Related Experiment Video
Updated: Sep 23, 2025
06:48
Automated Segmentation of Cortical Grey Matter from T1-Weighted MRI Images
Published on: January 7, 2019
9.0K
An optimal brain tumor segmentation algorithm for clinical MRI dataset with low resolution and non-contiguous slices
Dheerendranath Battalapalli1, B V V S N Prabhakar Rao1, P Yogeeswari2
1Department of Electrical and Electronics Engineering, Birla Institute of Technology and Science Pilani, Hyderabad Campus, Hyderabad, 500078, India.
BMC Medical Imaging
|May 15, 2022
Summary
Automated brain tumor segmentation using deep neural networks (deepmedic) shows promise but requires custom training for low-resolution clinical MRI scans. Deepmedic outperforms traditional methods like region growing and fuzzy C-means for accurate tumor delineation.
Area of Science:
- Medical Imaging
- Computer Vision
- Artificial Intelligence
Background:
- Manual segmentation of brain tumors from MRI is time-consuming and lacks reproducibility.
- Low-resolution clinical MRI scans pose challenges for accurate tumor delineation.
- Automated segmentation algorithms are needed to improve efficiency and precision in clinical practice.
Purpose of the Study:
- To evaluate and compare the performance of three segmentation methods: region growing, fuzzy C-means (FCM), and deep neural networks (DeepMedic).
- To assess these algorithms on both a public dataset (BRATS 2018) and a private clinical dataset of brain tumor MRIs.
Main Methods:
- Investigated region growing, fuzzy C-means, and DeepMedic algorithms.
- Evaluated performance on 48 patients from the BRATS 2018 dataset and 43 patients from a clinical dataset.
- Measured performance using Dice Similarity Coefficient, Hausdorff Distance, and volume metrics.
Main Results:
- Region growing performed poorly compared to FCM and DeepMedic.
- FCM and DeepMedic achieved comparable Dice Similarity Coefficient scores on both datasets.
- Accuracy for FCM and DeepMedic was generally below 70%.
Conclusions:
- DeepMedic, while accurate on high-resolution data, requires custom training for low-resolution clinical MRI scans.
- Large datasets are necessary for DeepMedic to be a standalone clinical application.
- DeepMedic demonstrates superior potential for brain tumor segmentation compared to region growing and FCM.
