Related Experiment Video
Updated: Oct 5, 2025
09:53
Quantifying the Brain Metastatic Tumor Micro-Environment using an Organ-On-A Chip 3D Model, Machine Learning, and Confocal Tomography
Published on: August 16, 2020
7.4K
Metabolite selection for machine learning in childhood brain tumour classification
Dadi Zhao1,2, James T Grist1,2, Heather E L Rose1,2
1Institute of Cancer and Genomic Sciences, University of Birmingham, Birmingham, UK.
NMR in Biomedicine
|January 28, 2022
Summary
Selecting specific metabolites using magnetic resonance spectroscopy (MRS) improves the machine learning accuracy for diagnosing childhood brain tumors. This metabolite selection method offers better tumor discrimination than principal component analysis.
Area of Science:
- Neuroimaging
- Biochemistry
- Machine Learning
Background:
- Magnetic Resonance Spectroscopy (MRS) aids in diagnosing pediatric brain tumors.
- Principal Component Analysis (PCA) is a common method for reducing metabolite profile dimensionality.
- Optimal metabolite set identification for multi-class problems remains underexplored.
Purpose of the Study:
- To investigate metabolite selection from in vivo MRS for classifying childhood brain tumors.
- To compare metabolite selection with PCA for dimensionality reduction.
Main Methods:
- Retrospective evaluation of multi-site 1.5T and 3T cohorts with ependymoma, medulloblastoma, and pilocytic astrocytoma.
- Dimensionality reduction using multi-class receiver operating characteristics (ROC) for metabolite selection.
- Classification accuracy assessed via leave-one-out and k-fold cross-validation.
Main Results:
- Key metabolites identified for 1.5T: myo-inositol, total lipids/macromolecules, and total creatine.
- Key metabolites identified for 3T: glycine, total N-acetylaspartate, and total choline.
- Metabolite selection significantly improved tumor discrimination compared to PCA (P < 0.05).
- Highest accuracies: 85% (1.5T, SVM) and 75% (3T, LDA with oversampling).
Conclusions:
- A curated set of crucial metabolites enhances the discrimination of childhood brain tumors.
- Metabolite selection via multi-class ROC analysis is a viable alternative to PCA.
- This approach shows promise for improving diagnostic accuracy in pediatric neuro-oncology.
