Brain Tumor Detection Using Hybrid Deep Learning Models

Abstract

Advanced methods of the deep learning have been recently applied in case of medical image analysis especially for classification of brain tumours. Nonetheless, many techniques remain liable to overfitting, have a high computational cost, and are not efficient on a small data sample. To address all these challenges, this work proposes a solution that entails using Convolutional Neural Network (CNN) with VGG16 with added extra layers for classification. The model can further go through image pre-processing techniques including normalization, cropping and augmentation thus improving the model performance given that it is intended for four categories: glioma, meningioma, pituitary and no tumor. The data set containing 5,712 brain MRI scans were augmented to contain 17,136 scans in order to improve the class distribution and its resilience. All the MRI images were also preprocessed to a standard size of uniform 224x224 pixels for the training of the model. Finally, using an intuitive Streamlit interface, the proposed model provides real-time tumor detection and classification with an accuracy of 99.79% and an F1 score of 0.98.thus showing the model’s prowess in multiclass tumor classification and outcompeting other general models.

Introduction

A brain tumor is an abnormal growth of cells in the brain that disrupts normal brain function. Brain tumors can be classified as primary (originating in the brain) or secondary (spreading from other parts of the body), and as benign or malignant. Since symptoms are often unclear, diagnosis typically relies on imaging techniques such as MRI and CT scans.

Traditional brain tumor detection methods include MRI, liquid biopsy, and Positron Emission Tomography (PET) scans. PET imaging with specialized tracers has improved tumor detection, boundary identification, and differentiation between recurrent tumors and treatment-related changes. However, these techniques can be expensive, may have limited availability, and can sometimes produce inaccurate results due to non-tumor-related tracer uptake.

To improve diagnosis, machine learning techniques such as Support Vector Machine (SVM), Naïve Bayes, and K-Nearest Neighbor (KNN) have been applied to MRI image analysis. These methods can classify tumors and assist in identifying tumor types such as astrocytoma, glioblastoma, and oligodendroglioma. However, traditional machine learning methods often struggle with complex image variations and large-scale medical data.

Recent advances in deep learning have significantly improved brain tumor detection and classification. Convolutional Neural Networks (CNNs) and advanced architectures such as VGG16, ResNet50, InceptionV3, and 3D Swin Transformer have demonstrated high classification accuracy. Nevertheless, many existing deep learning models suffer from overfitting, high computational costs, and limited ability to classify multiple tumor subtypes.

To address these limitations, the proposed study introduces an improved VGG16-based deep learning model using transfer learning, data augmentation, and regularization techniques. Key contributions include:

1. Refined VGG16 Architecture: Uses transfer learning with frozen early layers, fine-tuning of the last layers, dropout regularization, and resized MRI inputs (128×128) to reduce overfitting and improve generalization.
2. Efficient and Accurate Approach: Combines image normalization, augmentation, and partial fine-tuning to achieve high performance with lower computational requirements than transformer-based models.
3. Multiclass Classification: Classifies MRI scans into four categories—glioma, meningioma, pituitary tumor, and no tumor. Data augmentation expanded the dataset from 5,712 to 17,136 images, improving model robustness and diversity.

The methodology involves collecting MRI datasets, preprocessing images through resizing, normalization, and augmentation, and using the VGG16 convolutional base for feature extraction. Additional layers, including dropout and dense layers, are added for classification. The model is evaluated using metrics such as accuracy, precision, recall, F1-score, and confusion matrices, while hyperparameter tuning is performed to optimize performance.

Conclusion

Implementation of a CNN model with features learnt from a pre-trained network for multiple types of brain tumors for the identification of IDH1 mutation is very promising for the application of automated medical imaging. In this research, the proposed model has reached the desired level of success, achieving a classification accuracy of 99.79% that makes it suitable for differentiating glioma, meningioma, pituitary and no tumor cases. The result of the pre-processing that were applied to the MRI scans include resizing, normalization, cropping and augmentation of the images used in the construction of the model increased its efficiency. Besides, the ability to make predictions in a real-time manner through the application built on Streamlit makes the model quite useful for physicians in clinical settings. This reaffirms the importance of the model in the improvement of definitional accuracy of diagnosis and decision-making in the context of neuro-oncology. As the future work, it would be interesting to use other pre-trained architectures, extend the usage of the model to a greater and a range of datasets, and introduce segmentation functionality. They would also advance the practical use of AI in diagnosing brain tumors and significantly contribute to the field.

References

[1] Abd El Kader Isselmou, Shuai Zhang, Guizhi Xu, “A Novel Approach for Brain Tumor Detection Using MRI Images”, 2021. [2] Paul M Brennan , Holly J Butler , Loren Christie , Mark G Hegarty , Michael D Jenkinson , Catriona Keerie , John Norrie , Rachel O’Brien , David S Palmer , Benjamin R Smith , Matthew J Baker, “Early diagnosis of brain tumours using a novel spectroscopic liquid biopsy”, 2021. [3] Ji-hoon Jung, Byeong-Cheol Ahn, “Current Radiopharmaceuticals for Positron Emission Tomography of Brain Tumors”, 2018. [4] Sanjeev Kumar, Chetna Dabas2, Sunila Godara, “Classification of Brain MRI Tumor Images: A Hybrid Approach”,2017. [5] Divyamary.D, Gopika.S, Pradeeba.S, Bhuvaneswari.M, “Brain Tumor Detection from MRI Images using Naïve Classifier”, 2020. [6] Rafi Haidar Ramdlon , Entin Martiana Kusumaningtyas , Tita Karlita ,“BRAIN TUMOR CLASSIFICATION USING MRI IMAGES WITH K-NEAREST NEIGHBOR METHOD”,2019. [7] Hossam H. Sultan, Nancy M. Salem, Walid Al-Atabany,“Multi-Classification of Brain Tumor Images Using Deep Neural Network”,2019. [8] Md Ishtyaq Mahmud, Muntasir Mamun, Ahmed Abdelgawad, “A Deep Analysis of Brain Tumor Detection from MR Images Using Deep Learning Networks”, 2023. [9] K. Kavin Kumar, P. M. Dinesh, P. Rayavel, L. Vijayaraja, R. Dhanasekar, Rupa Kesavan, Kannadasan Raju, Arfat Ahmad Khan, Chitapong Wechtaisong, Mohd Anul Haq, Zamil S. Alzamil and Ahmed Alhussen, “Brain Tumor Identification Using Data Augmentation and Transfer Learning Approach”, 2023. [10] Jiang, Yuan Zhang, Xin Lin, Jinkun Dong, Tongtong Cheng and Jing Liang, “SwinBTS: A Method for 3D Multimodal Brain Tumor Segmentation Using Swin Transformer”, 2022.

Copyright

Copyright © 2026 Abhijit Singh, Nakul Jain, Rishabh Singh, Barkha Bhardwaj. 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.