Related Experiment Video
Updated: Sep 28, 2025
12:52
Isolation and Flow Cytometric Analysis of Glioma-infiltrating Peripheral Blood Mononuclear Cells
Published on: November 28, 2015
16.0K
Machine Learning in Differentiating Gliomas from Primary CNS Lymphomas: A Systematic Review, Reporting Quality, and
G I Cassinelli Petersen1,2, J Shatalov3, T Verma1,4
1From the Department of Radiology and Biomedical Imaging (G.I.C.P., T.V., H.S., R.C.B., S.M., T.Z., L.H.S., J.C., R.A.B., A.M., M.S.A.).
AJNR. American Journal of Neuroradiology
|April 1, 2022
Summary
Machine learning shows promise in distinguishing gliomas from primary CNS lymphoma, but studies often lack robust data and external validation, impacting reliability.
Area of Science:
- Neuro-oncology
- Radiology
- Artificial Intelligence
Background:
- Differentiating gliomas from primary CNS lymphoma is diagnostically challenging with significant therapeutic implications.
- While biopsy is standard, MR imaging combined with machine learning offers promising non-invasive diagnostic potential.
Purpose of the Study:
- To systematically evaluate the reporting quality and risk of bias in studies using machine learning for tumor differentiation.
- To assess the databases, algorithms, and performance metrics of machine learning models applied to differentiating gliomas and primary CNS lymphoma.
Main Methods:
- A comprehensive literature search was conducted across major databases (EMBASE, MEDLINE, Cochrane, Web of Science).
- 23 studies involving 2276 patients were included, focusing on machine learning models for differentiating primary CNS lymphoma from gliomas.
- Reporting quality and risk of bias were assessed using TRIPOD and ROB tools, with meta-analysis performed on a subset of studies.
Main Results:
- Logistic regression and support vector machines achieved high external validation performance (AUC 0.961, accuracy 91.2%) using radiomic features.
- Meta-analysis of machine learning classifiers yielded a mean AUC of 0.944.
- Significant deficiencies were noted in reporting quality (median TRIPOD 51.7%) and high risk of bias in 16 studies.
Conclusions:
- Machine learning demonstrates considerable potential for differentiating gliomas and primary CNS lymphoma.
- However, most studies suffer from small, imbalanced datasets and insufficient external validation.
- Identified reporting quality and bias issues limit the generalizability and reproducibility of current findings.
