Brain Tumor Detection and Classification Using MRI and CT Image Fusion

Jonnalagadda Cynthia Smiley , Rajdev Preselvan, Kakani Vamsidhar, Malin Bruntha and Kesava Reddy
Department of ECE, Karunya Institute of Technology and Sciences, Coimbatore, Tamil Nadu, India 
Correspondence to: Jonnalagadda Cynthia Smiley, cynthiajonnalagadda0283@gmail.com

DOI: https://doi.org/10.70389/PJS.100229

Additional information

• Ethical approval: N/a
• Consent: N/a
• Funding: No industry funding
• Conflicts of interest: N/a
• Author contribution: Jonnalagadda Cynthia Smiley, Rajdev Preselvan, Kakani Vamsidhar, Malin Bruntha and Kesava Reddy – Conceptualization, Writing – original draft, review and editing
• Guarantor: Jonnalagadda Cynthia Smiley
• Provenance and peer-review: Unsolicited and externally peer-reviewed
• Data availability statement: N/a

Keywords: Brain tumor classification, MRI-CT image fusion, Customized CNN, Fisher-yates shuffle algorithm, VGG16-based detection.

Peer Review
Received: 14 August 2025
Last revised: 22 October 2025
Accepted: 17 December 2025
Version accepted: 3
Published: 8 January 2026

Plain Language Summary Infographic

Introduction

Cancer is one of the most aggressive and life-threatening diseases, with brain tumors contributing significantly to cancer-related mortality.¹ Some of the most frequently diagnosed brain tumors are Meningiomas, Pituitary and Gliomas. Although these tumors are diagnosed, there have been significant challenges involved in their detection, classification, and treatment.² It is best understood that timely or early detection and its accurate classification are crucial for improving the rate of survival. Moreover, personalized treatment is also necessary for improving the survival rate. Brain tumor detection relies significantly on medical imaging technologies magnetic resonance imaging (MRI). MRI provides insights into the tumour’s boundaries and internal structures.^3–8 It has exceptional soft tissue contrast capability.⁹ In the traditional diagnostic techniques for brain tumors, the medical images are manually inspected by the radiologists, which is often a time-consuming process and highly prone to human error, and often requires a decision support tool. Furthermore, it is heavily dependent on the expertise of the clinician.^10,11 The push for more accurate brain tumor classification has increased the demand for automated and efficient classification models. Convolutional Neural Networks provides high accuracy for feature extraction and classification.¹²

The MRI dataset is required for brain tumor classification. The customized convolutional neural network (CNN) model is trained on the MRI dataset which predicts the type of brain tumor. We also compare the evaluation metrics with different pre-trained architectures such as VGG16 and ResNet. Our customized CNN model is developed to classify brain tumors into Glioma, Meningioma, Pituitary, or No Tumor categories. Recent advancements in AI and medical imaging have fostered this research. The availability of large-scale brain tumor datasets has made this research possible with high efficiency.¹³ Traditional Machine-learning models rely on feature extraction. However, deep learning enables automation and efficient feature learning from medical images.^14–17

Key Contributions to this Research

The Customized CNN model classifies the type of brain tumors into four categories, namely meningioma, glioma, pituitary, and no tumor. The customized CNN model’s performance is evaluated using metrics like accuracy, precision, recall, and F1 score. The diagnostic reliability is significantly improved by using the customized CNN model.

Proposed Methodology

The four key stages in brain tumor classification using a customized CNN are data collection, image preprocessing, feature extraction and tumor classification. Each of these stages is described in detail below.

Data Collection

The data is collected from Kaggle, licensed under CC0 Public Domain. We have chosen 7,023 grayscale MRI images. Figure 1 displays a sample of MRI image of brain. In our dataset, 70% is used for training and 10% for testing 20% for validation, Specifically, the number of MRI images used are 4,917 for training, 702 for testing, and 1,404 for validation, respectively. The dataset is categorized into Glioma, Meningioma, Pituitary, or No Tumor. The average original dimensions of these images were 232 × 277. The images are in JPEG format and have varying resolutions. They have been resized and standardised to a spatial dimension of 224 × 224 pixels. The MRI images identify the soft tissue variations and predict the type of brain tumor.

Image Preprocessing

In image preprocessing, we focus on improving the image quality and maintaining consistency while reducing irregularities. The following preprocessing steps are applied. Image Augmentation techniques used are rotation, width and height shift, shear transformation, which uses a shear range of x to distort the images, zooming, horizontal flipping, resizing to standardise image dimensions for uniformity, and Normalization where the values for pixel intensity are scaled between 0 and 1. We employed augmentation techniques to increase the number of training.

Feature Extraction

The dataset of MRI images is used for brain tumor prediction using the customized CNN model. This model automatically learns hierarchical features for tumor classification. The architecture of the customized CNN is shown in Figure 2. Figure 2 comprises four Conv2D layers with varying filter sizes. We can see that three pooling layers are used after each Conv2D layer, and a single flatten layer is used. The number of dense layers used is three, along with a single dropout layer. Each layer in the customized CNN model is as follows.

• Convolutional layers – The model has four Conv2D layers with 32, 64, 128, and 256 filters, respectively.
• Pooling Layers – Three pooling layers are used which are applied after each convolution layer (Conv2D).
• Flatten layer – A single flattened vector is created using the feature maps.
• Fully Connected Layers (Dense Layers) – The model consists of three dense layers, each with 512, 256, and 128 neurons. The hidden layers use ReLU activation, while the output layer uses softmax for brain tumor classification.
• Dropout – A single dropout layer is used with a rate of 50% to prevent overfitting.

The customized CNN model is inspired by the VGG16 model. The Conv2D layers and dropout rate were chosen manually by iteratively changing the values. The optimised number of layers consist of four Conv2D layers and a single dropout layer was determined because of better generalisation capability by trial and testing.

Brain Tumor Classification

The Customized CNN performs brain tumor prediction. It classifies the tumor into either of these categories Glioma (G), Meningioma (M), Pituitary (P), or No Tumor (N) and type corresponding tumor type is printed based on the MRI image. The training setup includes a detailed procedure in Figure 3. As seen in Figure 3, the MRI data is loaded it is then preprocessed. The train-test-validation split is performed on the dataset and data generators are obtained from MRI. The customized CNN model is built and trained with softmax activation. It performs brain tumor prediction using the MRI dataset. Its performance is evaluated and multi class prediction is performed which outputs the tumor type. Figure 4 outlines the workflow of brain tumor prediction using the customised CNN model. We use cross-entropy loss as the loss function. Adam optimizer is used for training with a batch size of 32, the learning rate is 0.001. The model is then trained over 50 epochs.

Results and Discussion

The customized CNN model is implemented using Python and TensorFlow. The computational efficiency is improved by utilizing Graphics Processing Unit (GPU) acceleration. In our dataset, 70% is used for training and 10% for testing 20% for validation. A balanced evaluation is performed to improve the model.

Dataset Details

The dataset contains MRI images for brain tumor prediction. The total number of MRI images are 7,023. We have used 4,917 MRI images for training, 702 for testing and 1,404 for validation. The MRI images are categorized as meningioma, pituitary, glioma, and no tumor. The greyscale images are in JPEG format. The average original dimensions are 232 × 277 which have been resized to 224 × 224.

Model Performance and Evaluation

The customized CNN’s performance model is compared to other pre-trained architectures like VGG16 and ResNet. Table 1 details the performance analysis of the Customized CNN, VGG16, and ResNet. The evaluation is done by using key classification metrics such as:

• Accuracy – measures the classification performance.
• Precision – determines the positively predicted tumors that are correct.
• Recall – The model’s ability to identify tumors correctly assessed.
• F1 score – The balance between precision and recall is represented.

Based on typical training behaviour and the structure of the notebook, the estimated total training and validation time for 20 epochs is approximately 200 seconds which is 3.33 minutes.

Table 1: Performance analysis of models.
Model	Accuracy	Precision	Recall	F1 Score
Customized CNN	91.2	94	93.5	90
VGG16	77	79	75.5	76
ResNet	59	60.2	58	57

Comparative Analysis of Existing and Proposed Methodologies

Lightweight Model – The large hybrid models in the existing Methods are replaced with our Lightweight customized CNN model.

Conclusion

A deep learning-based customized CNN multimodal is presented. The brain tumor classification is performed using MRI dataset by using a customized CNN model. The MRI provides soft tissue contrast. By leveraging the customized CNN for brain tumor prediction, high diagnostic accuracy is obtained to classify brain tumors. Our proposed model has high accuracy and it is suitable for clinical applications. The computation performance is efficient. Reliable classifications are achieved while maintaining low processing time per image. The processing time per image in the customized CNN model is 2 ms. The diagnostic uncertainty is significantly decreased. However, there can be dataset bias as the dataset is sourced from Kaggle and lacks demographic diversity. Deployment in a clinical setting requires high computational resources. Furthermore, the patient’s consent to share their medical data has to be considered with strict adherence to privacy and security regulations. The brain tumor prediction model can be used as a decision support tool rather than replacing the already existing medical methodologies. It is helpful for radiologists in clinical decision-making. The Customized CNN model significantly improves brain tumor classification accuracy.

References

1. Swati ZNK, Zhao Q, Kabir M, Ali F, Ali Z, Ahmed S, et al. Brain tumor classification for MR images using transfer learning and fine-tuning. Comput Med Imaging Graph. 2019;75:34–46. https://doi.org/10.1016/j.compmedimag.2019.05.001
2. Bakas S, Reyes M, Jakab A, Bauer S, Rempfler M, Crimi A, et al. Identifying the best machine learning algorithms for brain tumor segmentation, progression assessment, and overall survival prediction in the BRATS challenge; 2018. arXiv preprint arXiv:1811.02629.https://doi.org/10.48550/arXiv.1811.02629
3. Baid U, Ghodasara S, Mohan S, Bilello M, Calabrese E, Colak E, et al. The RSNA-ASNR-MICCAI BraTS 2021 benchmark on brain tumor segmentation and radiogenomic classification; 2021. arXiv preprint arXiv:2107.02314.https://doi.org/10.48550/arXiv.2107.02314
4. Rasool N, Bhat JI. A critical review on segmentation of glioma brain tumor and prediction of overall survival. Arch Comput Methods Eng.2025;32(3):1525–69. https://doi.org/10.1007/s11831-024-10188-2
5. Wang W, Chen C, Ding M, Li J, Yu H, Zha S. TransBTS: multimodal brain tumor segmentation using transformer. In: International conference on medical image computing and computer-assisted intervention. Cham: Springer International Publishing; 2021. https://doi.org/10.48550/arXiv.2103.04430
6. Amin J, Sharif M, Yasmin M, Fernandes SL. A distinctive approach in brain tumor detection and classification using MRI. Pat Recog Lett. 2020;139:118–27. https://doi.org/10.1016/j.patrec.2017.10.036
7. Campanacci M. Bone and soft tissue tumors: clinical features, imaging, pathology and treatment. New York, NY, USA: Springer Science & Business Media; 2013. https://doi.org/10.1007/978-3-7091-3846-5
8. Bauer S, Wiest R, Nolte LP, Reyes M. A survey of MRI-based medical image analysis for brain tumor studies. Phys Med Biol. 2013;58(13):R97. https://doi.org/10.1088/0031-9155/58/13/R97
9. Khan MSI, Rahman A, Debnath T, Karim MR, Nasir MK, Band SS, et al. Accurate brain tumor detection using deep convolutional neural network. Comput Struct Biotechnol J. 2022;20:4733–45.https://doi.org/10.1016/j.csbj.2022.08.039
10. Abiwinanda N, Hanif M, Tafwida Hesaputra S, Handayani A, Mengko TR. Brain tumor classification using convolutional neural network. In: World congress on medical physics and biomedical engineering 2018. Vol. 1. Singapore: Springer Nature Singapore; 2018. p. 3–8.
11. Rehman A, Khan MA, Saba T, Mehmood Z, Tariq U, Ayesha N. Microscopic brain tumor detection and classification using 3D CNN and feature selection architecture. Micro Res Tech. 2021;84(1):133–49. http://doi.org/10.1002/jemt.23597
12. Cè M, Irmici G, Foschini C, Danesini GM, Falsitta LV, Serio ML, et al. Artificial intelligence in brain tumor imaging: a step toward personalized medicine. Curr Oncol. 2023;30(3):2673–701. http://doi.org/10.3390/curroncol30030203
13. Chahal PK, Pandey S, Goel S. A survey on brain tumor detection techniques for MR images. Multimedia Tools Appl. 2020;79(29):21771–814.http://doi.org/10.1007/s11042-020-08898-3
14. Saeedi S, Rezayi S, Keshavarz H, Niakan Kalhori SR. MRI-based brain tumor detection using convolutional deep learning methods and chosen machine learning techniques. BMC Med Inform Decis Making. 2023;23(1):16. http://doi.org/10.1186/s12911-023-02114-6
15. Pereira S, Pinto A, Alves V, Silva CA. Brain tumor segmentation using convolutional neural networks in MRI images. IEEE Trans Med Imaging. 2016;35(5):1240–51.http://doi.org/10.1109/TMI.2016.2538465
16. Asif S, Yi W, Ain QU, Hou J, Yi T, Si J. Improving effectiveness of different deep transfer learning-based models for detecting brain tumors from MR images. IEEE Access. 2022;10:34716–30. http://doi.org/10.1109/ACCESS.2022.3153306
17. Mahmud MI, Mamun M, Abdelgawad A. A deep analysis of brain tumor detection from MR images using deep learning networks. Algorithms. 2023;16(4):176. https://doi.org/10.3390/a16040176.

Cite this article as:
Cynthia SJ, Rajdev PK, Vamsidhar K, Malin BP and Kesava R. Brain Tumor Detection and Classification Using MRI and CT Image Fusion. Premier Journal of Science 2025;15:100229

Export to EndNote Export to Mendeley