Quantum Hybrid Model Tested for ECG Arrhythmia Detection

Researchers from CQUniversity in Australia and Telkom University in Indonesia have published a study in Scientific Reports evaluating the potential of a hybrid quantum algorithm for classifying cardiac arrhythmias. The study, led by Melvin Yashnil Ramkhelawan and colleagues, investigates whether integrating quantum computing components into deep learning architectures can improve the accuracy of electrocardiogram (ECG) image analysis, particularly when spatial resolution is reduced. As quantum machine learning moves from theoretical proposals to practical applications, understanding the specific conditions under which quantum pathways offer advantages over classical methods is critical for the development of next-generation medical diagnostic tools. The team focused on a six-class arrhythmia classification task, a common benchmark in cardiological signal processing that requires distinguishing between normal heartbeats and various types of irregularities.

The core of the proposed system is a hybrid architecture that combines a fully fine-tuned ResNet50 convolutional neural network with a 10-qubit variational quantum circuit. In this design, the ResNet50 backbone serves as the feature extractor, processing the input ECG images to identify relevant patterns. These features are then passed to the quantum circuit, which acts as a classifier. This approach leverages the strengths of classical deep learning for high-dimensional data processing while attempting to exploit the potential computational benefits of quantum superposition and entanglement for the final decision-making step. The authors aimed to determine if this hybrid structure could maintain high classification performance even when the input images are downsampled, a scenario that might occur in resource-constrained environments or for rapid screening applications.

To rigorously assess the algorithm’s performance, the researchers utilized a dataset of 24,799 held-out ECG images. They evaluated the model at five controlled content resolutions: 80, 112, 160, 224, and 320 pixels. This multi-resolution approach is essential because it tests the robustness of the model against varying levels of spatial information. By comparing the hybrid quantum algorithm against a conventional ResNet50 classifier and a closely parameter-matched classical 10-dimensional bottleneck, the study isolates the specific contribution of the quantum component. The classical bottleneck serves as a fair baseline, ensuring that any observed differences in performance are attributable to the quantum processing rather than simply the architecture’s capacity or the number of parameters involved.

At the highest resolutions of 224 and 320 pixels, all three approaches reached ceiling-level classification performance. This indicates that when sufficient spatial detail is available, the choice of classifier—whether classical or hybrid quantum—has a minimal impact on the final accuracy. The models were able to extract enough information from the high-resolution images to distinguish between the six arrhythmia classes with high confidence. However, the study found that differences between the architectures became more apparent as the spatial information was reduced, highlighting the potential niche for quantum-enhanced processing in scenarios where data fidelity is compromised.

At a resolution of 112 pixels, the hybrid quantum algorithm demonstrated a mean accuracy of 0.9690 with a standard deviation of 0.0124 across three independent training seeds. In comparison, the conventional ResNet50 achieved a mean accuracy of 0.9396 with a standard deviation of 0.0044, while the classical bottleneck recorded a mean accuracy of 0.9580 with a standard deviation of 0.0354. Although the hybrid algorithm achieved the highest mean accuracy, the corresponding macro-F1 scores revealed a more nuanced picture. The hybrid algorithm scored 0.8638 ± 0.0417, the ResNet50 scored 0.8235 ± 0.0062, and the classical bottleneck scored 0.8783 ± 0.1033. These results suggest that while the hybrid model may offer higher accuracy, its performance can be less consistent across different training initializations compared to the classical counterparts.

The variability in performance across training seeds is a significant finding, indicating a sensitivity to optimization randomness in the hybrid quantum architecture. At 160 pixels, the architectures approached ceiling performance, suggesting that the quantum pathway does not provide a consistent classification advantage across all resolutions. At the lowest tested resolution of 80 pixels, the hybrid algorithm and the standard ResNet50 showed comparable classification behavior, with neither demonstrating a clear superiority. This implies that the benefits of the quantum component are most relevant in the intermediate range of spatial information reduction, where classical models begin to struggle with the loss of detail but have not yet reached the point of complete failure.

The study underscores the importance of considering both accuracy and stability when evaluating new machine learning models for medical applications. While a higher mean accuracy is desirable, a model that is highly sensitive to random initialization may pose challenges for reproducibility and clinical deployment. The classical bottleneck, despite having a lower mean accuracy than the hybrid model at 112 pixels, exhibited a much lower standard deviation in its macro-F1 score, suggesting a more stable performance profile. This trade-off between peak performance and consistency is a critical consideration for healthcare providers who require reliable diagnostic tools that perform predictably across different patient populations and data conditions.

It is important to note that the results reflect image-level benchmark performance without patient identifiers or external validation. The authors emphasize that broader generalizability requires patient-independent and external validation studies. In clinical settings, the ability of a model to generalize across different patients, demographics, and hospital environments is paramount. The current study provides a strong foundation for understanding the technical capabilities of the hybrid quantum algorithm, but further research is needed to confirm its utility in real-world clinical workflows. The lack of external validation means that the findings should be interpreted as a proof of concept rather than a definitive clinical recommendation.

This research contributes to the growing body of literature on quantum machine learning in healthcare, offering a detailed analysis of the conditions under which quantum components may be beneficial. By systematically varying the input resolution and comparing against well-established classical baselines, the authors provide valuable insights into the practical limitations and potential advantages of hybrid quantum-classical architectures. As quantum hardware continues to advance and become more accessible, such studies will be crucial for guiding the integration of quantum technologies into medical imaging and signal processing pipelines. The work highlights the need for careful benchmarking and the recognition that quantum advantage is not universal but context-dependent.

The implications of this study extend beyond arrhythmia classification to other areas of medical imaging where data resolution may vary. The hybrid approach could potentially be adapted for other diagnostic tasks, provided that the specific conditions for quantum advantage are met. Researchers and clinicians interested in quantum-enhanced diagnostics should consider the balance between computational complexity, data resolution, and model stability when designing their systems. The findings from this study serve as a cautionary tale against assuming that quantum computing will automatically outperform classical methods in all scenarios, while also demonstrating that there are specific niches where hybrid approaches may offer competitive performance. Future work should focus on expanding the dataset, incorporating external validation, and exploring more complex quantum circuits to further elucidate the potential of quantum machine learning in cardiology.

Subject of Research: Hybrid quantum algorithm performance in multi-resolution ECG arrhythmia classification

Article Title: A hybrid quantum algorithm for multi-resolution arrhythmia classification

Article References: Ramkhelawan, M. Y., Wibowo, S., Grandhi, S., & Mandala, S. (2026). A hybrid quantum algorithm for multi-resolution arrhythmia classification. Scientific Reports. https://doi.org/10.1038/s41598-026-76787-0

Image Credits: AI Generated

DOI: 10.1038/s41598-026-76787-0

Keywords: quantum computing, ECG analysis, arrhythmia detection, deep learning, medical imaging, hybrid algorithms, ResNet50, cardiology, machine learning, signal processing, hybrid, quantum