Related Experiment Video
Updated: Apr 7, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.7K
Overview of Deep Learning Algorithms and Optimizers for Brain Tumor Segmentation.
Nisha Purohit1, Chandi Prasad Bhatt2
1Department of Allied Health Sciences, Delhi Pharmaceutical Sciences and Research University, New Delhi, India.
Journal of Medical Physics
|April 6, 2026
Summary
Deep learning models significantly enhance brain tumor segmentation accuracy, achieving high Dice scores and outperforming traditional methods. Further research is needed to address challenges like computational complexity and improve clinical adoption.
Area of Science:
- Medical Imaging Analysis
- Artificial Intelligence in Healthcare
- Computational Neuroscience
Background:
- Brain tumor segmentation is critical for diagnosis, treatment planning, and monitoring.
- Evolution from manual to machine learning and deep learning (DL) techniques.
- Convolutional Neural Networks (CNNs) have revolutionized segmentation by enabling end-to-end learning.
Purpose of the Study:
- To review and analyze various DL architectures for brain tumor segmentation.
- To evaluate the performance of DL models optimized with different optimizers.
- To identify current challenges and future research directions in DL-based brain tumor segmentation.
Main Methods:
- Analysis of different deep learning architectures, including CNNs.
- Evaluation of model performance based on optimization techniques.
- Review of segmentation metrics such as Dice scores and Intersection over Union (IoU).
Main Results:
- DL models achieved high segmentation accuracy, outperforming traditional methods.
- Reported Dice scores up to 0.91 and validation accuracy of 98%.
- Specific Dice scores: 0.84 (enhancing tumor), 0.85 (tumor core), 0.91 (whole tumor); Mean IoU: 0.8665.
Conclusions:
- Deep learning significantly improves brain tumor segmentation accuracy and efficiency.
- Challenges include computational complexity, dataset imbalance, and generalization.
- Future work should focus on transfer learning, dataset diversity, and explainable AI for clinical integration.