Related Experiment Video
Updated: Jul 21, 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
7.3K
An Explainable MRI-Radiomic Quantum Neural Network to Differentiate Between Large Brain Metastases and High-Grade
Tony Felefly1,2,3, Camille Roukoz4, Georges Fares4,5
1Radiation Oncology Department, Hôtel-Dieu de France Hospital, Saint Joseph University, Beirut, Lebanon. tony.felefly@hotmail.com.
Journal of Digital Imaging
|July 28, 2023
Summary
This study introduces a quantum neural network (QNN) using quantum-annealed feature selection to differentiate brain tumors. The quantum-informed model shows comparable performance to classical methods for distinguishing large brain metastases from high-grade gliomas on MRI.
Area of Science:
- Neuroimaging
- Quantum Machine Learning
- Radiomics
Background:
- Distinguishing solitary large brain metastases (LBM) from high-grade gliomas (HGG) on MRI is challenging.
- Accurate differentiation is crucial as management strategies differ significantly.
- Non-invasive methods are needed to avoid invasive biopsies and surgeries.
Purpose of the Study:
- To evaluate the performance and interpretability of an MRI-radiomics variational quantum neural network (QNN).
- To utilize quantum-annealing for mutual information-based feature selection.
- To develop a quantum-informed model for differentiating LBM from HGG on contrast-enhanced T1-weighted (CE-T1) MRI.
Main Methods:
- Retrospective analysis of 423 patients with LBM or HGG on CE-T1 MRI.
- Extraction and selection of 1813 radiomic features using mutual information and quantum annealing.
- Development of a 2-qubit QNN model with 10 selected features.
- Benchmarking against dense neural networks (DNN) and extreme gradient boosting (XGB).
Main Results:
- The QNN achieved a test ROC-AUC of 0.76 and balanced accuracy (bACC) of 0.74.
- Quantum-annealing identified 10 optimal features (6 tumor, 4 peri-tumoral).
- Model performance was comparable to classical models (XGB: test ROC-AUC 0.79, bACC 0.72; DNN: test ROC-AUC 0.75, bACC 0.73).
- Shapley values provided interpretability for QNN predictions.
Conclusions:
- A quantum-informed QNN model can accurately differentiate LBM from HGG on CE-T1 brain MRI.
- Quantum-annealing offers an effective approach for radiomic feature selection.
- The developed QNN model demonstrates comparable performance to state-of-the-art classical machine learning methods.
