# US 12,178,621 B2 · Fast and low memory usage convolutional neural network for tomosynthesis data processing and feature detection

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 12,178,621 B2
- Document ID: US12178621B2
- Status: Granted
- Kind: B2
- Application number: 17/667,764
- Filed: 2022-02-09
- Published: 2023-08-10
- Granted: 2024-12-31
- Earliest claimed: 2022-02-09
- Assignees: GE Precision Healthcare LLC
- Applicants: GE Precision Healthcare LLC
- Inventors: Sylvain Bernard; Vincent Bismuth
- Web page: https://patentgrep.com/patent/US12178621B2

## CPC

- A61B 6/025: {Tomosynthesis}
- G06T 7/0012: {Biomedical image inspection}
- A61B 6/502: {for diagnosis of breast, i.e. mammography}
- G06T 7/0008: {checking presence/absence}
- G06T 12/20: Inverse problem, i.e. transformations from projection space into object space
- G16H 30/40: for processing medical images, e.g. editing
- G06T 2207/10081: Computed x-ray tomography [CT]
- G06T 2207/20084: Artificial neural networks [ANN]
- G06T 2207/30004: Biomedical image processing
- G06T 2207/30068: Mammography; Breast
- G06T 2207/30096: Tumor; Lesion
- A61B: DIAGNOSIS; SURGERY; IDENTIFICATION (usefulness limited to only animals A61D)
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G16H: HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA

## Abstract

A convolutional neural network (CNN) is employed in an automated feature and/or anomaly detection system for analyzing images provided by X-ray imaging system. The automated detection system operates in a manner that reduces the number of full-resolution CNN convolution layers required in order to speed up the network inference and learning processes for the detection system. To do so, the detection system utilizes as an input a more compact representation of the tomographic data to alleviate the CNN memory footprint and computation time issues in prior art X-ray systems.

## Claims

### Claim 1

A method for detecting an anomaly in one or more images obtained from an X-ray imaging procedure on an object, the method comprising the steps of:
a. providing a computing system on which an automatic anomaly detection system is provided, the anomaly detection system comprising:
i. a processing unit configured to receive and process projection images, the processing unit including a convolutional neural network;
ii. an electronic information storage device operably connected to the processing unit and containing operational instructions for the convolutional neural network; and
b. supplying a compact data representation input of the projection images of an object of interest to the automatic anomaly detection system;
c. analyzing the compact data representation input with the automatic anomaly detection system in at least one layer of the convolutional neural network to obtain an anomaly detection result; and
d. outputting an anomaly detection result from the automatic anomaly detection system,
wherein filters for the at least one layer of the convolutional neural network employed on the compact data representation input are at least partially learned by the automatic anomaly detection system.

### Claim 2 (depends on 1)

The method of claim 1, wherein the at least one layer of the convolutional neural network is one of a filtering operation, a reconstruction operation, a slabbing operation, a 2D synthetic image generation and combinations thereof.

### Claim 3 (depends on 1)

The method of claim 1, wherein the compact data representation input is the set of projection images.

### Claim 4 (depends on 3)

The method of claim 3, wherein the step of analyzing the compact data representation input comprises the steps of:
a. performing a filtering operation on the set of projection images to form filtered projection images;
b. performing a reconstruction operation on the filtered projection images to form filtered slices.

### Claim 5 (depends on 3)

The method of claim 3, wherein the step of analyzing the compact data representation input comprises the steps of:
a. performing a filtering operation on the set of projection images to form filtered projection images;
b. performing a reconstruction operation on the filtered projection images to form filtered slices; and
c. performing a slabbing operation on the filtered slices to obtain filtered slabs.

### Claim 6 (depends on 3)

The method of claim 3, wherein the step of analyzing the compact data representation input comprises the steps of:
a. performing a filtering operation on the set of projection images to form filtered projection images;
b. performing a reconstruction operation on the filtered projection images to form filtered slices;
c. performing a slabbing operation on the filtered slices to obtain filtered slabs; and
d. performing a synthetic 2D image generation operation on the filtered slices to obtain filtered 2D synthetic images.

### Claim 7 (depends on 6)

The method of claim 6, further comprising the steps of:
a. performing one or more detection operations on the filtered slabs to obtain a first lesion type output; and
b. performing one or more detection convolution operations on the filtered 2D synthetic images to obtain a second lesion type output.

### Claim 8 (depends on 3)

The method of claim 3, wherein the step of analyzing the compact data representation input comprises the steps of:
a. performing a filtering operation on the set of projection images to form filtered projection images;
b. performing a reconstruction operation on the filtered projection images to form filtered slices; and
c. performing a synthetic 2D image generation operation on the filtered slices to obtain filtered 2D synthetic images.

### Claim 9 (depends on 3)

The method of claim 3, wherein the step of analyzing the compact data representation input further comprises the steps of:
a. performing a filtering operation on the set of projection images to form filtered projection images;
b. performing a backprojection on the filtered projection images to form a backprojected filtered projection set;
c. performing an aggregation operation on the filtered backprojected filtered projection set to form a slice image set;
d. performing at least one of:
i. a reconstruction operation on the slice image set to form one or more viewable slices,
ii. a slabbing operation on the slice image set to form a slab image set and a convolutional operation on the slab image set to form one or more viewable slabs, or
iii. combinations thereof;
e. performing a detection operation on the slice image set or slab image set to form an anomaly detection result; and
f. outputting the one or more viewable slices, the one or more viewable slabs, and combinations thereof in conjunction with the anomaly detection result.

### Claim 10 (depends on 1)

The method of claim 1, wherein the compact data representation input is a set of precomputed synthetic 2D image and further comprises a step of performing a filtering convolution operation on a synthetic 2D images to obtain filtered synthetic 2D images.

### Claim 11 (depends on 1)

The method of claim 1, wherein the compact data representation input is a set of precomputed slices obtained from the projection images.

### Claim 12 (depends on 11)

The method of claim 11, wherein the step of analyzing the compact data representation input comprises the steps of:
a. performing a filtering operation on the set of precomputed slices to form filtered slices; and
b. performing a slabbing operation on the filtered slices to form filtered slabs.

### Claim 13 (depends on 1)

The method of claim 1, wherein the compact data representation input is a set of precomputed slabs and further comprises a step of performing a filtering operation on the set of slabs to form filtered slabs.

### Claim 14

An X-ray imaging system comprising:
a. an X-ray source configured to emit radiation beams towards an object;
b. an X-ray detector movably aligned with the X-ray source to receive the radiation beams from the X-ray source and generate image data;
c. a processing unit operably connected to the X-ray source and the X-ray detector to control the movement and operation of the X-ray source and X-ray detector, the processing unit configured to receive and process image data from the X-ray detector to form projection images;
d. a display operably connected to the processing unit for presenting information to a user;
e. a user interface operably connected to the processing unit to enable user input to the processing unit;
f. an automatic anomaly detection system operably connected to the controller; and
g. an electronic information storage device operably connected to the processing unit,
wherein the automatic anomaly detection system is formed of a convolutional neural network having and operational instructions contained on the storage device to be accessed by the processing unit, the automatic anomaly detection system configured to receive a compact data representation input of the projection images and to analyze the compact data representation input to obtain an anomaly detection result, and wherein filters for the at least one layer of the convolutional neural network employed by the automatic anomaly detection system on the compact data representation are at least partially learned by the automatic anomaly detection system.

### Claim 15 (depends on 14)

The method of claim 14, wherein the at least one layer of the convolutional neural network is one of a filtering operation, a reconstruction operation, a slabbing operation, a 2D synthetic image generation and combinations thereof.

### Claim 16 (depends on 14)

The X-ray imaging system of claim 14, wherein the compact data representation input is the set of projection images.

### Claim 17 (depends on 16)

The X-ray imaging system of claim 16, wherein the automatic anomaly detection system is configured to analyze the compact data representation input by performing a filtering operation on the set of projection images to form filtered projection images, and performing a reconstruction operation on the filtered projection images to form filtered slices.

### Claim 18 (depends on 17)

The X-ray imaging system of claim 17, wherein the automatic anomaly detection system is configured to analyze the compact data representation input by performing a slabbing operation on the filtered slices to obtain filtered slabs.

### Claim 19 (depends on 17)

The X-ray imaging system of claim 17, wherein the automatic anomaly detection system is configured to analyze the compact data representation input by performing a synthetic 2D image generation operation on the filtered slices to obtain filtered 2D synthetic images.

### Claim 20 (depends on 17)

The X-ray imaging system of claim 17, wherein the automatic anomaly detection system is configured to:
a. analyze the compact data representation input by:
i. performing a slabbing operation on the filtered slices to obtain filtered slabs; and
ii. performing a synthetic 2D image generation operation on the filtered slices to obtain filtered 2D synthetic image; and
b. perform one or more detection operations on the filtered slabs to obtain a first lesion type output; and
c. perform one or more detection operations on the filtered synthetic 2D to obtain a second lesion type output.

### Claim 21 (depends on 16)

The X-ray imaging system of claim 16, wherein the automatic anomaly detection system is configured to:
a. perform a filtering operation on the set of projection images to form filtered projection images;
b. perform a backprojection on the filtered projection images to form filtered slices;
c. perform an aggregation operation on the filtered slices to form an 3D volume image set;
d. perform at least one of a reconstruction operation to form a viewable slice volume from the 3D volume image set, and a slabbing operation to form a viewable slab volume from the 3D volume image set;
e. perform a detection operation on the 3D volume image set to form an anomaly detection result; and
f. output the viewable slices, viewable slabs, and combinations thereof in conjunction with the anomaly detection result on the display.