Method for Creating Super Resolution Medical Images

The deep learning algorithm retrospectively improves the resolution of routine medical images (e.g., CT, MRI, etc.) without the need for increased radiation or scan time.

Technology:
Clinical medical images often have low resolution due to radiation-dose or scan-time restrictions. Inventors developed a probabilistic deep learning model to improve resolution of routinely acquired medical images. The method is particularly suited for situations where high resolution imaging using MRI, CT, or PET is not clinically feasible. This method produces significantly better performance in terms of superior image quality, greater quantitative-measurement accuracy, and low bias compared to other methods using convolutional neural networks and generative adversarial networks (GAN). For example, this method allows the measurement of microstructural trabecular bone thickness, separation, number, and strength which can assist in evaluating bone health without the need for a bone biopsy.

Advantages: