Brain Tumor Analysis and Prediction from MRI & CT Scan Data

Abstract

Now adays,aneedofhavinganautomatedsystem for brain tumors analysis is growing for rapid examination of various reports such as MRI and CT scans. In today age, considering the progress in field of AI and Machine Learning especially deep learning, there is a hope for advancements of neurology and understanding of human brain. This technology allows us to segment, classify and give data driven diagnostics decisions.Acarefullyanalyzethescansandmapthemaccording tostatusofAI.Thenthekeypointsareanalyzedassegmentation against deep learning, classification of tumors against machine learning and multi-modal fusion techniques by discussing methods adopted, result obtained, and limitations thereof. Integration ofAI and ML in the field of medical imaging may lead to better accuracy in diagnostics and can help radiologists to be more productive and provide better assistance in decision making. However challenges are there that we must address: data variety, the need of supervised dataset and the underlying complexityofmeringdifferentdatasetstogettoconclusion.In this paper we have summarized the current finding from the research, identified gaps in existing tech and proposed further research directions to improve the integration ofAI and MLin brain tumor analysis.

Introduction

The document reviews the use of Artificial Intelligence (AI) and Computer Vision for brain tumour detection and segmentation, focusing on improving accuracy, efficiency, and clinical decision support through advanced deep learning techniques.

It explains that traditional medical diagnosis can be slow and highly dependent on human expertise, while AI systems—especially those using CNNs, attention mechanisms, and multimodal learning—can analyze MRI and CT scans more effectively. These models help detect tumours, handle complex cases, and potentially support personalized treatment planning.

The proposed approach introduces a multimodal deep learning framework that combines MRI and CT data using a multi-encoder architecture:

• MRI is processed using nnU-Net, which automatically optimizes segmentation.
• CT scans are processed using a 3D CNN (ResNet-based model).
• A cross-attention fusion module integrates features from both modalities, focusing on the most relevant tumor-related information.
• A decoder then generates the final segmentation mask or classification output.

The system is designed for real-time clinical use, allowing immediate analysis of new scans.

The methodology includes:

• Data preprocessing (registration, normalization, resampling)
• Training with loss functions like Dice loss + Cross-Entropy
• Evaluation using metrics such as Dice Similarity Coefficient (DSC)
• Implementation using frameworks like PyTorch, MONAI, SimpleITK, and nnU-Net, trained on high-performance GPUs.

Two tasks are developed:

1. Classification model (CNN and ResNet50) to identify tumour types (Glioma, Meningioma, Pituitary, No Tumour), with ResNet50 achieving higher accuracy (~94%).
2. Segmentation model (U-Net) to locate tumour regions in MRI scans, producing accurate tumour masks.

Key challenges include handling heterogeneous medical imaging data, aligning MRI and CT scans, and ensuring reliable multimodal fusion. The solution is a structured, reproducible pipeline that improves robustness and interpretability.

Conclusion

The current study shows an evident brain work plan.Analysis of tumor multimodal imaging and is not constrained by the restrictionsofsingle-modalityanalysis. Ournewfusionmodel is a combination of MRI and CT scans, which is a big step in making a development on the creation of a better comprehensive and helpful diagnosis instrument.The key part of our tasks will consist of designing a multi-encoder framework that makes use of a cross attention process and its integration of the different images that does not merely concatenate as the features do. Our model is able to learn complexrelationshiphencereturnsbettersegmentationswhich are more reliable. In addition, from the robust self-optimizing framework of nnU-Net, this work proposes a high-quality pipelinethatreducestheneedformanualtuningandmaximizes speed and efficiency in moving the research forward. The successful deployment of such a system implies the capacity for deep learning architectures, when designed ingeniously, to change the landscape of medical imaging. Moreover, with single view assessment evolving to a more comprehensive multimodalassessmentlaysthegroundworkformoreaccurate diagnoses, better patient outcomes, and the advent of next-generation intelligent clinical decision support systems in neuro-oncology.

References

[1] O. Ronneberger, P. Fischer, and T. Brox wrote a paper titled \'U-Net: Convolutional Networks for Biomedical ImageSegmentation,\'whichwaspresentedattheMedical Image Computing and Computer-Assisted Intervention (MICCAI) conference in 2015. [2] F.Isensee,P.F.Jaeger,S.A.A.Kohl,J.Petersen,andK. H. Maier-Hein discussed in their paper, \"nnU-Net: a selfconfiguring method for deep learning-based biomedical image segmentation,\" published in Nature Methods, volume 18, issue 2, pages 203–211 in 2021. [3] M. Havaei,A. Davy,D.Warde-Farley, A. Biard,A. Courville,Y.Bengio,C.Pal,P.Jodoin,andH.Larochelle, \"Braintumor Segmentation with Deep Neural Networks,\"Medical Image Analysis, vol. 35, pp. 18-31, 2017. [4] W.Wang,C.Chen,Y.Ding,J.Yu,C.P.Yu,andD.Zhang, \"TransBTS:MultimodalBraintumorSegmentationUsing Transformer,\" in Medical Image Computing and Computer-Assisted Intervention (MICCAI), 2021. [5] Kaggle Dataset: Brain Tumour MRI Dataset https://www.kaggle.com/datasets/masoudnickparvar/brain-tumor-mri-dataset [6] Goodfellow,I.,Bengio,Y.,&Courville,A.DeepLearning.MITPress,2016.

Copyright

Copyright © 2026 Neha Tyagi, Kshitij Chauhan, Shivanshu Kasaundhan, Priyanshi Negi. 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.