QC Ware and Roche - Quantum Neural Networks for Biomedical Image Analysis
QC Ware, a leading quantum software and services company, partnered with Roche Pharma Research and Early Development (pRED) to explore how quantum computing could support drug development through advanced biomedical image analysis. This collaboration focused specifically on using quantum neural networks to classify medical images for detecting and diagnosing diabetic retinopathy, a serious eye condition that can lead to vision loss in people with diabetes.
Roche faced significant challenges in analyzing and classifying complex biomedical images, particularly retinal scans used to detect diabetic retinopathy. This condition, which affects blood vessels in the retina, requires early detection and accurate staging to prevent vision loss, but analyzing retinal images is a complex and resource-intensive process.
Traditional machine learning approaches for medical image analysis require substantial computational resources and can struggle with the subtle features and patterns that indicate different stages of diabetic retinopathy. More accurate and efficient methods for analyzing these images could potentially improve diagnostic capabilities, accelerate drug development for eye conditions, and ultimately lead to better patient outcomes.
Additionally, as a leading biotechnology company, Roche needed to explore how emerging technologies like quantum computing might transform their research and development processes in the future. Understanding the potential applications and limitations of quantum computing for pharmaceutical research could provide a competitive advantage as the technology matures.
The QC Ware-Roche collaboration developed a quantum machine learning approach focused on using quantum neural networks for biomedical image classification. Their solution had several key components:
Quantum Vision Transformers. The teams developed novel quantum transformer models for analysing retinal images. These quantum-enhanced vision transformers leveraged the unique properties of quantum computing to process and classify medical images in ways that could potentially surpass classical approaches. This approach drew inspiration from classical transformer models, which have revolutionised natural language processing and computer vision, but implemented them using quantum computing principles to potentially achieve better performance with fewer parameters.
Hardware-Efficient Quantum Neural Networks. The solution employed quantum neural network architectures optimised for current and near-term quantum hardware. These networks were designed to operate efficiently on systems with limited qubit counts and subject to the noise and errors present in current quantum computers.
By tailoring the architecture to the constraints of available quantum hardware, the team created a practical implementation that could be tested on actual quantum computers rather than just simulated environments.
Hybrid Quantum-Classical Approach. Recognizing the limitations of current quantum hardware, the researchers developed a hybrid approach that combined the strengths of quantum computing with classical processing. This allowed them to implement their solution on available quantum systems while still addressing the complex requirements of biomedical image analysis.
The implementation of this quantum solution involved several phases, from theoretical development to practical testing on quantum hardware.
Quantum Algorithm Development. The QC Ware and Roche pRED Quantum Computing Taskforce collaborated to design quantum algorithms specifically for classifying biomedical images. This required translating complex image processing tasks into formats suitable for quantum computation while maintaining the accuracy needed for medical applications.
Testing on Quantum Hardware. The team tested their quantum vision transformer models on IBM’s 27-qubit superconducting quantum computer. They ran direct experiments with up to six qubits to evaluate the performance of their algorithms on actual quantum hardware. Additionally, they conducted simulations to assess how their approach might perform on larger quantum systems with up to 100 qubits, providing insights into the potential future capabilities of their solution as quantum hardware continues to advance . Prnewswire
Comparative Analysis
A critical aspect of the implementation was comparing the performance of their quantum neural networks against state-of-the-art classical machine learning models for the same image classification tasks. This allowed them to assess the potential advantages and limitations of quantum approaches for biomedical image analysis.
Documentation and Knowledge Sharing. The research findings were documented in a paper titled “Quantum Vision Transformers,” which was published on arXiv, making the results available to the broader scientific community. This reflected both organizations’ commitment to advancing the field of quantum computing for biomedical applications.
The collaboration between QC Ware and Roche yielded several significant outcomes with implications for both medical diagnostics and quantum computing applications.
Performance Achievements. The study demonstrated that the quantum transformer models matched—and in some cases outperformed—classical models for analyzing retinal images to detect diabetic retinopathy. This finding was particularly notable given the early stage of quantum computing technology. As Iordanis Kerenidis, QC Ware’s senior vice president of quantum algorithms, noted, these results were “extremely encouraging” and illustrated “the potential future of quantum computing in the acceleration of image analysis and medical diagnostics”.
Quantum Computing Insights. The project provided Roche with valuable insights into how quantum computing could support drug development processes. Marielle van de Pol, Global Head Scientific Solution Engineering and Architecture at Roche, emphasized that the result of their joint exploration was “to understand how quantum computing can support drug development”. This understanding of quantum computing’s potential applications and limitations in pharmaceutical research could inform Roche’s technology strategy and investments moving forward.
Industry Recognition. The collaboration positioned both QC Ware and Roche as pioneers in applying quantum computing to medical diagnostics. By successfully demonstrating practical applications of quantum neural networks for biomedical image analysis, they established themselves as leaders in this emerging field.
The QC Ware-Roche partnership outlined several directions for future development of quantum applications in biomedical research:
Expanding to Additional Medical Imaging Applications. Building on their success with diabetic retinopathy detection, the partners identified opportunities to apply their quantum neural network approach to other types of medical images, potentially including radiological images, pathology slides, and microscopy data . Prnewswire
Scaling to Larger Quantum Systems. As quantum hardware continues to advance, the teams planned to scale their algorithms to take advantage of systems with more qubits and lower error rates. Their simulation work with up to 100 qubits provided a roadmap for how their approach could evolve as hardware capabilities improve.
Integration with Drug Development Workflows. Roche aimed to explore how quantum computing capabilities could be integrated into broader drug development workflows, potentially accelerating multiple aspects of the pharmaceutical research and development process.
Quantum Machine Learning Framework Development. QC Ware planned to continue developing its quantum machine learning framework, incorporating insights from the Roche collaboration to enhance its capabilities for biomedical applications and other domains requiring complex image analysis.