# US 11,908,174 B2 · Methods and systems for image selection

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

- Number: US 11,908,174 B2
- Document ID: US11908174B2
- Status: Granted
- Kind: B2
- Application number: 17/646,609
- Filed: 2021-12-30
- Published: 2023-07-20
- Granted: 2024-02-20
- Earliest claimed: 2021-12-30
- Assignees: GE Precision Healthcare LLC
- Applicants: GE Precision Healthcare LLC
- Inventors: Mathieu Bedez; Marouane Belhaj; Vincent Pangon
- Web page: https://patentgrep.com/patent/US11908174B2

## CPC

- G06V 10/7715: {Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods}
- G06T 11/00: Two-dimensional [2D] image generation
- G06V 10/22: by selection of a specific region containing or referencing a pattern; Locating or processing of specific regions to guide the detection or recognition
- G06V 10/25: Determination of region of interest [ROI] or a volume of interest [VOI]
- G06V 10/764: using classification, e.g. of video objects
- G06V 10/82: using neural networks
- G06V 20/70: Labelling scene content, e.g. deriving syntactic or semantic representations
- G06V 40/10: Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G16H 30/40: for processing medical images, e.g. editing
- G16H 50/20: for computer-aided diagnosis, e.g. based on medical expert systems
- G06T 2210/21: Collision detection, intersection
- G06T 2210/41: Medical
- G06V 2201/03: Recognition of patterns in medical or anatomical images
- G16H 50/70: for mining of medical data, e.g. analysing previous cases of other patients
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- 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

Various methods and systems are provided for automatically classifying a plurality of image slices using body region bounding boxes identified from a localizer image. In one embodiment, a localizer image may be mapped to a plurality of bounding boxes, corresponding to a plurality of body regions, using a trained machine learning model. Coordinates of the plurality of bounding boxes may be used to determine body region boundaries, such that the body regions are non-intersecting and coherent. The body regions identified in the localizer image may then be correlated to image slice ranges, and image slices within each image slice range may be labeled as belonging to the corresponding body region.

## Claims

### Claim 1

A method comprising:
receiving a localizer image of an imaging subject and a plurality of image slices corresponding to the localizer image;
extracting a plurality of features from the localizer image;
mapping the plurality of features to one or more bounding boxes for one or more pre-determined regions of the imaging subject using a trained machine learning model;
resolving boundaries of the one or more bounding boxes to produce a plurality of non-intersecting regions by extracting boundary box borders and determining if the boundary box borders overlap; and
responding to a region of the plurality of non-intersecting regions being a target region by:
correlating coordinates of the region from a coordinate system of the localizer image to a coordinate system of the plurality of image slices to produce a target image slice coordinate range; and
labeling a subset of the plurality of image slices as belonging to the target region of the imaging subject based on the target image slice coordinate range.

### Claim 2 (depends on 1)

The method of claim 1, wherein the one or more bounding boxes include one or more of a head bounding box, a chest bounding box, an abdomen bounding box, a pelvis bounding box, and a leg bounding box.

### Claim 3 (depends on 2)

The method of claim 2, wherein resolving boundaries of the one or more bounding boxes to produce the plurality of non-intersecting regions, comprises one or more of:
setting a head/shoulder boundary as a bottom of the head bounding box;
setting a shoulder/chest boundary as a top of the chest bounding box;
setting a chest/abdomen boundary as a lesser of a bottom of the chest bounding box and a top of the abdomen bounding box;
setting an abdomen/pelvis boundary as a top of the pelvis bounding box, if the pelvis bounding box is present, else setting the abdomen/pelvis boundary as the bottom of the abdomen bounding box; and
setting a pelvis/leg boundary as a lesser of a bottom of the pelvis bounding box and a top of the leg bounding box.

### Claim 4 (depends on 1)

The method of claim 1, wherein the plurality of non-intersecting regions include one or more of a head region, a shoulder region, a chest region, an abdomen region, a pelvis region, and a leg region.

### Claim 5 (depends on 1)

The method of claim 1, wherein correlating coordinates of the region from the coordinate system of the localizer image to the coordinate system of the plurality of image slices to produce the target image slice coordinate range, comprises, mapping a first boundary and a second boundary of the region from a first axis in the coordinate system of the localizer image, to a second axis in the coordinate system of the plurality of image slices, to produce the target image slice coordinate range, wherein the plurality of image slices are indexed based on distance along the second axis.

### Claim 6

A method comprising:
receiving a localizer image of an imaging subject and a plurality of image slices corresponding to the localizer image;
extracting a plurality of features from the localizer image;
mapping the plurality of features to one or more bounding boxes for one or more pre-determined regions of the imaging subject using a trained machine learning model;
resolving boundaries of the one or more bounding boxes to produce a plurality of non-intersecting regions; and
responding to a region of the plurality of non-intersecting regions being a target region by:
correlating coordinates of the region from a coordinate system of the localizer image to a coordinate system of the plurality of image slices to produce a target image slice coordinate range; and
labeling a subset of the plurality of image slices as belonging to the target region of the imaging subject based on the target image slice coordinate range,
wherein correlating coordinates of the region from the coordinate system of the localizer image to the coordinate system of the plurality of image slices to produce the target image slice coordinate range, comprises, mapping a first boundary and a second boundary of the region from a first axis in the coordinate system of the localizer image, to a second axis in the coordinate system of the plurality of image slices, to produce the target image slice coordinate range, wherein the plurality of image slices are indexed based on distance along the second axis;
wherein distance along the first axis is measured in pixels, and distance along the second axis is measured in millimeters, and wherein a relationship between distance along the first axis and distance along the second axis is given by:
Z
k
=
Z
ref
+
Y
k
*
Spacing
Z
Y
*
cos
⁢
θ
where Zk is a distance of a point k along the second axis, Zref is an offset between an origin of the first axis and an origin of the second axis, Yk is a distance of the point k along the first axis, Spacing is a ratio between distance along the second axis and distance along the first axis, and θ is an angle between the first axis and the second axis, wherein when θ equals 0, the first axis is parallel to the second axis.

### Claim 7 (depends on 1)

The method of claim 1, the method further comprising:
displaying the subset of the plurality of image slices via a display device.

### Claim 8

A method, comprising:
receiving a localizer image;
pre-processing the localizer image to form a pre-processed image based on an architecture of a trained machine learning model;
mapping the pre-processed image to one or more bounding boxes;
resolving boundaries of the one or more bounding boxes to produce a plurality of non-intersecting regions by extracting boundary box borders and determining if the boundary box borders overlap; and
classifying a plurality of axial images into one or more classes based on the plurality of non-intersecting regions.

### Claim 9 (depends on 8)

The method of claim 8, wherein the localizer image and the plurality of axial images are acquired via a CT scanner.

### Claim 10 (depends on 8)

The method of claim 8, wherein pre-processing the localizer image comprises:
extracting pixel intensity data from the localizer image;
normalizing the pixel intensity data to a pre-determined range; and
duplicating the normalized pixel intensity data into three channels, corresponding to RGB channels of a natural image.

### Claim 11 (depends on 8)

The method of claim 8, wherein the trained machine learning model is YOLOv3.

### Claim 12 (depends on 8)

The method of claim 8, wherein classifying the plurality of axial images comprises:
selecting at least one region of the plurality of non-intersecting regions;
extracting coordinates of the at least one region;
converting coordinates of the at least one region into a slice location range; and
assigning each image slice within the slice location range to the at least one region.

### Claim 13

A system comprising:
a non-transitory memory, wherein the non-transitory memory stores instructions, and a trained machine learning model;
a display device; and
a processor, wherein the processor is communicably coupled to the non-transitory memory and the display device, and wherein the processor, when executing the instructions, is configured to:
receive a localizer image of an imaging subject and a plurality of image slices corresponding to the localizer image;
map the localizer image to one or more bounding boxes using the trained machine learning model;
post-process the one or more bounding boxes to produce a coherent set of bounding boxes;
resolve boundaries of the coherent set of bounding boxes to produce a plurality of non-intersecting regions by extracting boundary box borders and determining if the boundary box borders overlap;
correlate coordinates of a region of the plurality of non-intersecting regions from a coordinate system of the localizer image to a coordinate system of the plurality of image slices to produce an image slice coordinate range;
label a subset of the plurality of image slices as belonging to the region of the imaging subject based on the image slice coordinate range; and
displaying the subset of the plurality of image slices via the display device.

### Claim 14 (depends on 13)

The system of claim 13, wherein the processor is configured to post-process the one or more bounding boxes to produce the coherent set of bounding boxes by:
responding to detection of a first bounding box assigned to a first body region, and a second bounding box also assigned to the first body region by:
keeping the first bounding box in response to a determination that the first bounding box has a higher confidence score than the second bounding box, the confidence score indicating probability that the respective bounding box includes an object; and
discarding the second bounding box.

### Claim 15 (depends on 13)

The system of claim 13, wherein the processor is further configured to post-process the one or more bounding boxes to produce the coherent set of bounding boxes by:
removing duplicate bounding boxes based on a pre-determined body region order.

### Claim 16 (depends on 13)

The system of claim 13, wherein the processor is configured to resolve boundaries of the coherent set of bounding boxes to produce the plurality of non-intersecting regions, by, extracting coordinates of one or more of:
a bottom of a head bounding box;
a top of a chest bounding box;
a bottom of the chest bounding box;
a top of an abdomen bounding box;
a bottom of the abdomen bounding box;
a top of a pelvis bounding box;
a bottom of the pelvis bounding box; and
a top of a legs bounding box.

### Claim 17 (depends on 16)

The system of claim 16, wherein the processor is configured to resolve boundaries of the coherent set of bounding boxes to produce the plurality of non-intersecting regions, by:
setting a head/shoulder boundary as the bottom of the head bounding box;
setting a shoulder/chest boundary as the top of the chest bounding box;
setting a chest/abdomen boundary as a lesser of the bottom of the chest bounding box and the top of the abdomen bounding box;
setting an abdomen/pelvis boundary as the top of the pelvis bounding box, if the pelvis bounding box is present, else setting the abdomen/pelvis boundary as the bottom of the abdomen bounding box; and
setting a pelvis/leg boundary as a lesser of the bottom of the pelvis bounding box and the top of the legs bounding box.

### Claim 18 (depends on 13)

The system of claim 13, wherein the processor is configured to correlate coordinates of the region of the plurality of non-intersecting regions from the coordinate system of the localizer image to the coordinate system of the plurality of image slices to produce the image slice coordinate range by:
mapping a first boundary and a second boundary of the region from a first axis in the coordinate system of the localizer image, to a second axis in the coordinate system of the plurality of image slices, to produce the image slice coordinate range.

### Claim 19

A system comprising:
a non-transitory memory, wherein the non-transitory memory stores instructions, and a trained machine learning model;
a display device; and
a processor, wherein the processor is communicably coupled to the non-transitory memory and the display device, and wherein the processor, when executing the instructions, is configured to:
receive a localizer image of an imaging subject and a plurality of image slices corresponding to the localizer image;
map the localizer image to one or more bounding boxes using the trained machine learning model;
post-process the one or more bounding boxes to produce a coherent set of bounding boxes;
resolve boundaries of the coherent set of bounding boxes to produce a plurality of non-intersecting regions;
correlate coordinates of a region of the plurality of non-intersecting regions from a coordinate system of the localizer image to a coordinate system of the plurality of image slices to produce an image slice coordinate range;
label a subset of the plurality of image slices as belonging to the region of the imaging subject based on the image slice coordinate range; and
displaying the subset of the plurality of image slices via the display device;
wherein the processor is configured to correlate coordinates of the region of the plurality of non-intersecting regions from the coordinate system of the localizer image to the coordinate system of the plurality of image slices to produce the image slice coordinate range by:
mapping a first boundary and a second boundary of the region from a first axis in the coordinate system of the localizer image, to a second axis in the coordinate system of the plurality of image slices, to produce the image slice coordinate range,
wherein a mapping between the first axis and the second axis is given by:
Z
k
=
Z
ref
+
Y
k
*
Spacing
Z
Y
*
cos
⁢
θ
where Zk is a distance of a point k along the second axis, Zref is an offset between an origin of the first axis and an origin of the second axis, Yk is a distance of the point k along the first axis, Spacing is a ratio between distance along the second axis and distance along the first axis, and θ is an angle between the first axis and the second axis.

### Claim 20 (depends on 13)

The system of claim 13, wherein the processor is configured to label the subset of the plurality of image slices as belonging to one of a head region, a shoulder region, a chest region, an abdomen region, a pelvis region, and a legs region.