Using hybrid quantum-classical convolutional neural networks to identify cancer

Artificial intelligence increases in importance in medical diagnostics. In this critical context, accurate and reliable predictions are crucial. Training of machine learning algorithms typically requires large, annotated datasets, especially for computer vision tasks. Clinical studies typically achieve sample sizes of around 100 to 1000, which is often not sufficient for ML approaches. Quantum computing assisted algorithms promise to achieve high prediction accuracy even with limited amounts of data. Within this talk we present a joint project between the Fraunhofer Institute for Cognitive Systems and the Department of Radiology, University Hospital, Ludwig-Maximilians-University Munich, where we have developed a quantum-classical convolutional neural network to identify cancer in imaging data from various modalities. This data includes ultrasound images to diagnose breast cancer and computer tomography images to identify malign lesions in the lung. The proof-of-principle analysis shows a promising performance of hybrid quantum-classical convolutional neural networks and encourages further studies into this direction.