# US 2024/0177308 A1 · Series selection for computer vision machine learning models analyzing medical images

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

- Number: US 2024/0177308 A1
- Document ID: US20240177308A1
- Status: Published application
- Kind: A1
- Application number: 18/515,368
- Filed: 2023-11-21
- Published: 2024-05-30
- Earliest claimed: 2022-11-25
- Assignees: Aidoc Medical Ltd
- Applicants: Aidoc Medical Ltd
- Inventors: Ido REISS; Ronny Sharon LEVANDA; Uri Goren HORESH; Idan BASSUKEVITZ
- Web page: https://patentgrep.com/patent/US20240177308A1

## CPC

- G06T 7/0012: {Biomedical image inspection}
- G06V 10/776: Validation; Performance evaluation
- G06T 2207/20084: Artificial neural networks [ANN]
- G06T 2207/30004: Biomedical image processing
- G06V 2201/03: Recognition of patterns in medical or anatomical images
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING

## Abstract

There is provided a computer implemented method of scheduling analysis of at least one series of a study of medical images of a subject, comprising: predicting a time when at least one series of the study which is not yet available for processing, will be available for processing, predicting at least one parameter of the at least one series which is not yet available for processing, and obtaining the at least one parameter for series which are available, selecting a target series according to a combination of the predicted time and the at least one parameter, in response to the target series not yet available for processing, waiting for the target series to become available for processing, and in response to the target series being available for processing, feeding the target series into an image analysis machine learning model.

## Claims

### Claim 1

A computer implemented method of scheduling analysis of at least one series of a study of medical images of a subject, comprising:
predicting a time when at least one series of the study which is not yet available for processing, will be available for processing;
predicting at least one parameter of the at least one series which is not yet available for processing, and obtaining the at least one parameter for series which are available;
selecting a target series according to a combination of the predicted time and the at least one parameter;
in response to the target series not yet available for processing, waiting for the target series to become available for processing; and
in response to the target series being available for processing, feeding the target series into an image analysis machine learning model.

### Claim 2 (depends on 1)

The computer implemented method of claim 1, wherein the study includes at least one series which is not yet available for processing and at least one series which is available for processing.

### Claim 3 (depends on 1)

The computer implemented method of claim 1, further comprising:
predicting resource utilization of a computer implementing inference by the image analysis machine learning model for each series of the study, wherein the combination for selecting the target series further includes the predicted resource utilization.

### Claim 4 (depends on 1)

The computer implemented method of claim 1, further comprising:
predicting a performance of the image analysis machine learning model for each series of the study including not yet available series and available series;
wherein the combination for selecting the target series further includes the predicted performance.

### Claim 5 (depends on 1)

The computer implemented method of claim 1, wherein the at least one parameter includes metadata.

### Claim 6 (depends on 1)

The computer implemented method of claim 1, wherein the at least one parameter is selected from a group comprising: reconstruction plane, convolutional kernels, slice thickness, contrast phase, protocol name, reason for study, reason for the series, description of study, description of the series, contrast phase, anatomy depicted in the study, and reconstruction parameters.

### Claim 7 (depends on 1)

The computer implemented method of claim 1, wherein the selecting and the feeding are performed for each of a plurality of image analysis machine learning models according to a specific image analysis machine learning model of the plurality of image analysis machine learning models, wherein the target series is fed into the specific image analysis machine learning model.

### Claim 8 (depends on 1)

The computer implemented method of claim 1, further comprising computing at least one respective weight for the predicted time for a respective series and/or for the at least one parameter of the respective series, wherein the combination for the respective series is computed according to the at least one respective weight.

### Claim 9 (depends on 8)

The computer implemented method of claim 8, wherein the at least one weight for the subsequent series is dynamically adapted according to outcome of the image analysis machine learning model analyzing at least one preceding series.

### Claim 10 (depends on 1)

The computer implemented method of claim 1, further comprising computing a score for each series of the study as a function of the combination of the predicted time and the at least one parameter; and wherein selecting comprises selecting the target series according to the score.

### Claim 11 (depends on 10)

The computer implemented method of claim 10, further comprising: in response to a score of a certain series being higher than a score of a current series being processed by the image analysis machine learning model, interrupt processing of the current series by the image analysis machine learning model, and instructing processing of the certain series by the image analysis machine learning model.

### Claim 12 (depends on 10)

The computer implemented method of claim 10, further comprising: in response to a score of a predicted series which is not yet available for processing being higher than a score of a current series which is available for processing, not feeding the current series into the image analysis machine learning model and waiting for the predicted series to be available for processing.

### Claim 13 (depends on 10)

The computer implemented method of claim 10, further comprising:
for each series, computing a respective sub-score of a plurality of sub-scores for each of the predicted time, resource utilization of a computer implementing inference by the image analysis machine learning model, and performance of the image analysis machine learning model; and
computing the score as an aggregation of the plurality of sub-scores.

### Claim 14 (depends on 10)

The computer implemented method of claim 10, further comprising re-computing the score of the predicted series in response to at least one of: the predicted series being available, or after a predefined amount of time has elapsed.

### Claim 15 (depends on 1)

The computer implemented method of claim 1, further comprising in response to a certain series which was previously unavailable for processing becoming available for processing, analyzing the certain series to determine actual value of at least one parameter, and correcting the predicted at least one parameter to the actual value.

### Claim 16 (depends on 1)

The computer implemented method of claim 1, wherein the estimated at least one parameter of the at least one series which is not yet available for processing and/or a prediction of the time when at least one series of the study which is not yet available for processing will be available for processing, is obtained as an outcome of a prediction ML model that is fed at least one parameter of the study, and at least one parameter and time when each at least one series became available for processing.

### Claim 17 (depends on 16)

The computer implemented method of claim 16, wherein the prediction ML model comprises:
(i) a series encoder that receives the at least one parameter of the study as input, and generates an outcome of an encoded vector representing the series;
(ii) a time stamped encoder that receives the encoded vector as input and the time when each at least one series became available for processing, and generates an encoded study;
(iii) a decoder that receives the encoded study as input, and generates an outcome of a prediction of a time when at least one series will become available for processing and/or a prediction of parameter(s) of the at least one series which is not yet available for processing, until an end of the study.

### Claim 18 (depends on 16)

The computer implemented method of claim 16, wherein the prediction ML model is training on a training dataset of a plurality of records, wherein a record includes at least one parameter of a sample study, at least one parameter of a sample series of the sample study, and a time when the sample series became available for processing.

### Claim 19 (depends on 1)

The computer implemented method of claim 1, further comprising:
in response to a certain series which was not yet available for processing, being available for processing, repeating the predicting, the predicting at least one parameter, the obtaining at least one parameter, the selecting, and the feeding.

### Claim 20 (depends on 1)

The computer implemented method of claim 1, wherein selecting comprises feeding the combination into a decision machine learning model, and obtaining the target series as an outcome of the decision machine learning model, wherein the machine learning model is trained on a training dataset of a plurality of records, wherein a record is for a sample study, and includes time when a certain sample series became available for processing, at least one parameter of the certain sample series, and a ground truth indication of the target sample series.

### Claim 21 (depends on 1)

The computer implemented method of claim 1, further comprising:
performing a relevancy check on each series of the study, wherein the relevancy check is performed using the predicted at least one parameter for the series which is not yet available for processing, and using the obtained at least one parameter for the series which is available and/or by analyzing the series which is available; and
excluding a certain series from the selecting and the feeding according to the relevancy check.

### Claim 22 (depends on 1)

The computer implemented method of claim 1, further comprising predicting an end-of-study (EOS), wherein the target series is selected in view of the predicted EOS.

### Claim 23

A system for scheduling analysis of at least one series of a study of medical images of a subject, comprising:
at least one processor executing a code for:
predicting a time when at least one series of the study which is not yet available for processing, will be available for processing;
predicting at least one parameter of the at least one series which is not yet available for processing, and obtaining the at least one parameter for series which are available;
selecting a target series according to a combination of the predicted time and the at least one parameter;
in response to the target series not yet available for processing, waiting for the target series to become available for processing; and
in response to the target series being available for processing, feeding the target series into an image analysis machine learning model.

### Claim 24

A computer implemented method of scheduling analysis of at least one series of a study of medical images of a subject, comprising:
in response to an end-of-study (EOS) of a plurality of series, for each respective series of the study, obtaining at least one parameter of metadata and/or properties of medical images of the respective series, and predicting a performance of an image analysis machine learning model being fed the respective series;
selecting a target series from the plurality of series of the study according to a combination of the at least one parameter and the predicted performance computed for each respective series; and
feeding the target series into the image analysis machine learning model.

### Claim 25 (depends on 24)

The computer implemented method of claim 24, further comprising predicting resource utilization of a computer implementing inference by the image analysis machine learning model for each respective series, wherein the combination for selecting the target series further includes the predicted resource utilization of the respective series.
