Tumor Detection Using RestNet50

Abstract

Brain tumor’s are major threat to human health and life. It is better for early detection and treatment the tumor’s to enhance survival rates. Traditional methods of detecting and categorizing brain tumor, such as manual segmentation and feature extraction, they are both time consuming and prone to inaccuracies. Early diagnosis of brain tumors plays an important role in a patient’s treatment and makes it easy to save his/her life. The conventional method of manually detecting brain tumors from brain magnetic resonance imaging (MRI) scans can be problematic and erroneous. This study aims to compare Restnets50 performance on publicity available Brats dataset, an exclusive collection of brain tumor which is divided into 4 categories no_tumor , glioma, pituitary, meningioma. Trained on a dataset of 7023 images, the model achieves exceptional accuracy 98% in both training and validation datasets, with a focus on precision. Leveraging techniques such as data augmentation, transfer learning with ResNet50, and regularization ensuresstabilityandgeneralizability.Magnetic resonance imaging (MRI) is a well-known used imaging technique to detect brain tumor . Thisstudydemonstratesthepotentialofdeeplearninginearlybraintumor diagnosis, surpassing conventional methods and laying a robust foundation for future research in neural network-based classification algorithms for brain tumor.

Introduction

1. Introduction

A brain tumor is an abnormal mass of cancerous cells in the brain, which can severely impair brain function and lead to death. Early and accurate detection is vital for improving patient outcomes. With over 120 types of brain tumors classified by the WHO into four grades of malignancy, common symptoms include headaches, seizures, vision issues, vomiting, memory loss, and balance problems.

Traditional diagnosis using MRI or CT scans is time-consuming, prone to human error, and depends heavily on radiologist expertise. To overcome these challenges, deep learning—particularly Convolutional Neural Networks (CNNs)—is being used to automate tumor detection from MRI images.

2. Related Work

• Ramanagiri used ResNet-50, achieving 90.04% accuracy by leveraging deep residual learning for better classification.

• Dipu et al. employed YOLO V3–V5 models for object detection, with YOLO V5 reaching 95.07% mAP, though it required more manual data preparation.

• Sinha focused on image segmentation with traditional thresholding methods, which were less robust compared to deep learning models.

Comparison: ResNet-50 outperforms YOLO and traditional CNNs in accuracy, robustness, and feature extraction, making it more suitable for clinical applications.

3. Proposed Method

The study proposes a ResNet-50-based deep learning system to classify brain tumors as benign or malignant using MRI scans.

Key Steps:

• Data Preprocessing: MRI images are resized (224x224), normalized, and augmented (rotation, flipping, scaling) to increase robustness and handle class imbalance.

• Model Architecture:

  • Base model: Pre-trained ResNet-50.

  • Replaced top layers with:

    • Global Average Pooling

    • Dense layers with ReLU

    • Batch Normalization

    • Dropout (0.5)

    • Softmax for classification

  • Optimizer: Adam

  • Loss function: Categorical Cross-Entropy

• Training and Optimization:

  • EarlyStopping and ReduceLROnPlateau callbacks used.

  • Data split into training and validation sets for performance monitoring.

• Model Evaluation:

  • Evaluated using accuracy, confusion matrix, and ROC curves.

  • Achieved 98% classification accuracy on a dataset of 7,023 MRI images.

4. System Performance

• High accuracy (98%) and strong performance across all tumor categories.

• Efficient classification with low false positives and fast processing, suitable for real-world clinical deployment.

5. Future Scope

Planned improvements include:

• Multi-class tumor classification

• Integration of Explainable AI (XAI) for transparency

• Web-based platforms and hospital system integration

• Application expansion to other diseases and estimation of treatment parameters

Conclusion

The ResNet-50-based model for brain tumor classification achieved 98% accuracy on a 7,023-image Kaggle MRI dataset, demonstrating high reliability in distinguishing benign and malignant tumors. The study emphasizes early detection for timely treatment and evaluates various segmentation algorithms, identifying multilevel thresholding and OTSU thresholding as the most effective. A modified CNN approach significantlyimproved classification accuracy

References

[1] A.Ramanagiri, M.Mukunthan, Dr.G.Balamurugan, “Enhanced Brain Tumor Detection using Rest NET 50. 2024 10th International Conference on Communication and Signal Processing (ICCSP) DOI: 10.1109/ICCSP60870.2024 [2] Nadim Mahmud Dipu ,Sifatul Alam Shohan , and K. M. A Salam, “Brain Tumor Detection Using Various Deep Learning Algorithms”, 2021 International conference on science and contemporary technologies 5-7 Aug 2021 [3] Avigyan Sinha @, Aneesh R P $, Malavika Suresh $, Nitha Mohan R*, Abinaya D*, Ashwin G Singerji*, “Brain Tumour Detection Using Deep Learning” , 2021 Seventh International conference on Bio Signals, Images, and Instrumentation . [4] Shivansh Rathore, Tanishka Rana, Utkarsh Mittal, Tejasvi Gupta, Nahid Malik, Mohammed Shariff A S, “Brain Tumor Detection by using ResNet-50 and Image Processing Tools”, 2023 3rd International Conference on Advance Computing and Innovative Technologies in Engineering [5] Toufique A. Soomro Ahmed Ali , Member, IEEE, Lihong Zheng , Member, IEEE, Shafiullah Soomro , Member, IEEE, Ahmed J. Afifi , Ming Yin , , Member, IEEE, and Junbin Gao, “Image Segmentation for MR Brain Tumor Detection Using Machine Learning: A Review”, 1937-3333 © 2022 IEEE.

Copyright

Copyright © 2025 Dr. R. Rashmi, Pooja Zawar, Aashishkumar Bhagat, Chaitanya Sali, Kaiwalya Alkari. 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.