In a paper published in the journal Scientific Reports, researchers presented a quantum hybrid classical convolutional neural network (QCCNN) approach for breast cancer diagnosis. This approach utilizes quantum computing's data processing power to enhance diagnostic accuracy and efficiency.
Study: Quantum classical hybrid convolutional neural networks for breast cancer diagnosis. Image Credit: Po.Kanyaluck/Shutterstock.com
The QCCNN seeks to improve learning efficiency and overcome the challenges posed by conventional CNNs, which demand significant computational power, by incorporating a quantum convolutional layer. Simulation results on German breast cancer study group (GBSG), surveillance, epidemiology, results program (SEER), and Wisconsin diagnostic breast cancer (WDBC) datasets showed that QCCNN outperformed CNN and logistic regression models, highlighting its potential to enhance breast cancer diagnostics.
Background
Past work highlighted breast cancer as the second most common malignancy globally, affecting over 2.2 million women. Machine learning (ML), particularly CNNs, has significantly advanced breast cancer diagnostics, aiding early detection through improved image analysis. However, CNNs face computational limitations with large, complex datasets, creating demand for more efficient methods. With its powerful parallel processing, Quantum computing offers new potential for medical diagnostics by enhancing data processing speed and accuracy, although its application in breast cancer diagnosis is still emerging.
Innovative Model for Breast Cancer
This study explores the potential of a QCCNN for breast cancer diagnosis, comparing its performance with traditional models like CNN and logistic regression. The GBSG, SEER, and WDBC datasets were employed to train and validate each model's diagnostic accuracy.
These datasets, sourced from the Kaggle repository, provide comprehensive data on breast cancer cases, including patient records and tumor features. For example, the GBSG dataset includes 686 samples with 11 prognostic features, SEER consists of 4024 samples from female patients diagnosed between 2000 and 2017, and WDBC has 569 samples with 30 tumor-specific features.
Data preprocessing included normalization to standardize feature ranges, enabling effective model training across different scales. The CNN architecture comprises convolutional layers, pooling layers, and fully connected layers, all structured to identify and categorize features from images.
CNN's approach involved feature extraction using sliding filters over images and pooling layers to reduce spatial dimensions. This model was trained on the three datasets with classification accuracy as a performance metric. QCCNN, however, added a quantum layer to this structure, integrating quantum computations with classical layers for potentially enhanced image recognition.
Constructed using the PennyLane framework, the QCCNN included a quantum convolutional layer that employed quantum circuits to encode features and perform entanglement operations. CNN and QCCNN were optimized using the Adam optimizer with a learning rate 0.1 and evaluated using cross-entropy loss.
Logistic regression was utilized as a reference for classification, acting as a linear model that estimates probabilities through a linear combination of the input variables. Its structure included a parameter layer with weights and a bias term, optimized similarly with the Adam optimizer at a 0.01 learning rate.
Performance was also accurately measured. Logistic regression's architecture was comparatively simple, mapping input features to probabilities through a sigmoid function, yielding insights into the model's diagnostic capability relative to CNN and QCCNN. The experimental setup for QCCNN and CNN leveraged PyTorch and PennyLane libraries for efficient training and evaluation.
Overall, the QCCNN aimed to capitalize on the strengths of both quantum and classical layers, showing promise in breast cancer diagnostic applications. Evaluation criteria included accuracy based on true and false classifications, with cross-entropy loss quantifying differences between predicted and actual results, providing a comprehensive approach to model assessment.
Hybrid Quantum-Classical Diagnostic Model
This study validated the QCCNN model’s efficiency in breast cancer diagnosis by comparing it with CNN and logistic regression on the GBSG, SEER, and WDBC datasets. Results showed QCCNN consistently achieved the highest accuracies, with 74.375% on GBSG, 90.088% on SEER, and 96.155% on WDBC validation sets, outperforming the other models. Each model was trained using 70% of the dataset and validated with the remaining 30%, with accuracy as the primary performance metric.
The findings demonstrate the QCCNN model’s potential in medical imaging and highlight the advantages of integrating quantum computing for disease diagnosis. Compared to traditional approaches, the QCCNN model leverages quantum layers to enhance feature extraction and improve classification performance.
Using the PennyLane framework, QCCNN incorporated quantum entanglement in its convolutional layers, significantly boosting diagnostic accuracy across datasets. This hybrid approach underscores the potential of quantum-classical models in medical image analysis, suggesting promising applications in complex diagnostic tasks where traditional models may have limitations.
Conclusion
To sum up, this study proposed a QCCNN model for breast cancer diagnosis, demonstrating its superior classification accuracy compared to traditional models like CNN and logistic regression across GBSG, SEER, and WDBC datasets. Although the research faced limitations regarding quantum circuit design and the gap between conventional and quantum computing, it highlighted the promising potential of QCCNN in medical diagnosis.
Future research aims to explore breast cancer staging applications and assess computational efficiency, enhancing the model's real-world applicability. Comprehensive evaluations of quantum ML techniques were also expected to advance the field of medical diagnosis.
Journal Reference
Xiang, Q. et al. (2024). Quantum classical hybrid convolutional neural networks for breast cancer diagnosis. Scientific Reports, 14:1, 1-13. DOI: 10.1038/s41598-024-74778-7, https://www.nature.com/articles/s41598-024-74778-7
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.