# US 2009/0268952 A1 · Method and apparatus for automatically developing a high performance classifier for producing medically meaningful descriptors in medical diagnosis imaging

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

- Number: US 2009/0268952 A1
- Document ID: US20090268952A1
- Status: Published application
- Kind: A1
- Application number: 11/721,999
- Filed: 2005-12-13
- Published: 2009-10-29
- Earliest claimed: 2004-12-17
- Assignees: Koninklijke Philips Electronics N.V.
- Inventors: James David Schaffer; Walid Ali; Larry Eshelman; Claude Cohen-Bacrie; Jean-Michel Lagrange; Claire Levrier; Nicolas Francois Villain; Robert Randall Entrekin
- Web page: https://patentgrep.com/patent/US20090268952A1

## CPC

- G06F 18/21: Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

A method for determining the presence or absence of malignant features in medical images, wherein a plurality of base comparison or training images of various types of lesions taken of actual patient is examined by one or more image reading experts to create a first database array. Low-level features of each of the lesions in the same plurality of base comparisons or training images arc determined using one or more image processing algorithms to obtain a second database array set. The first and second database array set are combined to create a training database array set which is input to a learning system that discovers/learns a classifier that maps from a subset of the low-level features to the expert's evaluation in the first database array set. The classifier is used to determine the presence of a particular mid-level feature in an image of lesion in a patient based solely on the image.

## Claims

### Claim 1

A method for determining the presence of at least one medically meaningful feature in an object in an input medical image, comprising the steps of:
creating a first database comprising evaluations by at least one expert of at least one object in a plurality of training medical images;
analyzing by a data processor the at least one object in the training images to identify features therein for creating a second database comprising the identified features;
inputting the first and second database into a classifier data processor;
training the classifier to emulate a predefined perception of at least one medically meaningful feature based on the first and second database; and
analyzing the medical image in the data processor and the trained classifier to output at least one indication of a presence of the at least one medically meaningful feature of an object in the medical image.

### Claim 2 (depends on 1)

The method of claim 1, wherein the at least one object is a lesion.

### Claim 3 (depends on 1)

The method of claim 1, wherein the classifier is a neural network.

### Claim 4 (depends on 1)

The method of claim 1, wherein each evaluation comprises a classification of features of each object within the medical image.

### Claim 5 (depends on 1)

The method of claim 1, wherein the identified features are a low-level features of the object.

### Claim 6 (depends on 1)

The method of claim 1, wherein the low-level features are represented by at least one of Fourier descriptors, moments, shape, and stochastic, co-occurrence and autocorrelation parameters.

### Claim 7 (depends on 1)

The method of claim 1, wherein the at least one medically meaningful feature is a mid-level feature of the object.

### Claim 8 (depends on 7)

The method of claim the method of claim 7, wherein the mid-level feature is at least one malignant characteristic of the object.

### Claim 9 (depends on 8)

The method of claim 8, wherein the malignant characteristic is at least one of spiculation of the lesion, angular margins of the lesion, microlobulation of the lesion, aspect ratio of the lesion, calcifications in the lesion, duct extension and branching patterns in the lesion, calcifications, hypoechoic relative to fat, and heterogeneous texture of the lesion.

### Claim 10 (depends on 7)

The method of claim 7, wherein the mid-level feature is at least one benign characteristic of the lesion.

### Claim 11 (depends on 10)

The method of claim 10, wherein the benign characteristic is at least one of an extent to which the lesion is wider than taller, lobulation of the lesion, continuousness of the lesion, smooth capsule nature of the lesion, and homogeneous texture of the lesion.

### Claim 12 (depends on 4)

The method of claim 4, wherein the plurality of training medical images encompasses a full spectrum of expected malignancies.

### Claim 13 (depends on 1)

The method of claim 1, wherein the identified features are developed by using a genetic algorithm to select a gene subset of features based on the identified features in the training medical images.

### Claim 14 (depends on 12)

The method of claim 12, wherein the genetic algorithm is an adaptive search algorithm.

### Claim 15 (depends on 12)

The method of claim 12, wherein the identified features are low-level features which correspond to numbers in the gene subset.

### Claim 16 (depends on 15)

The method of claim 15, wherein the gene subset belongs to a set of genes of a chromosome.

### Claim 17 (depends on 1)

The method of claim 1, wherein said training step includes the step of generating a high performance classifier for mapping the identified features of the object to the at least one medically meaningful feature.

### Claim 18 (depends on 16)

The method of claim 16, wherein said generating step comprises the steps of:
providing a set of input-output pairs to the classifier,
dividing the input-output pairs into a training set and an evaluation set;
adapting the classifier with the training set, and
evaluating the adapted classifier with the training set.

### Claim 19 (depends on 1)

The method of claim 1, wherein the classifier is a Naïve Bayes classifier.

### Claim 20 (depends on 7)

The method of claim 7, wherein an individual classifier is used to generate each mid-level feature.

### Claim 21 (depends on 7)

The method of claim 7, wherein an individual classifier is used to generate each mid-level feature.

### Claim 22 (depends on 10)

The method of claim 10, wherein an individual classifier is used to generate each mid-level feature.

### Claim 23 (depends on 4)

The method of claim 4, wherein the object is a lesion.

### Claim 24

A method for determining the presence of at least one medically meaningful feature within an object in an image, comprising:
creating a first database having evaluations by at least one expert of at least one object in at least one training image;
analyzing the at least one object in the at least one training image to identify features therein for creating a second database of identified features;
inputting the first and second database array sets into a classifier;
training the classifier to emulate a predefined perception of at least one medically meaningful feature based on the first and second database; and
analyzing a medical image with the trained classifier to output at least one indication of a presence of the at least one medically meaningful feature of an object in the medical image.

### Claim 25

An apparatus for determining the presence of at least one medically meaningful feature in an object in an input medical image, comprising:
a first database for storing evaluations by at least one expert of at least one object in a plurality of training medical image;
a second data base for storing identified features of the medical image;
a data processor operatively coupled to the second and first databases, said processor being further operable to perform the steps of:
providing evaluations by at least one expert of at least one object in a plurality of training medical images for storage in the first database;
analyzing by the data processor the at least one object in the training images to identify features therein for storage in the second database;
inputting the first and second database into a classifier data processor;
training the classifier to emulate a predefined perception of at least one medically meaningful feature based on the first and second database; and
analyzing the medical image in the data processor and the trained classifier to output at least one indication of a presence of the at least one medically meaningful feature of an object in the medical image

### Claim 26 (depends on 25)

The apparatus of claim 25, wherein the classifier is a neural network or a Naïve Bayes classifier.
