CONVERTING LOW-DOSE TO HIGHER DOSE 3D TOMOSYNTHESIS IMAGES THROUGH MACHINE-LEARNING PROCESSES

• US 20170071562A1
• Filed: 11/23/2016
• Published: 03/16/2017
• Est. Priority Date: 01/15/2014
• Status: Active Grant

• Current Assignee: Alarma Systems Incorporated
• Original Assignee: Alarma Systems Incorporated
• Inventors: SUZUKI, Kenji
• Granted Patent: US 10,610,182 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: A61B 6/025 Tomosynthesis
• CPC Class Codes: A61B 6/461 Displaying means of special interest
• CPC Class Codes: A61B 6/502 for diagnosis of breast, i.e. mammography
• CPC Class Codes: A61B 6/5205 involving processing of raw data to produce diagnostic data
• CPC Class Codes: A61B 6/5211 involving processing of medical diagnostic data
• CPC Class Codes: A61B 6/542 involving control of exposure
• CPC Class Codes: G06F 18/217 Validation; Performance evaluation; Active pattern learning techniques
• CPC Class Codes: G06T 11/005 Specific pre-processing for tomographic reconstruction, e.g. calibration, source positioning, rebinning, scatter correction, retrospective gating
• CPC Class Codes: G06T 2207/10116 X-ray image
• CPC Class Codes: G06T 2207/20081 Training; Learning
• CPC Class Codes: G06T 2207/30068 Mammography; Breast
• CPC Class Codes: G06T 5/00 Image enhancement or restoration
• CPC Class Codes: G06T 5/50 using two or more images, e.g. averaging or subtraction
• CPC Class Codes: G06T 7/0014 using an image reference approach
• CPC Class Codes: G06V 10/776 Validation; Performance evaluation
• CPC Class Codes: G16H 50/20 for computer-aided diagnosis, e.g. based on medical expert systems

Abstract

A method and system for converting low-dose tomosynthesis projection images or reconstructed slices images with noise into higher quality, less noise, higher-dose-like tomosynthesis reconstructed slices, using of a trainable nonlinear regression (TNR) model with a patch-input-pixel-output scheme called a pixel-based TNR (PTNR). An image patch is extracted from an input raw projection views (images) of a breast acquired at a reduced x-ray radiation dose (lower-dose), and pixel values in the patch are entered into the PTNR as input. The output of the PTNR is a single pixel that corresponds to a center pixel of the input image patch. The PTNR is trained with matched pairs of raw projection views (images together with corresponding desired x-ray radiation dose raw projection views (images) (higher-dose). Through the training, the PTNR learns to convert low-dose raw projection images to high-dose-like raw projection images. Once trained, the trained PTNR does not require the higher-dose raw projection images anymore. When a new reduced x-ray radiation dose (low dose) raw projection images is entered, the trained PTNR outputs a pixel value similar to its desired pixel value, in other words, it outputs high-dose-like raw projection images where noise and artifacts due to low radiation dose are substantially reduced, i.e., a higher image quality. Then, from the “high-dose-like” projection views (images), “high-dose-like” 3D tomosynthesis slices are reconstructed by using a tomosynthesis reconstruction algorithm. With the “virtual high-dose” tomosynthesis reconstruction slices, the detectability of lesions and clinically important findings such as masses and microcalcifications can be improved.

Claims

1. A x-ray breast imaging system comprising:

a low x-ray dose breast imaging structure configured to image a patient'"'"'s breast with an x-ray beam to provide low-dose x-ray breast images each taken at an x-ray dose substantially below a standard x-ray dose for an otherwise comparable x-ray image and having image quality below that of an otherwise comparable breast image taken at said standard x-ray dose;

a computer-implemented processor configured to apply computer processing to the low-dose x-ray images, said processing using training with breast images that have the image quality of comparable breast images taken at least at said standard x-ray dose, to thereby convert the low-dose x-ray images to respective higher-quality x-ray breast images comparable in image quality to otherwise comparable breast images taken at said standard x-ray dose; and

a display configured to display said higher-quality x-ray images of the breast or breast images derived from said higher-quality breast images.

2. The x-ray breast imaging system of claim 1, in which the breast imaging structure is configured to take the low-dose x-ray images at doses corresponding to a dose that is between approximately 90% and 10% of a standard x-ray dose for otherwise comparable x-ray images.

3. The x-ray breast imaging system of claim 1, in which the low-dose x-ray images are two-dimensional tomosynthesis projection images taken at respective angles of an imaging x-ray beam with respect to the breast and the higher-quality breast images are higher-quality versions of said projection images, and in which the computer-implemented image processor further includes a computer-implemented image reconstruction system configured to process the higher-quality projection images into breast slice images comprising said breast images derived from said higher-quality breast images.

4. The x-ray breast imaging system of claim 1 in which the breast imaging structure is configured to take each of the low-dose x-ray images at an x-ray dose that is between 50% and 10% of the standard x-ray dose.

5. The x-ray breast imaging system of claim 1 in which the processor is configured to apply non-linear regression processing to the low-dose x-ray images and thereby convert them to the higher-quality images.

6. The x-ray breast imaging system of claim 1 in which the processor is configured to process the low-dose images by processing each of respective multi-pixel patches of the low-dose images into a single pixel of the higher-quality images.

7. The x-ray breast imaging system of claim 1 in which the processor is further configured to extract morphologic elements corresponding to impulse areas of the low-dose images and use the extracted elements to improve converting the low-dose images into the higher-quality images.

8. A method of processing a mammogram, comprising:

obtaining input lower-quality breast images from an x-ray breast imaging system;

acquiring plural image patches from the input images;

entering the image patches into a computer-implemented, trainable regression model as input and obtaining from the model output pixel values corresponding to respective image patches; and

arranging the output pixel values from the regression model into output breast images of higher image quality than the input images.

9. The method of claim 8, wherein the input images are tomosynthesis projection images taken at x-ray doses below a standard x-ray dose for otherwise comparable images.

10. The method of claim 8, wherein the input images are obtained from one or more of a breast tomosynthesis system, a mammography system, computer storage, a viewing workstation, a picture archiving and communication system, cloud computing, website, and the Internet.

11. The method of claim 10, wherein the tomosynthesis system or the mammography system is operating in a low-dose mode.

12. The method of claim 8, wherein the trainable regression model is a previously-trained regression model.

13. The method of claim 8, wherein the trainable regression model is at least one of an artificial neural network regression model, a support vector regression model, a nonlinear Gaussian process regression model, and a machine-learning regression model.

14. The method of claim 8, wherein the trainable regression model is a trainable regression model that was trained with lower image quality images and paired higher image quality images.

15. The method of claim 14, wherein the lower-quality images are lower-dose tomosynthesis projection images or mammograms, and the higher-quality images are higher-dose tomosynthesis projection images or mammograms.

16. The method of claim 8, further comprising:

extraction of morphologic elements that extracts morphologic elements from the input images and acquiring

the plural image patches from both the input images and from the morphologic-elements-extracted images.

17. The method of claim 16, wherein the extraction of morphologic elements includes extraction of small high brightness regions.

18. The method of claim 16, wherein the plural image patches acquired from the input images are equal to or larger than the plural image patches acquired from the morphologic-elements-extracted image.

19. A method of processing x-ray breast images, comprising:

obtaining pairs of an input breast image and a desired breast image from a system;

acquiring plural image patches from the input images;

entering the image patches into a computer-implemented, trainable regression model as input to convert them into respective output pixel values;

calculating a difference between the output pixel values from the trainable regression model and corresponding desired pixel values from the paired desired breast images; and

adjusting parameters in the trainable regression model based on the calculated difference.

20. The method of claim 19, wherein the input image of a pair is a tomosynthesis projection image taken at an x-ray dose below a standard dose and the desired breast image is a projection tomosynthesis image taken at the standard dose or at a dose higher than the standard dose.

21. The method of claim 19, wherein the trainable regression model is one of an artificial neural network regression model, a support vector regression model, nonlinear Gaussian process regression model, and a machine-learning regression model.

22. The method of claim 19, wherein the calculated difference is a mean absolute error between output pixel values and corresponding desired pixel values, or a mean squared error between output pixel values and corresponding desired pixel values.

23. The method of claim 19, wherein the adjusting parameters in the trainable regression model comprise at least one of an error-back propagation algorithm, a steepest descent method, Newton'"'"'s algorithm, and an optimization algorithm.

24. The method of claim 19, further comprising:

extraction of morphologic elements that extracts morphologic elements from the input images; and

wherein the plural image patches are acquired from both the input images and the morphologic-elements-extracted images.

25. The method of claim 24, wherein the extraction of morphologic elements includes extraction of small high brightness regions.

26. A system comprising:

a source of lower image quality input breast x-ray images;

a computer-implemented processor configured to acquire plural image patches from the input images;

said processor being further configured to apply a trained regression model processing to the acquired image patches and provide output pixel values each corresponding to a respective image patch;

said processor being further configured to arrange the output pixel values from the regression model into output breast images of higher image quality than the respective input images; and

a display associated with the processor to receive and display the output images.

27. The system of claim 26, in which said processor is configured to acquire respective patches from plural regions in which the input images are segmented, and is further configured to apply differently trained regression models to the patches from the respective regions.

29. The system of claim 26, in which said model facility is configured to acquire respective patches from plural regions in which the input image and the desired image are segmented, and to apply respective regression model processing to the patches from the respective regions.

28. A system comprising:

a source configured to provide a pair of an input breast x-ray image and a desired breast image;

a computer-implemented trainable regression model facility configured to acquire plural image patches from the input image and apply regression model processing thereto to produce an output image, calculate a difference between the desired image and the output image and change parameters of the regression model to reduce the difference, and repeat the steps of applying the regression model, calculating the difference and changing parameters until a threshold condition is met, and outputting a final output breast image upon meeting the threshold condition; and

a display facility selectively displaying the output breast image.