Quantum annealing feature selection on light-weight medical image datasets
Merlin A Nau1, Luca A Nutricati2,3, Bruno Camino4
1Pattern Recognition Lab, Friedrich-Alexander-Universität Erlangen-Nürnberg, 91052, Erlangen, Germany. merlin.nau@fau.de.
Scientific Reports
|August 7, 2025
Summary
Quantum computing, specifically quantum annealing, shows promise for feature selection in medical imaging. This approach tackles complex selection problems, offering potential for high-dimensional data analysis despite current hardware limitations.
Area of Science:
- Quantum Computing
- Medical Imaging
- Computational Optimization
Background:
- Feature selection is crucial for analyzing light-weight medical image datasets, but its computational complexity (k of n problem) hinders scalability.
- Classical methods struggle with the combinatorial explosion inherent in large-scale feature selection.
- Quantum annealers are theoretically suited for optimization problems like feature selection.
Purpose of the Study:
- To investigate the application of quantum computing algorithms on real quantum hardware for feature selection in medical imaging.
- To develop and demonstrate a scalable method for solving larger feature selection instances than previously achieved on quantum annealers.
- To compare the efficacy of quantum annealing-based feature selection against various classical and learning-based methods.
Main Methods:
- A novel method combining a linear Ising penalty mechanism with subsampling and thresholding was developed for enhanced scalability.
- The method was tested on a toy problem involving feature selection for reconstructing small-scale medical images.
- Performance was benchmarked against randomized baselines, classical algorithms, and autoencoder-based feature representations.
Main Results:
- Quantum annealing-based feature selection proved effective in the simplified use case, demonstrating potential for high-dimensional optimization.
- The proposed method enabled solving larger feature selection instances on commercial quantum annealers than previously reported.
- While autoencoders offered superior reconstruction, quantum annealing provided greater interpretability and direct feature selection control.
Conclusions:
- Quantum annealing presents a viable approach for feature selection in specific medical imaging applications, particularly where interpretability is key.
- Current quantum hardware limitations necessitate further development for broader real-world applicability.
- This study highlights the potential of quantum computing to address computationally intensive challenges in medical data analysis.
