Related Experiment Video
Updated: Jul 19, 2025
07:15
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
6.9K
Differentiating Radiation Necrosis and Metastatic Progression in Brain Tumors Using Radiomics and Machine Learning
Elahheh Salari1, Haitham Elsamaloty2, Aniruddha Ray3,4
1Departments of Radiation Oncology.
American Journal of Clinical Oncology
|August 15, 2023
Summary
This study developed an automated radiomics and machine learning technique to differentiate radiation necrosis (RN) from brain metastasis progression. The method shows high accuracy in distinguishing these conditions on MRI, potentially improving patient treatment outcomes.
Area of Science:
- Radiology
- Medical Imaging
- Machine Learning
Background:
- Differentiating radiation necrosis (RN) from metastatic progression is challenging due to imaging similarities.
- Accurate differentiation is critical for determining appropriate patient treatment and improving outcomes.
Purpose of the Study:
- To establish an automated technique for differentiating RN from brain metastasis progression.
- To utilize radiomics and machine learning for improved diagnostic accuracy in post-radiosurgery brain metastases.
Main Methods:
- Eighty-six patients with brain metastasis post-stereotactic radiosurgery were analyzed.
- Radiomics features were extracted from post-contrast T1-weighted MRI using various filters (Discrete wavelets transform, Laplacian-of-Gaussian, Gradient, Square).
- Machine learning classifiers (Random Forest, Logistic Regression, Support Vector Classification) were trained and validated.
Main Results:
- Random Forest classification with a Gradient filter achieved the highest performance (AUC=0.910±0.047, accuracy=0.8±0.071).
- Support Vector Classification with wavelet_HHH showed strong results (AUC=0.890±0.89).
- Logistic Regression yielded lower performance metrics compared to the other models.
Conclusions:
- Machine learning-based radiomics analysis can accurately distinguish RN from tumor recurrence on MRI.
- This automated approach may obviate the need for biopsy, potentially enhancing therapeutic outcomes.
