3-D Quantum-Inspired Self-Supervised Tensor Network for Volumetric Segmentation of Medical Images.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Quantum Computing

Background:

• Medical image segmentation is crucial for diagnosis and treatment planning.
• Classical supervised and self-supervised networks often face slow convergence issues.
• Existing methods require extensive training data and computational resources.

Purpose of the Study:

• To introduce a novel shallow 3-D self-supervised tensor neural network (3-D-QNet) for medical image segmentation.
• To leverage quantum formalism and tensor decomposition for faster convergence and obviated training/supervision.
• To evaluate the performance of 3-D-QNet on benchmark datasets.

Main Methods:

• Developed a 3-D quantum-inspired self-supervised tensor neural network (3-D-QNet).
• Incorporated quantum neurons (qubits) and tensor decomposition within a shallow network architecture.
• Utilized an S-connected third-order neighborhood topology for voxelwise processing.
• Tested on the BRATS 2019 (Brain MR) and LiTS17 (Liver Tumor) datasets.

Main Results:

• 3-D-QNet demonstrated promising Dice Similarity (DS) for volumetric segmentation.
• Achieved faster convergence compared to classical supervised and self-supervised networks.
• Outperformed or showed comparable results to established models like 3D-UNet, VoxResNet, DRINet, and 3D-ESPNet.

Conclusions:

• The proposed 3-D-QNet offers an efficient and effective self-supervised approach for medical image semantic segmentation.
• Quantum formalism and tensor decomposition accelerate network convergence, reducing computational burden.
• 3-D-QNet shows potential as a viable alternative to time-intensive supervised methods in medical imaging.