Deep Learning-Based Brain Tumor Detection and Classification from MRI Images with Explainable AI and GIS Spatial Analysis: A Study in the Odisha Healthcare Context

Abstract

Brain tumors represent one of the most severe neurological disorders, demanding early and precise diagnosis for effective clinical management. Manual interpretation of Magnetic Resonance Imaging (MRI) scans is time-consuming, subject to inter-observer variability, and highly dependent on specialized radiologist expertise — resources that remain scarce across large parts of Odisha, India. This paper presents a comprehensive, end-to-end deep learning framework for automatic brain tumor segmentation, classification, and explainability, with integrated Geographic Information System (GIS) spatial analysis of tumor incidence patterns across Odisha\'s 30 districts. The segmentation component applies and comparatively evaluates four unsupervised clustering algorithms — K-Means, Self-Organizing Maps (SOM), Hierarchical Clustering, and Fuzzy C-Means (FCM) — on MRI images converted to CIELab color space, establishing FCM as the optimal method (pixel accuracy: 94.8%, Dice score: 0.928). The classification component employs transfer learning with VGG-16 pre-trained on ImageNet, achieving 97.8% macro-averaged accuracy across four clinically relevant tumor classes (Glioma, Meningioma, Pituitary Tumor, No Tumor). Explainable AI via Grad-CAM generates visual heatmaps of classification decisions, clinically validated by radiologists who rated 90%+ of outputs as anatomically consistent. GIS spatial analysis reveals statistically significant geographic clustering of tumor incidence (Moran\'s I = 0.312, p < 0.05), identifying underserved southern tribal districts with likely unmet diagnostic need. The integrated framework is deployed as a web-based clinical decision support application suitable for Odisha\'s heterogeneous healthcare infrastructure.

Introduction

Brain tumors are abnormal growths in the brain that range from benign tumors to highly aggressive cancers such as glioblastoma. Diagnosing brain tumors using MRI scans is challenging because MRI interpretation is time-consuming, requires expert radiologists, and often suffers from variability in manual tumor identification. In regions like Odisha, India, limited access to specialized neuroimaging services further complicates timely diagnosis. To address these challenges, this study proposes an AI-based clinical decision support system that combines MRI preprocessing, clustering-based tumor segmentation, deep learning classification, Explainable AI (Grad-CAM), and GIS-based spatial analysis.

The study aims to evaluate clustering algorithms (K-Means, Self-Organizing Maps, Hierarchical Clustering, and Fuzzy C-Means) for tumor segmentation, develop a VGG-16 transfer learning model for classifying MRI scans into four categories (Glioma, Meningioma, Pituitary Tumor, and No Tumor), integrate Grad-CAM for visual explanations, analyze tumor incidence across Odisha using GIS tools, and deploy the complete framework as a web-based application.

A dataset of 7,023 MRI images, supplemented with 412 anonymized cases from Odisha hospitals, was used. Images underwent preprocessing steps including noise reduction, skull stripping, normalization, and data augmentation. Tumor segmentation was performed using clustering algorithms in the CIELab color space, while classification was carried out using a fine-tuned VGG-16 convolutional neural network. Grad-CAM generated heatmaps highlighting regions responsible for classification decisions, improving interpretability and clinician trust. GIS analysis mapped 1,247 brain tumor cases across Odisha districts to identify spatial patterns and healthcare disparities.

Results showed that Fuzzy C-Means (FCM) achieved the best segmentation performance with 94.8% pixel accuracy, 87.3% IoU, and a Dice score of 0.928, while K-Means provided faster execution with slightly lower accuracy. The CNN classifier achieved 97.8% overall accuracy, with high precision, recall, and F1-scores across all tumor classes. On Odisha-specific cases, accuracy remained high at 96.9%, demonstrating strong regional applicability. Grad-CAM explanations aligned with tumor regions in over 91% of cases, and radiologists found them clinically consistent. GIS analysis revealed significant geographic variation in tumor incidence, highlighting areas with potential under-diagnosis and unmet healthcare needs.

Conclusion

This paper presented a comprehensive deep learning-based framework for automatic brain tumor detection, segmentation, classification, and explainability, with integrated GIS spatial analysis validated for the Odisha healthcare context. The principal findings are: • FCM outperforms K-Means, SOM, and Hierarchical Clustering for MRI tumor segmentation (94.8% pixel accuracy, 0.928 Dice score), validating soft-partitioning approaches for inherently fuzzy tissue boundaries in medical imaging. • VGG-16 transfer learning with two-phase fine-tuning achieves 97.8% macro-averaged four-class classification accuracy, representing state-of-the-art performance with region-specific validation on Odisha hospital data. • Grad-CAM provides clinically validated visual explanations, rated anatomically consistent in over 90% of cases by independent radiologists, enhancing clinical trust and adoptability. • GIS spatial analysis reveals statistically significant geographic inequity in brain tumor diagnostic access across Odisha (Moran\'s I = 0.312, p < 0.05), identifying southern tribal districts with likely substantial unmet diagnostic need — findings with direct policy implications for healthcare resource allocation and telemedicine investment. Important limitations include reliance on predominantly international training data, 2D slice analysis rather than full 3D volumetric processing, and the absence of prospective clinical validation. Future work will address these limitations through 3D CNN and transformer-based architectures (TransUNet, Swin-UNETR), multimodal T1/T2/FLAIR fusion, federated learning across Odisha hospitals for privacy-preserving multi-site training, mobile-optimized deployment for remote healthcare facilities, and longitudinal treatment monitoring.

References

[1] S.R. Kannan, Jan. 2005. \"Segmentation of MRI using new Unsupervised Fuzzy C-Means Algorithm.\" ICGST-GVIP Journal, Vol. 5, Issue 1, pp. 45–52. [2] J. Horvath, 2006. \"Image Segmentation using Fuzzy C-Means.\" SAMI 2006 — 4th Slovakian-Hungarian Joint Symposium on Applied Machine Intelligence. [3] L. Jiang, W. Yang, 2003. \"A Modified Fuzzy C-Means Algorithm for Magnetic Resonance Images.\" Proc. VII Digital Image Computing: Techniques and Applications, Sydney. [4] C. Reyes-Aldasoro, 2004. \"Image Segmentation with Kohonen Neural Network SOM.\" CIIST 2004. [5] C. Ambroise, F. Badran, S. Thiria, G. Seze, 2000. \"Hierarchical Clustering of SOM for Cloud Classification.\" NeuroComputing, Vol. 30, pp. 47–52. [6] K.S. Fu and J.K. Mui, 1981. \"A Survey on Image Segmentation.\" Pattern Recognition, Vol. 13, pp. 3–16. [7] R.C. Gonzalez and R.E. Woods, 2002. Digital Image Processing. Pearson Education, 2nd Ed. [8] R.M. Haralick and L.G. Shapiro, 1985. \"Survey: Image Segmentation Techniques.\" Computer Vision, Graphics and Image Processing, Vol. 29, pp. 100–132. [9] J. Goldberger, S. Gordon, H. Greenspan, 2006. \"Unsupervised Image-Set Clustering Using an Information Theoretic Framework.\" IEEE Trans. Image Processing, Vol. 15, No. 2, pp. 449–458. [10] N. Pal and S. Pal, 1993. \"A Review on Image Segmentation Techniques.\" Pattern Recognition, Vol. 26, pp. 1277–1294. [11] P. Tamije Selvy, V. Palanisamy, T. Purusothaman, 2011. \"Performance Analysis of Clustering Algorithms in Brain Tumor Detection of MR Images.\" J. Computer Science, Vol. 7, No. 8, pp. 1174–1181. [12] R.R. Selvaraju, M. Cogswell, A. Das, R. Vedantam, D. Parikh, D. Batra, 2017. \"Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localization.\" Proc. ICCV, pp. 618–626. [13] B. Zhou, A. Khosla, A. Lapedriza, A. Oliva, A. Torralba, 2016. \"Learning Deep Features for Discriminative Localization.\" Proc. CVPR, pp. 2921–2929. [14] [M. Sajjad, S. Khan, K. Muhammad, W. Wu, A. Ullah, S.W. Baik, 2019. \"Multi-Grade Brain Tumor Classification using Deep CNN with Extensive Data Augmentation.\" J. Computational Science, Vol. 30, pp. 174–182. [15] Y. LeCun, Y. Bengio, G. Hinton, 2015. \"Deep Learning.\" Nature, Vol. 521, pp. 436–444. [16] K. Simonyan, A. Zisserman, 2014. \"Very Deep Convolutional Networks for Large-Scale Image Recognition.\" arXiv preprint arXiv:1409.1556. [17] B.H. Menze et al., 2015. \"The Multimodal Brain Tumor Image Segmentation Benchmark (BRATS).\" IEEE Trans. Medical Imaging, Vol. 34, No. 10, pp. 1993–2024. [18] P. Afshar, A. Mohammadi, K.N. Plataniotis, 2019. \"Brain Tumor Type Classification via Capsule Networks.\" Proc. 25th IEEE ICIP, pp. 3129–3133. [19] World Health Organization (WHO), 2021. Classification of Tumours of the Central Nervous System, 5th Edition. IARC Press: Lyon, France. [20] Indian Council of Medical Research (ICMR), 2022. National Cancer Registry Programme Report. ICMR: New Delhi, India.

Copyright

Copyright © 2026 Madhumita Swain, Dr. Brahamara Bar Biswal. 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.