A New Convolutional Neural Network Using Xception Architecture for Automatic Detection of Brain Tumors in Magnetic Resonance Imaging Images

Abstract

Brain diseases are mainly caused by abnormal growth of brain cells that may damagethe brain structure, and eventually will lead to malignant brain cancer and other discussed models in this paper is conducted. The classification of brain tumors from medical imaging is pivotal for accurate medical diagnosis but remains challenging due to the intricate morphologies of tumors and the precision required. Existing methodologies, including manual MRI evaluations and computer-assisted systems, primarily utilize conventional machine learning and pre-trained deep learning models. These systems often suer from overfitting due to modest medical imaging datasets and exhibit limited generalizability on unseen data, alongside substantial computational demands that hinder real-time application. To enhance diagnostic accuracy and reliability, this research introduces an advanced model utilizing the Xception architecture, enriched with additional batch normalization and dropout layers tomitigate overfitting. This model is further refined by leveraging large-scale datathrough transfer learning and employing a customized dense layer setup tailoredto efectively distinguish between meningioma, glioma, and pituitary tumorcategories. This hybrid method not only capitalizes on the strengths of pre-trained network features but also adapts specific training to a targeted dataset,thereby improving the generalization capacity of the model across differentimaging conditions. Demonstrating an important improvement in diagnosticperformance, the proposed model achieves a classification accuracy of 98.039%on the test dataset, with precision and recall rates above 96% for all categories. These results underscore the possibility of the MRI Images model as a reliable diagnostic toolin clinical settings, significantly surpassing existing diagnostic protocols for braintumors.

Introduction

1. Background

Brain tumors are abnormal growths within the brain or spinal canal and are categorized as:

• Primary tumors (originating in the brain).

• Secondary/metastatic tumors (spread from other body parts).

Main tumor types discussed:

• Meningiomas: Arise from the meninges, usually benign and surgically treatable.

• Gliomas: Originate from glial cells, vary in malignancy; high-grade gliomas (like glioblastoma) are aggressive and hard to remove.

• Pituitary tumors: Affect hormone regulation; generally benign but can cause systemic symptoms.

2. Research Motivation

MRI is the standard imaging technique for brain tumors, but traditional diagnosis:

• Is time-consuming,

• Relies on radiologist expertise,

• Is prone to human error and inconsistency.

Goal: Improve accuracy, efficiency, and generalizability of brain tumor diagnosis using deep learning models, particularly Convolutional Neural Networks (CNNs) and the Xception architecture, enhanced by modern techniques like transfer learning and regularization.

3. Research Objectives

1. Develop a model to classify meningioma, glioma, and pituitary tumors.

2. Use techniques like batch normalization and dropout to prevent overfitting.

3. Benchmark the proposed model against existing models like VGG16, VGG19, and CNN-SVM.

4. Promote deep learning adoption in medical diagnostics.

4. Related Work

• Early approaches: Manual image analysis and basic machine learning with limited success.

• CNNs revolutionized imaging by learning complex patterns directly from raw data.

• Deep models (e.g., AlexNet, VGG, GoogLeNet) significantly improved accuracy and automation in medical imaging.

5. Methodology

Model Architecture:

• Based on Xception CNN, enhanced with:

  • Batch normalization for stable training,

  • Dropout to reduce overfitting,

  • L1/L2 regularization for robust feature learning,

  • Softmax output for multi-class classification.

Transfer Learning:

• Starts with a model pre-trained on natural images.

• Fine-tuned using brain MRI scans to adapt to the medical domain.

6. Dataset

• Derived from a publicly available MRI dataset by Jun Chenge.

• Includes T1-weighted contrast-enhanced MRI scans from 299 patients.

• Covers 3 tumor types with hundreds of images each.

• Images were converted from .mat to .png format and resized to 229x229 pixels for uniformity.

7. Data Preprocessing & Augmentation

• Cropping removes irrelevant areas (texts, black borders).

• Normalization scales pixel values to [0,1].

• Augmentation (e.g., flipping, rotating, scaling) simulates real-world imaging variations, increasing model robustness.

8. Deep Learning Enhancements

• Regularization techniques (L1, L2) reduce overfitting by penalizing large weights.

• Dropout simulates multiple networks during training.

• Batch normalization helps stabilize and accelerate training.

• Softmax layer enables classification into 3 tumor categories by converting logits into probabilities.

9. Key Benefits

• Improved diagnostic accuracy and speed.

• Reduced reliance on human interpretation.

• Adaptability across imaging conditions via transfer learning.

• Potential for real-world clinical application with proper validation.

Conclusion

This study developed and validated a deep learning model using the Xception architecture to classify brain tumors from MRI images, demonstrating high accuracy, precision, recall, and AUC scores. The comprehensive methodology encompassed data preprocessing, the application of an advanced convolutional neural network, and rigorous evaluation using diverse metrics, proving the model’s ability to differentiate various types of brain tumors effectively. Looking forward, enhancing the model through the integration of larger and more diverse datasets could improve robustness and accuracy, particularly for complex or rare tumor types. Future work could also explore additional transfer learning strategies and fine-tuning approaches to enhance performance. Collaboration with medical professionals for clinical validation could confirm model’s utility in real-world settings, ensuring compliance with clinical standards. Moreover, incorporating multimodal data, such as genetic information and patient demographics, could offer a more comprehensive diagnostic tool, suggesting that deep learning could significantly enhance diagnostic processes in healthcare, providing tools that support radiologists and contribute to more personalized and precise medical treatments.

References

[1] Kollem S, Reddy KR, Prasad R, Chakraborty A, Ajayan J, Sreejith S, et al. AlexNet- NDTL: classification of MRI brain tumor images using modified AlexNet with deep transfer learning and Lipschitz-based data augmentation. Int J Imag Syst Technol. (2023) 33:1306–22. doi: 10.1002/ima.22870 [2] Ruba T, Tamilselvi R, Beham MP. Brain tumor segmentation in multimodal MRI images using novel LSIS operator and deep learning. J Ambient Intell Humaniz Comput. (2023) 14:13163–77. doi: 10.1007/s12652-022-03773-5. [3] Albalawi E, Thakur A, Ramakrishna TM, Khan BS, SankaraNarayanan S, Almarri B, et al.Oral squamous cell carcinoma detection using EfficientNet on histopathological images. Front Med. (2024) 10:1349336. doi: 10.3389/fmed.2023.1349336 [4] Al Moteri M, Mahesh TR, Thakur A, Vinoth Kumar V, Khan SB, Alojail M. Enhancing accessibility for improved diagnosis with modified EfficientNetV2-S and cyclic learning rate strategy in women with disabilities and breast cancer. Front Med. (2024) 11:1373244. doi: 10.3389/fmed.2024.1373244 [5] Thakur A, Gupta M, Sinha DK, Mishra KK, Venkatesan VK, Guluwadi S. Transformative breast cancer diagnosis using CNNs with optimized ReduceLROnPlateau and early stopping enhancements. Int J Comp Intellig Syst. (2024) 17:14. doi: 10.1007/s44196-023-00397-1 [6] Arledge AC, Crowe NW, Wang L, Bourland DJ, Topaloglu U, Habib AA, et al. Transfer learning approach to vascular permeability changes in brain metastasispost-whole-brain radiotherapy. Cancers. (2023) 15:2703. doi: 10.3390/cancers15102703 [7] Solanki S, Singh U, Chouhan S, Jain S. Brain tumor detection and classification using intelligence techniques: an overview. IEEE Access. (2023) 11:12870– 86. doi: 10.1109/ACCESS.2023.3242666 [8] Özkaraca O, Bagriaçik OI, Gürüler H, Khan F, Hussain J, Khan J, et al. Multiple brain tumor classification with dense CNNarchitecture using brainMRI IMAGES. Life. (2023) 13:349. doi: 10.3390/life13020349 [9] Thomas AW, Lindenberger U, Somek M, Müller KR. Evaluating deep transfer learning for whole-brain cognitive decoding. J Franklin Institute. (2023) 360:9754– 87. doi: 10.1016/j.jfranklin.2023.07.015 [10] Pedada KR, Rao B, Patro K, Allam J, Janjoom M, Samee N. A novel approach for brain tumour detection using deep learning-based technique. Biomed Signal Proc Control. (2023) 82:104549. doi: 10.1016/j.bspc.2022.104549DATASET.

Copyright

Copyright © 2025 M. Saranya, Arunesh A, Kaviya M, Kaviya S. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.