# US 2021/0174957 A1 · Characterizing and identifying biological structure

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

- Number: US 2021/0174957 A1
- Document ID: US20210174957A1
- Status: Published application
- Kind: A1
- Application number: 16/462,870
- Filed: 2017-11-21
- Published: 2021-06-10
- Earliest claimed: 2016-11-21
- Assignees: Sensome SAS
- Applicants: Sensome SAS
- Inventors: Gor Lebedev; Pierluca Messina
- Web page: https://patentgrep.com/patent/US20210174957A1

## CPC

- 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 5/0538: invasively, e.g. using a catheter
- A61B 5/7225: {Details of analogue processing, e.g. isolation amplifier, gain or sensitivity adjustment, filtering, baseline or drift compensation (input circuits for detecting, measuring, or recording bioelectric or biomagnetic signals A61B5/30; specific diagnostic methods using bioelectric or biomagnetic signals A61B5/316)}
- A61B 5/7267: {involving training the classification device}
- G16H 20/40: relating to mechanical, radiation or invasive therapies, e.g. surgery, laser therapy, dialysis or acupuncture
- G16H 40/63: for local operation
- A61B 2562/02: Details of sensors specially adapted for in-vivo measurements
- G16H 50/50: for simulation or modelling of medical disorders
- G16H: HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- A61B: DIAGNOSIS; SURGERY; IDENTIFICATION (usefulness limited to only animals A61D)

## Abstract

Embodiments described relate to techniques for identifying and characterizing biological structures using machine learning techniques. These techniques may be employed to enable a device to identify the particular type of tissue and/or cells (e.g., platelets, smooth muscle cells, or endothelial cells) in, for example, a biological structure, which may be a tissue or a lesion of a duct (e.g., vasculature) in an animal (e.g., a human or non-human animal), among other structures. The machine learning techniques may use raw impedance spectroscopy measurement data in addition to values derived from that raw data. In addition, the machine learning techniques may be used to select frequencies at which to measure impedance and select features to extract from the measured impedance at the selected frequencies to arrive at a small set of frequencies that allow for reliable differentiation.

## Claims

### Claim 1

A method of training a system, the method comprising:
receiving training data including a plurality of sets of impedance measurements at least for a type of biological structure, each set of the plurality including impedance measurements for multiple different frequencies;
selecting, based at least in part on the plurality of sets of impedance measurements, a set of frequencies with which to configure a medical device for in vivo measurement of biological structures of the type, the set of frequencies being less than all of the multiple different frequencies for which impedance measurements are included in the plurality of sets of impedance measurements, wherein the selecting comprises:
identifying a first plurality of subsets of the training data including a first subset of impedance measurements from each set of the plurality of sets of impedance measurements, each first subset of the first plurality of subsets including impedance measurements for a first subset of frequencies, the first subset of frequencies including less than all of the multiple different frequencies for which impedance measurements are included in the plurality of sets of impedance measurements;
identifying a first plurality of features from each first subset of the first plurality of subsets of the training data, the first plurality of features including for each first subset at least one derived feature that is not present in the first subset of impedance measurements and is derived from the first subset of the training data; and
training a model using at least one machine learning technique with the first plurality of identified features to create a first trained model.

### Claim at position 2

-3. (canceled)

### Claim 4 (depends on 1)

The method of claim 1, wherein training the model comprises training the model to identify at least one characteristic of a biological structure the at least one characteristic of the biological structure includes a composition of the biological structure.

### Claim 5 (depends on 1)

The method of claim 1, further comprising:
configuring the medical device to obtain in vivo measurements of impedance of biological structures at frequencies of the first subset of frequencies; and
using the first trained model to identify at least one characteristic of a biological structure.

### Claim 6 (depends on 5)

The method of claim 5, wherein using the first trained model to identify the at least one characteristic of the biological structure includes:
capturing at least one impedance measurement of the biological structure; and
identifying that least one characteristic of the biological structure based on the captured at least one impedance measurement of the biological structure using at least one machine learning parameter associated with the trained model.

### Claim 7 (depends on 1)

The method of claim 1, further comprising:
determining whether the first trained model meets at least one performance target; and
configuring the medical device to obtain in vivo measurements of impedance of biological structures of the type at frequencies of the first subset of frequencies and using the first trained model to identify at least one characteristic of a biological structure, responsive to the first trained model meeting the at least one performance target.

### Claim 8 (depends on 7)

The method of claim 7, further comprising. responsive to the first trained model failing to meet the at least one performance target:
identifying a second plurality of subsets of the training data including a second subset of impedance measurements from each set of the plurality of sets of impedance measurements, each second subset of the second plurality of subsets including impedance measurements for a second subset of frequencies, the second subset of frequencies including less than all of the multiple different frequencies for which impedance measurements are included in the plurality of sets of impedance measurements and being a different set of frequencies than the first subset;
identifying a second plurality of features from each second subset of the second plurality of subsets of the training data, the second plurality of features including at least one derived feature that is not present in the second subset of impedance measurements and is derived from the identified second subset of the training data;
training the model using the at least one machine learning technique with the second plurality of identified features to create a second trained model;
determining whether the second trained model meets the at least one performance target; and
using the second trained model to identify the at least one characteristic of the biological structure responsive to the second trained model meeting the at least one performance target, wherein using the second trained model comprises configuring the medical device to obtain in vivo measurements of impedance of biological structures of the type at frequencies of the second subset of frequencies.

### Claim 9 (depends on 8)

The method of claim 8, wherein identifying the second subset comprises selecting the second subset of frequencies based at least in part on a genetic algorithm that receives as input the first subset of frequencies and a performance metric for the first trained model.

### Claim at position 9

-27. (canceled)

### Claim 28 (depends on 1)

The method of claim 1, wherein:
the first plurality of features includes at least one feature present in each first subset of the first plurality of subsets of training data and the at least one derived feature for each first subset;
the at least one feature present in the subset comprises a value of an impedance measurement at a given frequency; and
the at least one derived feature for each first subset comprises a value determined from performing at least one computation on one or more impedance values of the first subset.

### Claim 29 (depends on 28)

The method of claim 28, wherein performing the at least one computation comprises performing one or more statistical analyses on the one or more impedance values of the first subset.

### Claim 30 (depends on 28)

The method of claim 28, wherein performing the at least one computation comprises determining a value indicative of a change between impedance measurements of the subset.

### Claim at position 13

-37. (canceled)

### Claim 38 (depends on 1)

The method of claim 1, wherein:
training the model comprises training at least one model to identify at least one characteristic of biological structures of the type;
the plurality of sets of impedance measurements include impedance measurements for a plurality of types of biological structures and each set of the plurality of sets indicating a type of biological structure to which the impedance measurements relate:
the training the at least one model comprises training a first model using the first plurality of identified features and, for each first subset of the plurality of subsets from which the first plurality of identified features were identified, an indication of a biological structure to which the first subset of impedance measurements relate;
the training comprises training the first model, based at least in part on the impedance measurements, to differentiate impedance measurements for the type of biological structure from impedance measurements for one or more other types of biological structures; and
the training the at least one model further comprises training a second model, at least in part by applying the at least one machine learning technique to impedance measurements for the first type of biological structure, to identify the at least one characteristic of the type of biological structure.

### Claim 39 (depends on 38)

The method of claim 38, further comprising operating the at least one processor to carry out acts of:
filtering an input set of impedance measurements of the plurality of sets using the first model to yield a filtered input set of impedance measurements relating to a first biological structure of the type; and
determining the at least one characteristic for the first biological structure using the second model.

### Claim at position 16

-48. (canceled)

### Claim 49

An apparatus comprising:
at least one medical device for obtaining in vivo impedance measurements of a biological structure, the at least one medical device comprising at least one impedance sensor to obtain impedance measurements of the biological structure in vivo using the at least one impedance sensor at a fixed set of frequencies; and
at least one control circuit configured to process the impedance measurements at least in part by evaluating the impedance measurements at the fixed scat of frequencies for the biological structure using at least one trained model that is trained to distinguish between biological structures having different characteristics based on the impedance measurements at the fixed set of frequencies, wherein using the at least one trained model comprises deriving at least one derived feature from the impedance measurements and processing the at least one derived feature using the at least one trained model.

### Claim 50 (depends on 49)

The apparatus of claim 49, wherein:
the at least one trained model is at least one first trained model;
the at least one control circuit is further figured to filter one or more sets of impedance measurements, using at least one second trained model, to produce filtered impedance measurements, the received impedance measurements including at least some impedance measurements corresponding to the biological structure, the at least one second trained model trained to distinguish impedance measurements corresponding to the biological structure from impedance measurements not corresponding to the biological structure; and
the at least one control circuit is configured to evaluate the impedance measurements using the at least one first trained model at least in part by evaluating the filtered impedance measurements using the at least one first trained model.

### Claim at position 19

-52. (canceled)

### Claim 53 (depends on 50)

The method of claim 50, wherein:
the at least one second trained model is further configured o distinguish between erroneous and non-erroneous impedance measurements; and
filtering the one or more sets of impedance measurements comprises, using the at east one second trained model, detecting and removing any erroneous measurements,

### Claim 54 (depends on 49)

The apparatus claim 49, wherein the at least one circuit is configured to evaluate the impedance measurements using the at least one trained model at least in part by evaluating the impedance measurements using the at least one trained model to determine the manner in which to treat the lesion.

### Claim 55 (depends on 54)

The apparatus of claim 54, wherein the at least one circuit is configured to evaluate the impedance measurements using the at least one trained model at least in part by performing one or more computations on the impedance measurements and on coefficients of the at least one trained model, wherein a result of the one or more computations indicates a manner in which the lesion is to be treated.

### Claim 56 (depends on 1)

The method of claim 1, further comprising:
configuring the medical device to obtain in vivo measurements of impedance of biological structures of the type at frequencies of the first subset of frequencies; and
using the first trained model to identify a manner in which to treat a biological structure of the type.

### Claim at position 24

-58. (canceled)

### Claim 60

At least one non-transitory computer-readable storage medium having encoded thereon executable instructions that, when executed by at least one processor, cause the at least one processor to carry out a method comprising:
receiving training data including a plurality of sets of impedance measurements at least for a type of biological structure, each set of the plurality including impedance measurements for multiple different frequencies;
selecting, based at least in part on the plurality of sets of impedance measurements, a set of frequencies with which to configure an invasive probe for in vivo measurement of biological structures of the type, the set of frequencies being less than all of the multiple different frequencies for which impedance measurements are included in the plurality of sets of impedance measurements, wherein the selecting comprises:
identifying a first plurality of subsets of the training data including a first subset of impedance measurements from each set of the plurality of sets of impedance measurements, each first subset of the first plurality of subsets including impedance measurements for a first subset of frequencies, the first subset of frequencies including less than all of the multiple different frequencies for which impedance measurements are included in the plurality of sets of impedance measurements
identifying a first plurality of features from each first subset of the first plurality of subsets of the training data, the first plurality of features including for each first subset at least one derived feature that is not present in the first subset of impedance measurements and is derived from the first subset of the training data; and
training a model using at least one machine learning technique with the first plurality of identified features to create a first trained model.

### Claim 61 (depends on 56)

The method of claim 56, wherein using the first trained model to identify a manner in which to treat the biological structure comprises identifying the manner in which to treat without identifying or characterizing the biological structure.

### Claim 62 (depends on 55)

The apparatus of claim 55, wherein the at least one control circuit is a component of the at least one medical device.

### Claim 63 (depends on 49)

The apparatus of claim 49, wherein the at least one control circuit is at least one processor.
