Related Experiment Video
Updated: Apr 13, 2026
06:48
On-Site Sampling and Extraction of Brain Tumors for Metabolomics and Lipidomics Analysis
Published on: May 31, 2020
6.5K
Machine learning methods for the classification of gliomas: Initial results using features extracted from MR
G Ranjith1, R Parvathy2, V Vikas3
1SCTIMST, Sri Chitra Tirunal Institute of Medical Sciences and Technology, Trivandrum, Kerala, India ranjithg@sctimst.ac.in.
The Neuroradiology Journal
|April 30, 2015
Summary
Machine learning algorithms show promise in classifying brain gliomas using MRI data. Random forest achieved the highest accuracy, suggesting a potential shift from invasive methods for tumor classification.
Area of Science:
- Artificial Intelligence in Medicine
- Medical Imaging Analysis
- Machine Learning for Healthcare
Background:
- Radiologists face increasing data volumes from advanced imaging modalities.
- Automated and intelligent systems are crucial for modern medical diagnosis.
- Machine learning (ML) is vital for medical image analysis, including computer-aided diagnosis.
Purpose of the Study:
- To classify gliomas into benign and malignant types.
- Utilize magnetic resonance imaging (MRI) data for classification.
- Explore the efficacy of ML algorithms in brain tumor classification.
Main Methods:
- Retrospective analysis of MRI data from 28 glioma patients.
- Classification of WHO Grade II (benign) vs. Grade III/IV (malignant) gliomas.
- Feature extraction from MR spectroscopy and application of four ML algorithms: multilayer perceptrons, support vector machine, random forest, and locally weighted learning.
Main Results:
- Three ML algorithms achieved an area under the ROC curve (AUC) greater than 0.80.
- Random forest demonstrated the best performance with an AUC of 0.911.
- Locally weighted learning yielded the highest sensitivity at 86.1%.
Conclusions:
- ML algorithms show significant promise for glioma classification using MRI.
- Integration of features from additional MR sequences could further enhance performance.
- ML offers a potential non-invasive alternative to conventional histopathological analysis.
