Quality of Medical Images Using Multi-Contrast and Deep Learning

Abstract

A method of improving diagnostic and functional imaging is provided by obtaining at least two input images of a subject, using a medical imager, where each input image includes a different contrast, generating a plurality of copies of the input images using non-local mean (NLM) filtering, using an appropriately programmed computer, where each input image copy of the subject includes different spatial characteristics, obtaining at least one reference image of the subject, using the medical imager, where the reference image includes imaging characteristics that are different form the input images of the subject, training a deep network model, using data augmentation on the appropriately programmed computer, to adaptively tune model parameters to approximate the reference image from an initial set of the input and reference images, with the goal of outputting an improved quality image of other sets of low SNR low resolution images, for analysis by a physician.

Description

FIELD OF THE INVENTION[0001]The invention relates to medical imaging. More specifically the invention relates to improving the quality of medical images using multi-contrast imaging, multi-lateral filters, and deep learning methods.BACKGROUND OF THE INVENTION[0002]With medical image denoising, multiple methods have been proposed including, Gaussian filtering, wavelet filtering, and non-local means (NLM) algorithms, where experimentation has shown the NLM (possibly combining Wavelet) is the superior method. However, all these methods still share some disadvantages such as the dependency of parameter tuning for different images. In one instance, a proposed method used the redundancy in and relationships of multi-contrast images as a prior for image denoising. Related works have been used to combine a blurry and a noisy pair of images for CMOS sensors and cameras. A further implementation used Group-Sparsity representation for image denoising, which also used the multi-contrast informa...

AI Technical Summary

Benefits of technology

The patent text describes a method that shortens the time needed for medical imaging by using a deep learning network to approximate a reference image from a set of input images. The method uses non-local mean filtering and data augmentation to train the network model, which can output an improved quality image without the need for a reference image. The method can also use multi-contrast information to improve regularized denoising and over-smoothing avoidance. The main technical effect of this method is to decrease the time required for medical imaging while maintaining image quality.

Problems solved by technology

However, all these methods still share some disadvantages such as the dependency of parameter tuning for different images.

However, it is not clear that the model can be better trained for medical imaging, since there are much fewer data sets available for training, and deep networks typically need thousands or millions of samples due to the number of parameters in the model.

Further, it is not clear what network structure is the best for medical images due to the intrinsic properties of medical images in that they are not the same as recognizing common objects within photos.

Finally, it is not fully known how to make sure the model does not introduce artifacts that are not in the image or miss the detail of pathology that the model has not seen from the training data.

Additionally, the relatively few samples can already capture sufficient variability of natural images patches.

SR works try to achieve better performance for aesthetic perception but does not address the need to avoid while preserving details and pathology in for medical images.

However, ASL perfusion maps typically suffer from low SNR due to its signal subtraction.

However this repeating of the scans significantly increases the testing time.

However, acquiring different delays further increases the time cost and results in even lower SNR and resolution due to the time constraints.

Images