Probability density function estimation

• US 9,076,197 B2
• Filed: 04/29/2011
• Issued: 07/07/2015
• Est. Priority Date: 04/30/2010
• Status: Active Grant

• Current Assignee: iCAD Incorporated
• Original Assignee: VuCOMP, Inc.
• Inventors: Wehnes, Jeffrey C.
• Primary Examiner(s): FUJITA, KATRINA R
• Application Number:
  US13/695,351
• Publication Number:
  US 20130208967A1
• Time in Patent Office: 1,530 Days
• US Class Current: 1/1
• CPC Class Codes: G06F 18/24 Classification techniques
• CPC Class Codes: G06T 2207/10116 X-ray image
• CPC Class Codes: G06T 2207/20016 Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
• CPC Class Codes: G06T 2207/30068 Mammography; Breast
• CPC Class Codes: G06T 2207/30096 Tumor; Lesion
• CPC Class Codes: G06T 5/10 using non-spatial domain filtering
• CPC Class Codes: G06T 5/70 Denoising; Smoothing
• CPC Class Codes: G06T 7/0012 Biomedical image inspection
• CPC Class Codes: G06T 7/11 Region-based segmentation
• CPC Class Codes: G06T 7/143 involving probabilistic approaches, e.g. Markov random field [MRF] modelling
• CPC Class Codes: G06V 10/454 Integrating the filters into a hierarchical structure, e.g. convolutional neural networks [CNN]

Abstract

A PDF estimator for determining a probability that a detected object is a specific type of object is provided. Training data from a known set is used to functionally describe the relevant neighborhood for specific representation points. The neighborhood is selected based on the measured features of the object to be classified and weights are calculated to be applied to the representation points. A probability is determined based upon the stored training data, the measured features of the object to be classified, and the weights.

Claims

1. A method for classifying an anomaly in a digital image, the method comprising:

receiving training data comprising a training feature value for each of a plurality of anomaly classification features for each of a plurality of training cases;

defining a neighborhood size for each of a plurality of representation points in feature space for each anomaly classification feature based on the training data;

receiving measured data comprising a measured feature value at an evaluation point in feature space for each anomaly classification feature for a measured case;

determining a scale parameter vector for at least some of the representation points near the evaluation point for each anomaly classification feature to define a respective neighborhood size for that anomaly classification feature;

determining a weight factor for the at least some of the representation points using the respective scale parameter vector; and

applying the weight factor for the at least some of the representation points to the training data at the plurality of representation points to generate a classification probability for the anomaly for the measured case at the evaluation point, wherein the scale parameter vector for a respective anomaly classification feature indicates the respective neighborhood size used to generate the classification probability at the evaluation point in feature space.

2. The method of claim 1, wherein receiving the training data comprises retrieving the training data from a database.

3. The method of claim 1, further comprising storing probability density function (PDF) estimates for the classification probability for the measured case in a database.

4. The method of claim 1, further comprising performing the receiving measured data, determining the scale parameter vector, determining and applying the weight factor, and generating the classification probability for a plurality of measured cases.

5. The method of claim 1, wherein the at least some of the representation points each represent multiple samples of the training data.

6. The method of claim 1, wherein receiving the training data and defining the neighborhood size are performed offline, and the neighborhood size for each of the plurality of representation points is stored in a database for later use in determining the scale parameter vector.

7. A system for classifying anomalies in a digital image, the system comprising:

a neighborhood definition unit receiving training data comprising a training feature value for each of a plurality of anomaly classification features for each of a plurality of training cases, and defining a neighborhood size for each of a plurality of representation points in feature space based on the training data;

a neighborhood determination unit receiving measured data comprising a measured feature value at an evaluation point in feature space for each anomaly classification feature for a measured case, and determining a scale parameter vector for at least some of the representation points near the evaluation point for each anomaly classification feature to define a respective neighborhood size for that anomaly classification feature;

a weight determination unit determining a weight factor for the at least some of the representation points using the respective scale parameter vector; and

a local estimator applying the weight factor for the at least some of the representation points to the training data at the plurality of representation points to generate a probability density function (PDF) estimate for an anomaly at the evaluation point, wherein the scale parameter vector for a respective anomaly classification feature indicates a respective neighborhood size used to generate the PDF estimate at the evaluation point in feature space.

8. The system of claim 7, wherein the neighborhood definition unit retrieves the training data from a database.

9. The system of claim 7, wherein the local estimator stores the PDF estimate for the measured case in a database.

10. The system of claim 7, wherein the at least some of the representation points each represent multiple samples of the training data.

11. The system of claim 7, wherein the neighborhood definition unit stores the neighborhood size for each of the plurality of representation points in a database for later use by the neighborhood determination unit.

12. A computer program product for classifying anomalies, the computer program product having a non-transitory computer-readable medium with a computer program embodied thereon, the computer program comprising:

computer program code for receiving training data comprising a training feature value for each of a plurality of anomaly classification features for each of a plurality of training cases;

computer program code for defining a neighborhood size for each of a plurality of representation points in feature space for each anomaly classification feature based on the training data;

computer program code for receiving measured data comprising a measured feature value at an evaluation point in feature space for a measured case;

computer program code for determining a scale parameter vector for at least some of the representation points near the evaluation point for each anomaly classification feature to define a respective neighborhood size for that anomaly classification feature;

computer program code for determining a weight factor for the at least some of the representation points using the respective scale parameter vector; and

computer program code for applying the weight factor for the at least some of the representation points to the training data at the plurality of representation points to generate a probability density function (PDF) estimate for an anomaly for the measured case at the evaluation point, wherein the scale parameter vector for a respective anomaly classification feature indicates a respective neighborhood size used to generate the PDF estimate at the evaluation point in feature space.

13. The computer program product of claim 12, wherein the computer program code for receiving the training data comprises computer program code for retrieving the training data from a database.

14. The computer program product of claim 12, further comprising computer program code for storing the PDF estimate for the measured case in a database.

15. The computer program product of claim 12, further comprising computer program code for performing the receiving measured data, determining the scale parameter vector, determining and applying the weight factor, and generating the PDF estimate for a plurality of measured cases.

16. The computer program product of claim 12, wherein the at least some of the representation points each represent multiple samples of the training data.