Related Experiment Video
Updated: Jul 18, 2025
07:15
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
6.9K
A Multidimensional Connectomics- and Radiomics-Based Advanced Machine-Learning Framework to Distinguish Radiation
Yilin Cao1,2, Vishwa S Parekh3,4, Emerson Lee1
1Department of Radiation Oncology and Molecular Radiation Sciences, Johns Hopkins University School of Medicine, Baltimore, MD 21231, USA.
Cancers
|August 26, 2023
Summary
Tumor connectomics combined with multiparametric radiomics effectively differentiates radiation necrosis from true progression in brain metastases. This novel MRI-based approach offers improved diagnostic accuracy for treatment response assessment.
Area of Science:
- Neuro-oncology
- Radiology
- Graph Theory
- Machine Learning
Background:
- Distinguishing radiation necrosis (RN) from true progression (TP) in brain metastases post-stereotactic radiosurgery (SRS) is clinically challenging.
- Accurate differentiation is crucial for appropriate treatment planning and patient management.
- Current imaging modalities often lack the specificity to reliably differentiate these conditions.
Purpose of the Study:
- To introduce and validate a novel MRI-based framework, tumor connectomics, integrated with multiparametric radiomics (mpRad).
- To develop a machine-learning model capable of distinguishing between RN and TP in brain metastases.
- To assess the diagnostic performance of this integrated approach.
Main Methods:
- A complex graph theory framework (tumor connectomics) was developed using MRI data.
- Multiparametric radiomics (mpRad) features were extracted from T1 post-contrast and T2 FLAIR sequences.
- An Isomap support vector machine (IsoSVM) model within the Integrated Radiomics Informatics System (IRIS) was trained and validated using leave-one-out cross-validation on pathologically confirmed cases.
Main Results:
- The study analyzed 135 lesions in 110 patients, with 43 cases of RN and 92 of TP.
- Top-performing connectomics features included centrality measures (degree, betweenness, eigenvector).
- The optimized IsoSVM model, combining connectomics and mpRad features, achieved a sensitivity of 0.87, specificity of 0.84, AUC-ROC of 0.89, and AUC-PR of 0.94.
Conclusions:
- Tumor connectomics, a novel MRI-based graph theory approach, shows significant promise in differentiating RN from TP.
- The integration of tumor connectomics with mpRad and machine learning provides a robust tool for assessing treatment response in brain metastases.
- This advanced imaging analysis framework has the potential to improve diagnostic accuracy and guide clinical decision-making.
