Invention Description
Accurate classification of brain tumors from MRI scans is important for diagnosis and treatment planning, but conventional deep learning models can require extensive training and computational resources. Image preprocessing can improve performance by highlighting important tumor features, yet traditional methods may not capture the complex patterns present in medical images. Emerging quantum computing techniques offer new ways to extract and represent these features. There is a need to explore whether combining quantum and classical image processing can improve tumor classification while reducing training requirements.
Researchers at Arizona State University have developed a quantum-assisted preprocessing pipeline that extracts features from MRI image patches using small quantum circuits and combines them with original image data and classical features from Gabor filters, Local Binary Patterns, and Sobel edge detection. Across a combined dataset of 9,439 MRI images covering glioma, meningioma, and pituitary tumors, adaptive fusion achieved 98.871% validation accuracy and reached 90% accuracy in an average of 20.3 epochs, compared with 28.1 epochs for classical preprocessing alone. The results demonstrate the potential of quantum-assisted medical image analysis, although further validation on real quantum hardware and in clinical settings is needed.
This hybrid quantum–classical image preprocessing pipeline can help neural networks classify brain tumors from MRI scans with high accuracy while reaching key accuracy milestones in fewer training epochs.
Potential Applications
- Computer-aided brain tumor detection and classification
- Medical imaging software for classifying glioma, meningioma, and pituitary tumors
- Quantum machine-learning research for healthcare and diagnostic imaging
- Future diagnostic research using higher-resolution or multimodal medical images and quantum hardware
Benefits and Advantages
- Adaptive quantum–classical fusion achieved the strongest reported validation accuracy: 98.871% ± 0.30%
- Reached 90%, 95%, and 98% accuracy in fewer epochs than the classical baseline
- Combines quantum-derived nonlinear features with established classical image features
- Simulated depolarizing noise up to 10% did not shift validation results outside the methods’ confidence intervals
- Reported consistent performance across randomized data splits
For more information about this opportunity, please see