International Journal of Innovative Research in Computer and Communication Engineering
ISSN Approved Journal | Impact factor: 8.771 | ESTD: 2013 | Follows UGC CARE Journal Norms and Guidelines
| Monthly, Peer-Reviewed, Refereed, Scholarly, Multidisciplinary and Open Access Journal | High Impact Factor 8.771 (Calculated by Google Scholar and Semantic Scholar | AI-Powered Research Tool | Indexing in all Major Database & Metadata, Citation Generator | Digital Object Identifier (DOI) |
| TITLE | Precision-Optimized 3d U-Net for Brain Tumor Segmentation Using Slice-Focused MRI Analysis |
|---|---|
| ABSTRACT | Accurate delineation of brain tumors from multi-parametric Magnetic Resonance Imaging (MRI) is a critical prerequisite for glioma diagnosis, treatment planning and post-operative monitoring, yet manual voxel-wise annotation by radiologists remains slow, subjective and difficult to scale to growing clinical imaging volumes. While 3D U-Net architectures have become a standard approach for volumetric medical image segmentation, their direct application is constrained by heavy GPU memory demands, sensitivity to the severe class imbalance between small tumor sub-regions and the dominant background, and the computational cost of processing full high-resolution volumes. This paper presents a precision-optimized 3D U-Net for brain tumor segmentation that deliberately favours architectural simplicity over added complexity, combining three targeted enhancements: a hybrid Dice-Cross-Entropy loss to counter class imbalance while preserving both regional overlap and voxel-level accuracy; tumor-aware slice selection that restricts training to the axial slice range empirically found to contain most tumor tissue, reducing memory footprint and computational overhead; and Automatic Mixed Precision (AMP) training to accelerate convergence on mid-range GPUs. The framework is trained and evaluated on the BraTS 2020 dataset, comprising 369 subjects imaged across four MRI modalities (T1, T1ce, T2, FLAIR), predicting four voxel classes corresponding to background, necrotic/non-enhancing core, edema and enhancing tumor. Despite retaining a compact, residual- and attention-free encoder-decoder design with only a few million trainable parameters, the proposed model attains an average Dice Similarity Coefficient of 98.9% across the Whole Tumor, Tumor Core and Enhancing Tumor sub-regions on the validation split, exceeding several more complex published baselines. These results indicate that careful preprocessing, loss design and memory-aware training can match or surpass the accuracy of heavier architectures while remaining fast enough - under four seconds of inference per volume - for deployment on modest clinical hardware. |
| AUTHOR | M. KALYANI, K. SUPRIYA Department of Computer Science and Engineering, St. Mary's Women's Engineering College, Budampadu, Guntur, Andhra Pradesh, India |
| VOLUME | 186 |
| DOI | DOI: 10.15680/IJIRCCE.2026.1407023 |
| pdf/23_Precision-Optimized 3d U-Net for Brain Tumor Segmentation Using Slice-Focused MRI Analysis.pdf | |
| KEYWORDS | |
| References | [1] M. Havaei et al., "Brain tumor segmentation with deep neural networks," Medical Image Analysis, vol. 35, pp. 18-31, 2017. [2] O. Ronneberger, P. Fischer, and T. Brox, "U-Net: Convolutional Networks for Biomedical Image Segmentation," Proc. MICCAI, pp. 234-241, 2015. [3] O. Cicek et al., "3D U-Net: Learning Dense Volumetric Segmentation from Sparse Annotation," Proc. MICCAI, pp. 424-432, 2016. [4] S. Alam et al., "3D Deep Residual U-Net for Brain Tumor Segmentation," Procedia Computer Science, vol. 200, pp. 1000-1009, 2023. [5] Z. Huang et al., "GCAUNet: Group Cross-Channel Attention Residual UNet for Brain Tumor Segmentation," Biomedical Signal Processing and Control, vol. 68, 2021. [6] D. Maji et al., "Attention Res-UNet with Guided Decoder for Semantic Segmentation of Brain Tumors," Biomedical Signal Processing and Control, vol. 76, 2022. [7] F. Isensee et al., "nnU-Net: A Self-Configuring Method for Deep Learning-Based Biomedical Image Segmentation," Nature Methods, vol. 18, pp. 203-211, 2021. [8] K. Kamnitsas et al., "Efficient Multi-scale 3D CNN with Fully Connected CRF for Accurate Brain Lesion Segmentation," Medical Image Analysis, vol. 36, pp. 61-78, 2017. [9] D. Wang et al., "Automatic Brain Tumor Segmentation Using Cascaded Anisotropic Convolutional Neural Networks," Proc. MICCAI BrainLes Workshop, pp. 178-190, 2017. [10] S. Bakas et al., "Advancing the Cancer Genome Atlas Glioma MRI Collections with Expert Segmentation Labels and Radiomic Features," Scientific Data, vol. 4, 2017. [11] J. Long, E. Shelhamer, and T. Darrell, "Fully Convolutional Networks for Semantic Segmentation," Proc. CVPR, pp. 3431-3440, 2015. [12] H. R. Roth et al., "DeepOrgan: Multi-level Deep Convolutional Networks for Automated Pancreas Segmentation," Proc. MICCAI, pp. 556-564, 2015. |
