Quantum Lung Segmentation: QCU-Net Applied to Chest X-Ray Images

Neural network-driven ML has thrived in applications like image recognition and precision medicine, exemplified by U-Net. Despite advancements like Transformer-integrated models, complexity increases training time and hardware demands, exacerbated by limited medical imaging datasets. Proposing QCU-Net, inspired by quantum variational circuits and CNNs, we address image segmentation challenges. QCU-Net combines U-Net’s aptitude with small medical datasets and quantum computing, innovating through quantum convolutions for efficient feature extraction and dimensionality reduction, streamlining the architecture and cutting training costs. It runs on just four qubits via circuit reuse, practical in the NISQ era. We train and evaluate QCU-Net on annotated chest X-rays from Shenzhen No. 3 People’s Hospital, demonstrating effective lung segmentation with reduced parameters and small-sample training akin to CNNs. Ablation studies reveal tripled U-Net training times sans convolutional layers. Experiments prove quantum computing’s potential and explore its application in medical image segmentation. They further showcase QCU-Net’s robust generalization on previously unseen pulmonary nodule cases, underscoring quantum-assisted medical image analysis benefits.