# US 2020/0390342 A1 · Hemodynamic analysis of vessels using recurrent neural network

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

- Number: US 2020/0390342 A1
- Document ID: US20200390342A1
- Status: Published application
- Kind: A1
- Application number: 16/891,192
- Filed: 2020-06-03
- Published: 2020-12-17
- Earliest claimed: 2019-06-11
- Assignees: Siemens Healthcare GmbH
- Applicants: Siemens Healthcare GmbH
- Inventors: Arnaud Arindra ADIYOSO
- Web page: https://patentgrep.com/patent/US20200390342A1

## CPC

- A61B 5/02028: {Determining haemodynamic parameters not otherwise provided for, e.g. cardiac contractility or left ventricular ejection fraction}
- G16H 50/20: for computer-aided diagnosis, e.g. based on medical expert systems
- A61B 5/02007: {Evaluating blood vessel condition, e.g. elasticity, compliance}
- A61B 5/029: Measuring blood output from the heart, e.g. minute volume
- G16H 50/30: for calculating health indices; for individual health risk assessment
- G16H 50/50: for simulation or modelling of medical disorders
- A61B: DIAGNOSIS; SURGERY; IDENTIFICATION (usefulness limited to only animals A61D)
- G16H: HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA

## Abstract

A method of and an Artificial Intelligence (AI) system for predicting hemodynamic parameters for a target vessel, in particular of an aorta, as well as to a computer-implemented method of training an AI unit of the AI system are disclosed. A vessel shape model of the target vessel and a corresponding flow profile of the target vessel are received. At least one hemodynamic parameter is predicted by the AI unit based on the received vessel shape model and the received flow profile. The AI unit is arranged and configured to predict at least one hemodynamic parameter based on a received vessel shape model and a received flow profile of the target vessel (aorta).

## Claims

### Claim 1

A method of predicting hemodynamic parameters for a target vessel, comprising:
receiving a vessel shape model of the target vessel;
receiving a flow profile corresponding to the target vessel received; and
predicting at least one hemodynamic parameter based on the vessel shape model received and the flow profile received, via an Artificial Intelligence (AI) unit.

### Claim 2 (depends on 1)

The method of claim 1, wherein in the predicting of the at least one hemodynamic parameter includes predicting at least one vessel shape point of the vessel shape model received.

### Claim 3 (depends on 2)

The method of claim 2, wherein the predicting of the at least one hemodynamic parameter includes iteratively executing the predicting for each of at least two consecutive vessel shape points of the vessel shape model received.

### Claim 4 (depends on 3)

The method of claim 3, wherein the predicting includes:
providing, to a first input block of the AI unit, one vessel shape point of the vessel shape model received at each iteration;
providing, to a second input block of the AI unit, the flow profile received at a first iteration;
generating at least one current hidden state, in a deduction block of the AI unit, based on the one vessel shape point provided and on the flow profile provided and received at the first iteration, or at least one previous hidden state of the deduction block at any iteration subsequent to the first iteration; and
determining the at least one hemodynamic parameter in an output block of the AI unit, based at least on a relatively last generated current hidden state of the at least one generated current hidden state.

### Claim 5 (depends on 4)

The method of claim 4, wherein the generating comprises:
generating a first current hidden state in a first deduction layer of the AI unit, based on the one vessel shape point provided and based on the flow profile provided at the first iteration or a provided first previous hidden state of the first deduction layer at any iteration subsequent to the first iteration; and
generating the relatively last current hidden state in a relatively last deduction layer of the AI unit, based on the current hidden state of the previous deduction layer and on either an initial state of the relatively last deduction layer or a provided relatively last previous hidden state of the relatively last deduction layer.

### Claim 6 (depends on 1)

The method of claim 1, wherein the vessel shape model includes a centreline along a main direction of the target vessel and wherein each consecutive vessel shape point includes a consecutive centreline point on a centreline and a radius of the target vessel at a respective centreline point.

### Claim 7 (depends on 6)

The method of claim 6, wherein the vessel shape model includes a centreline along a main direction of the target vessel and wherein each consecutive vessel shape point includes one consecutive centreline point on the centreline and a number of surrounding points surrounding each centreline point in a form of a sphere with the respective centreline point in the centre of the respective sphere,
wherein the generating includes:
generating first current hidden states in a first deduction layer of the AI unit, based on the provided one vessel shape point including the respective centreline point and the surrounding points surrounding the centreline point and on the provided flow profile at the first iteration or provided first previous hidden states of the first deduction layer at any iteration subsequent to the first iteration; and
generating the relatively last current hidden states in a relatively last deduction layer of the AI unit, based on the current hidden states of the previous deduction layer and on either initial states of the relatively last deduction layer or provided last previous hidden states of the relatively last deduction layer, and
wherein, in the determining, the at least one hemodynamic parameter is determined in the output block of the AI unit based on the relatively last generated current hidden states.

### Claim 8 (depends on 1)

The method of claim 1, wherein the flow profile includes:
an inlet blood flow velocity u in a direction x of a corresponding image dataset;
an inlet blood flow velocity v in a direction y of the image dataset orthogonal to the direction x; and
an inlet blood flow velocity w in a direction z of the image dataset orthogonal to the directions x and y.

### Claim 9 (depends on 1)

The method of claim 1, wherein at least one of the vessel shape model and the flow profile is derived from an image dataset.

### Claim 10

An artificial intelligence (AI) system for predicting hemodynamic parameters for a target vessel, comprising:
a first interface arranged and configured to receive a vessel shape model of the target vessel;
a second interface arranged and configured to receive a corresponding flow profile of the target vessel; and
an AI unit communicatively connected to the first interface and communicatively connected to the second interface, arranged and configured to predict at least one hemodynamic parameter based on the vessel shape model received and the flow profile received.

### Claim 11 (depends on 10)

The AI system of claim 10, wherein the AI unit comprises a Recurrent Neural Network (RNN).

### Claim 12 (depends on 10)

The AI system of claim 10, wherein the AI unit comprises:
a first input block communicatively connected to the first interface and arranged and configured to get provided one vessel shape point, of the vessel shape model received at each iteration;
a second input block communicatively connected to the second interface and arranged and configured to get provided the flow profile received at a first iteration;
a deduction block communicatively connected to the first input block and communicatively connected to the second input block and arranged and configured to generate at least one current hidden state based on the one vessel shape point provided and based on the flow profile provided at the first iteration or at least one previous hidden state of the deduction block provided at any iteration subsequent to the first iteration; and
an output block communicatively connected to the deduction block and arranged and configured to determine the at least one hemodynamic parameter based at least on a relatively last generated current hidden state of the at least one generated current hidden state.

### Claim 13 (depends on 12)

The AI system of claim 12, wherein the AI unit further comprises:
a first deduction layer communicatively connected to the first input block and communicatively connected to the second input block and arranged and configured to generate a first current hidden state based on the one vessel shape point provided and based on the flow profile provided at the first iteration or a provided first previous hidden state of the first deduction layer at any iteration subsequent to the first iteration;
at least one further deduction layer communicatively connected to the first deduction layer and arranged and configured to generate at least one further current hidden state based on the current hidden state of the previous deduction layer and based on an initial state of a memory cell of the at least one further deduction layer at the first iteration or a provided further previous hidden state of the at least one further deduction layer at any iteration subsequent to the first iteration; and
a last deduction layer communicatively connected to either the first deduction layer or a relatively last of the at least one further deduction layer and arranged and configured to generate the last current hidden state based on the current hidden state of the previous deduction layer and on either an initial state of the memory cell of the relatively last deduction layer at the first iteration or a provided relatively last previous hidden state of the last deduction layer at any iteration subsequent to the first iteration.

### Claim 14 (depends on 13)

The AI system of claim 13, wherein the first input block is arranged and configured to get one vessel shape point provided including one consecutive centreline point on a centreline along a main direction of the target vessel and a number of surrounding points surrounding each centreline point in a form of a sphere with each respective centreline point in the centre of each respective sphere at each respective iteration,
wherein the first deduction layer is arranged and configured to generate first current hidden states based on the provided one vessel shape point including the respective centreline point and the surrounding points surrounding the centreline point and on the flow profile provided at the first iteration or provided first previous hidden states of the first deduction layer at any iteration subsequent to the first iteration,
wherein the at least one further deduction layer is arranged and configured to generate at least one set of further current hidden states based on the current hidden states of the previous deduction layer and based on initial states of memory cells of the at least one further deduction layer at the first iteration or provided further previous hidden states of the at least one further deduction layer at any iteration subsequent to the first iteration,
wherein the last deduction layer is arranged and configured to generate the last current hidden states based on the current hidden states of the previous deduction layer and on either initial states of memory cells of the last deduction layer at the first iteration or provided relatively last previous hidden states of the relatively last deduction layer at any iteration subsequent to the first iteration, and
wherein the output block is arranged and configured to determine the at least one hemodynamic parameter based at least on the relatively last generated current hidden states.

### Claim 15 (depends on 12)

The AI system of claim 12, wherein the output block comprises:
at least one concatenation layer communicatively connected to the deduction block; and
an output layer communicatively connected to the last of the at least one concatenation layer and communicatively connected to the deduction block,
wherein the at least one concatenation layer and the output layer are jointly arranged and configured to determine the at least one hemodynamic parameter based at least on the at least one generated current hidden state.

### Claim 16 (depends on 1)

A non-transitory computer-readable medium storing a program which, when the program is executed by a computer, cause the computer to carry out the method of claim 1.

### Claim 17

A computer-implemented method of training an Artificial Intelligence (AI) unit, comprising:
receiving an input training dataset including training vessel shape models and corresponding training flow profiles;
receiving an output training dataset corresponding to the training input dataset received including corresponding at least one training hemodynamic parameters; and
training the AI unit to predict at least one hemodynamic parameter of a target vessel with the training vessel shape models and corresponding training flow profiles received and the set of corresponding training outputs received.

### Claim 18 (depends on 17)

The computer-implemented method of claim 17, comprising initial steps of:
determining a shape variability and a geometric variability of a vessel within a given population by applying a statistical shape analysis on a set of real vessel shape models of the vessel;
synthesising at least one synthetic vessel shape model of the vessel based on the shape variability and geometric variability of the vessel within the given population determined;
determining corresponding training flow profiles for at least one of the at least one synthetic vessel shape model and the set of real vessel shape models;
aggregating the input training dataset with the set of real shape models, the at least one synthetic shape model and the corresponding training flow profiles determined; and
computing corresponding at least one training hemodynamic parameters of the output training set based on at least one of the set of real shape models and the at least one synthetic vessel shape model, and based on the corresponding flow profiles of the input training dataset.

### Claim 19 (depends on 17)

The computer-implemented method of claim 17, wherein the training hemodynamic parameters of the output training set are computed using Computational Fluid Dynamic (CFD) simulations.

### Claim 20 (depends on 17)

A non-transitory computer-readable medium storing a program which, when the program is executed by a computer, cause the computer to carry out the computer-implemented method of claim 17.
