Related Experiment Video
Updated: Mar 8, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.6K
Multimodal brain tumor segmentation and classification based on optimized DeepLabV3 + and fused fire module with
Muhammad Sami Ullah1, Muhammad Attique Khan2, Yunyoung Nam3
1Department of Computer Science, HITEC University, Taxila, Pakistan.
European Journal of Medical Research
|March 6, 2026
Summary
We developed SMDeepNet, a novel deep learning model for brain tumor segmentation and classification. This AI achieved high accuracy in segmenting tumors and classifying modalities, outperforming existing methods.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computational Neuroscience
Background:
- Accurate brain tumor segmentation and classification are crucial for effective treatment planning.
- Deep learning models have shown promise but require further optimization for complex medical imaging tasks.
Purpose of the Study:
- To propose SMDeepNet, a novel deep learning architecture for enhanced brain tumor segmentation and classification.
- To evaluate the performance of SMDeepNet on the BraTS 2023 dataset.
Main Methods:
- Developed SMDeepNet, integrating an optimized DeepLabV3+ for segmentation and a Fused Fire Module with Self-Attention for classification.
- Employed ResNet-50 backbone with dynamic hyperparameter initialization and Atrous Spatial Pyramid Pooling (ASPP) for feature extraction.
- Utilized parallel Fire-Residual Bottleneck (Fire-RB) and Hybrid Efficient Attention (Hybrid-EA) frameworks, incorporating self-attention for classification.
Main Results:
- Achieved segmentation accuracy of 0.9871, Dice Score of 0.9420, and Intersection over Union (IOU) of 0.8951.
- Attained modality classification accuracy of 0.9920.
- Demonstrated superior performance compared to recent state-of-the-art techniques.
Conclusions:
- SMDeepNet offers a robust and effective deep learning solution for brain tumor segmentation and classification.
- The proposed architecture significantly improves upon existing methods, paving the way for advanced neuro-oncology diagnostics.