis heavily rely on invasive biopsy for definitive diagnosis of brain tumor, which is prone to infection, bleeding and complications [ 2 ] . These limitations beg for the urgent need for robust, automated, and non-invasive diagnostic solutions. 1.2 Research Gaps In recent years, at the height of Machine Learning and Artificial Intelligence, the diagnosis of any disease, cancer and tumor has been much easier and time-effective. Not only that, it has significantly increased the detection accuracy [ 8 ] . Recently MIT has successfully developed a deep-learning model that can predict the possibility of developing breast cancer up to five years in advance [ 9 ] . It has recently broken the internet showing the future possibilities of machine learning and artificial intelligence. Unlike traditional diagnostic workflows that rely heavily on manual interpretation of imaging scans, ML models can analyze vast amounts of imaging data efficiently. It is able to detect subtle patterns that humans can perceive. Techniques such as support vector machines (SVMs), random forests, k-nearest neighbors and logistic regression were initially applied to extract handcrafted features such as; texture, intensity, and shape from medical images to assist in tumor diagnosis [ 10 ] . Even with all the progress in using deep learning, there are still some big hurdles in the field for classifying brain tumors. For starters, getting and preparing medical images is tricky and expensive, thus making it hard to build up really good datasets. Class imbalances are another frequent issue, where some tumor types show up way more than others, throwing off how well models perform and generalize [ 11 ] . Overall, there’s a real need for tougher, more accurate systems for classifying brain tumors. Challenges like narrow dataset variety, uneven classes, weak feature extraction in single models, not-so-great accuracy and high resource demands keep popping up. A lot of existing methods lack clear explanations. They act like black boxes, without showing how they make decisions or pinpointing tumors exactly, which is key for doctors to trust and understand them. Many rely too much on data augmentation to fix dataset problems, but that can add fake elements that hurt the model’s true toughness and ability to work on new data. Testing the generalizability of the model across different dataset is often skipped or done poorly. So models shine on their training data but fail on fresh data from elsewhere, limiting their use in the real world. Full evaluations for different levels of complexity, like 2-class, 3-class, or 4-class setups, aren’t common; most stick to simple yes/no classifications that miss the range of tumor varieties. 1.3 Contributions of This Work • We design an architecture that integrates VGG16 and Xception with novel FGA block, combining fine-grained local details and global contextual features to improve tumor classification. • Unlike the most prior works that rely heavily on augmentation, our model achieves state-of-the-art accuracy without any augmentation which reduces preprocessing requirements and highlighting its practical usability in real-world clinical settings. • We validate the performance of our model on an independent dataset (3K-DS), demonstrating the model’s robustness across different data distributions. • We make comparative evaluation against CBAM baselines, showing FGA’s superior accuracy and interpretability through metrics and Grad-CAM visualizations. • We visualize the models prediction using explainable AI, enabling interpretability and clinical trust by showing how the model’s predictions made based on the tumor cells not any irrelevant area of the images. • We design a lightweight graphical user interface (GUI) that integrates classification and Grad-CAM visualization, enabling

Read the complete text