# US 2023/0274436 A1 · Automated Medical Image and Segmentation Quality Assessment for Machine Learning Tasks

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

- Number: US 2023/0274436 A1
- Document ID: US20230274436A1
- Status: Published application
- Kind: A1
- Application number: 17/680,799
- Filed: 2022-02-25
- Published: 2023-08-31
- Earliest claimed: 2022-02-25
- Assignees: International Business Machines Corporation
- Applicants: International Business Machines Corporation
- Inventors: Yi-Qing Wang; Giovanni John Jacques Palma
- Web page: https://patentgrep.com/patent/US20230274436A1

## CPC

- G06T 7/0016: {involving temporal comparison}
- G06T 7/10: Segmentation; Edge detection (motion-based segmentation G06T7/215)
- G06T 7/62: of area, perimeter, diameter or volume
- G06T 2207/10076: 4D tomography; Time-sequential 3D tomography
- G06T 2207/10081: Computed x-ray tomography [CT]
- G06T 2207/10088: Magnetic resonance imaging [MRI]
- G06T 2207/20081: Training; Learning
- G06T 2207/20084: Artificial neural networks [ANN]
- G06T 2207/30056: Liver; Hepatic
- G06T 2207/30096: Tumor; Lesion
- G06T 2207/30168: Image quality inspection
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL

## Abstract

Mechanisms for identifying inconsistencies between volumes of medical images are provided. A plurality of volumes of medical images are received, each having a plurality of medical images of an anatomical structure. A plurality of first representation data structures are generated, each corresponding to a volume and having dimensional measurements of the anatomical structure at various locations. A reference data structure is generated, for the anatomical structure based on the first representation data structures, having second dimensional measurements derived from the first dimensional measurements. A discrepancy is detected between a second representation data structure and the reference data structure based on a comparison and a notification of the discrepancy where the notification identifies a type of the discrepancy.

## Claims

### Claim 1

A method, in a medical image computer system, for identifying inconsistencies between volumes of medical images, the method comprising:
receiving, by the medical image computer system, a plurality of volumes of medical images, each volume of medical images comprising a plurality of medical images, of an anatomical structure of a patient, captured by medical imaging equipment;
generating, by the medical image computer system, a plurality of first representation data structures, each first representation data structure corresponding to a volume in the plurality of volumes of medical images, wherein each first representation data structure comprises a first dimensional measurements of the anatomical structure at various locations;
generating, by the medical image computer system, a reference data structure for the anatomical structure of the patient based on the first representation data structures, wherein the reference data structure comprises second dimensional measurements derived from the first dimensional measurements;
detecting, by the medical image computer system, a discrepancy between a second representation data structure and the reference data structure based on a comparison of the second representation data structure and the reference data structure; and
outputting, by the medical image computer system, a notification of the discrepancy to a computing system in response to detecting the discrepancy.

### Claim 2 (depends on 1)

The method of claim 1, wherein generating the reference data structure comprises:
selecting a first representation data structure from the plurality of first representation data structures to be a candidate reference data structure;
aligning other first representation data structures in the plurality of first representation data structures, to the selected first representation data structure; and
generating the reference data structure based on the aligned other first representation data structures.

### Claim 3 (depends on 2)

The method of claim 2, wherein generating the reference data structure based on the aligned other first representation data structures further comprises, iteratively until a stopping condition is reached:
generating a median reference data structure based on a median of the aligned other first representation data structures;
selecting the median reference data structure as the new reference data structure for a next iteration; and
aligning the first representation data structures to the new reference data structure, to thereby generate the aligned other first representation data structures for the next iteration.

### Claim 4 (depends on 3)

The method of claim 3, wherein the stopping condition is either the new reference data structure for the next iteration is stabilized with regard to a reference data structure of a previous iteration, or a predetermine number of iterations have been performed.

### Claim 5 (depends on 1)

The method of claim 1, wherein detecting the discrepancy comprises:
determining a distance between data points of the second representation data structure, at given locations of the anatomical structure, and data points of the reference data structure at the same given locations; and
classifying the discrepancy into one of a plurality of predetermined classifications based on the distances between the data points of the second representation data structure and the reference data structure.

### Claim 6 (depends on 1)

The method of claim 1, wherein detecting the discrepancy further comprises classifying the discrepancy as one of:
a span disparity in response to the comparison of the second representation data structure and the reference data structure indicating that the second representation data structure does not contain dimensional measurements for a span of the anatomical structure represented in the reference data structure;
a cross-sectional size disagreement in response to the comparison of the second representation data structure and the reference data structure indicating one or more cross-sectional sizes of the anatomical structure that are greater than or less than an expected cross-sectional size specified in the reference data structure for a same location of the anatomical structure; or
insufficient structure representation in response to a percentage of the anatomical structure represented in the second representation data structure being below a predetermined threshold.

### Claim 7 (depends on 1)

The method of claim 1, wherein classifying the discrepancy further comprises classifying a source of the discrepancy as one of:
a partial acquisition in response to the discrepancy being classified as a span disparity and there being no more medical images in a volume, corresponding to the second representation data structure, beyond an end point of a plot of a curve of data points of the second representation data structure;
a partial segmentation in response to the discrepancy being classified as a span disparity and there being more medical images in the volume, corresponding to the second representation data structure, beyond an end point of a plot of a curve of data points of the second representation data structure;
a medical procedure discrepancy in response to the discrepancy being classified as a cross-sectional size disagreement and one or more dimensional measurements in the second representation data structure being smaller than one or more corresponding dimensional measurements in the reference data structure; or
a medical condition discrepancy in response to the discrepancy being classified as a cross-sectional size disagreement and one or more dimensional measurements in the second representation data structure being larger than one or more corresponding dimensional measurements in the reference data structure.

### Claim 8 (depends on 1)

The method of claim 1, further comprising discarding the volume from inclusion in a training dataset of a machine learning computer model in response to detecting the discrepancy.

### Claim 9 (depends on 1)

The method of claim 1, further comprising outputting a notification indicating a potential faulty output of a trained machine learning computer model in response to detecting the discrepancy and a volume of medical images corresponding to the second representation data structure being input to the trained machine learning computer model.

### Claim 10 (depends on 1)

The method of claim 1, wherein the first representation data structures are scan profiles of the volumes in the plurality of volumes, wherein each volume has an associated scan profile and each scan profile comprises, for each medical image in a corresponding volume, a cross-sectional dimension of the anatomical structure at a given location corresponding to the medical image.

### Claim 11

A computer program product comprising a computer readable storage medium having a computer readable program stored therein, wherein the computer readable program, when executed on a medical image computer system, causes the medical image computer system to identify inconsistencies between volumes of medical images at least by:
receiving a plurality of volumes of medical images, each volume of medical images comprising a plurality of medical images, of an anatomical structure of a patient, captured by medical imaging equipment;
generating a plurality of first representation data structures, each first representation data structure corresponding to a volume in the plurality of volumes of medical images, wherein each first representation data structure comprises a first dimensional measurements of the anatomical structure at various locations;
generating a reference data structure for the anatomical structure of the patient based on the first representation data structures, wherein the reference data structure comprises second dimensional measurements derived from the first dimensional measurements;
detecting a discrepancy between a second representation data structure and the reference data structure based on a comparison of the second representation data structure and the reference data structure; and
outputting a notification of the discrepancy to a computing system in response to detecting the discrepancy.

### Claim 12 (depends on 11)

The computer program product of claim 11, wherein generating the reference data structure comprises:
selecting a first representation data structure from the plurality of first representation data structures to be a candidate reference data structure;
aligning other first representation data structures in the plurality of first representation data structures, to the selected first representation data structure; and
generating the reference data structure based on the aligned other first representation data structures.

### Claim 13 (depends on 12)

The computer program product of claim 12, wherein generating the reference data structure based on the aligned other first representation data structures further comprises, iteratively until a stopping condition is reached:
generating a median reference data structure based on a median of the aligned other first representation data structures;
selecting the median reference data structure as the new reference data structure for a next iteration; and
aligning the first representation data structures to the new reference data structure, to thereby generate the aligned other first representation data structures for the next iteration.

### Claim 14 (depends on 13)

The computer program product of claim 13, wherein the stopping condition is either the new reference data structure for the next iteration is stabilized with regard to a reference data structure of a previous iteration, or a predetermine number of iterations have been performed.

### Claim 15 (depends on 11)

The computer program product of claim 11, wherein detecting the discrepancy comprises:
determining a distance between data points of the second representation data structure, at given locations of the anatomical structure, and data points of the reference data structure at the same given locations; and
classifying the discrepancy into one of a plurality of predetermined classifications based on the distances between the data points of the second representation data structure and the reference data structure.

### Claim 16 (depends on 11)

The computer program product of claim 11, wherein detecting the discrepancy further comprises classifying the discrepancy as one of:
a span disparity in response to the comparison of the second representation data structure and the reference data structure indicating that the second representation data structure does not contain dimensional measurements for a span of the anatomical structure represented in the reference data structure;
a cross-sectional size disagreement in response to the comparison of the second representation data structure and the reference data structure indicating one or more cross-sectional sizes of the anatomical structure that are greater than or less than an expected cross-sectional size specified in the reference data structure for a same location of the anatomical structure; or
insufficient structure representation in response to a percentage of the anatomical structure represented in the second representation data structure being below a predetermined threshold.

### Claim 17 (depends on 16)

The computer program product of claim 16, wherein classifying the discrepancy further comprises classifying a source of the discrepancy as one of:
a partial acquisition in response to the discrepancy being classified as a span disparity and there being no more medical images in a volume, corresponding to the second representation data structure, beyond an end point of a plot of a curve of data points of the second representation data structure;
a partial segmentation in response to the discrepancy being classified as a span disparity and there being more medical images in the volume, corresponding to the second representation data structure, beyond an end point of a plot of a curve of data points of the second representation data structure;
a medical procedure discrepancy in response to the discrepancy being classified as a cross-sectional size disagreement and one or more dimensional measurements in the second representation data structure being smaller than one or more corresponding dimensional measurements in the reference data structure; or
a medical condition discrepancy in response to the discrepancy being classified as a cross-sectional size disagreement and one or more dimensional measurements in the second representation data structure being larger than one or more corresponding dimensional measurements in the reference data structure.

### Claim 18 (depends on 11)

The computer program product of claim 11, wherein the computer readable program further causes at least one of the medical image computer system to:
discard the volume from inclusion in a training dataset of a machine learning computer model in response to detecting the discrepancy; or
output a notification indicating a potential faulty output of a trained machine learning computer model in response to detecting the discrepancy and a volume of medical images corresponding to the second representation data structure being input to the trained machine learning computer model.

### Claim 19 (depends on 11)

The computer program product of claim 11, wherein the first representation data structures are scan profiles of the volumes in the plurality of volumes, wherein each volume has an associated scan profile and each scan profile comprises, for each medical image in a corresponding volume, a cross-sectional dimension of the anatomical structure at a location corresponding to the medical image.

### Claim 20

An apparatus comprising:
at least one processor; and
at least one memory coupled to the at least one processor, wherein the at least one memory comprises instructions which, when executed by the at least one processor, cause the at least one processor to identify inconsistencies between volumes of medical images at least by:
receiving a plurality of volumes of medical images, each volume of medical images comprising a plurality of medical images, of an anatomical structure of a patient, captured by medical imaging equipment;
generating a plurality of first representation data structures, each first representation data structure corresponding to a volume in the plurality of volumes of medical images, wherein each first representation data structure comprises a first dimensional measurements of the anatomical structure at various locations;
generating a reference data structure for the anatomical structure of the patient based on the first representation data structures, wherein the reference data structure comprises second dimensional measurements derived from the first dimensional measurements;
detecting a discrepancy between a second representation data structure and the reference data structure based on a comparison of the second representation data structure and the reference data structure; and
outputting a notification of the discrepancy to a computing system in response to detecting the discrepancy, wherein the notification identifies a type of the discrepancy.
