Assessing features for classification

• US 8,693,788 B2
• Filed: 08/06/2010
• Issued: 04/08/2014
• Est. Priority Date: 08/06/2010
• Status: Expired due to Fees

• Current Assignee: STRATA Skin Sciences, Inc.
• Original Assignee: Mela Sciences, Inc.
• Inventors: Wojton, Maciej, Kabelev, Nikolai
• Primary Examiner(s): Bella, Matthew
• Assistant Examiner(s): YANG, WEI WEN
• Application Number:
  US12/852,195
• Publication Number:
  US 20120033863A1
• Time in Patent Office: 1,341 Days
• Field of Search: 382126-133, 382/224, 382226-230, 382/205, 382/209
• US Class Current: 382/209
• CPC Class Codes: A61B 2576/02 specially adapted for a particular organ or body part
• CPC Class Codes: A61B 5/0013 Medical image data A61B1/00011, A61B6/56, A61B8/56 take precedence
• CPC Class Codes: A61B 5/0077 Devices for viewing the surface of the body, e.g. camera, magnifying lens
• CPC Class Codes: A61B 5/441 Skin evaluation, e.g. for skin disorder diagnosis
• CPC Class Codes: G06F 18/2113 by ranking or filtering the set of features, e.g. using a measure of variance or of feature cross-correlation
• CPC Class Codes: G06V 10/758 Involving statistics of pixels or of feature values, e.g. histogram matching
• CPC Class Codes: G06V 10/771 Feature selection, e.g. selecting representative features from a multi-dimensional feature space
• CPC Class Codes: G16H 30/40 for processing medical images, e.g. editing
• CPC Class Codes: G16H 50/20 for computer-aided diagnosis, e.g. based on medical expert systems

Abstract

Among other things, a degree is determined to which values of a feature that characterizes cases correctly predict classes to which cases belong, the degree depending on a combination of at least two factors that include: (a) uncertainty about the values that the feature will have depending on knowledge about classes to which cases belong, and (b) comparison of one or more first values of the feature that are associated with cases that belong to a first class with one or more second values of the feature associated with cases belonging to a second class.

Claims

1. A computer-implemented method comprising:

determining, by a processor, a degree to which values of a feature that characterizes cases correctly predict classes to which the cases belong, the degree depending on a mathematical combination of at least two factors that include;

(a) uncertainty about the values that the feature will have depending on knowledge about classes to which the cases belong, wherein the uncertainty comprises an entropy of the values of the feature, and

(b) comparison of one or more first values of the feature that are associated with cases that belong to a first class with one or more second values of the feature associated with cases belonging to a second class.

2. The method of claim 1 in which the uncertainty is represented by a mutual information measure.

3. The method of claim 1 in which the comparison comprises a distance measure.

4. The method of claim 3 in which the distance measure is normalized.

5. The method of claim 1 in which the uncertainty about the values of the feature comprises an uncertainty when the classes to which the cases belong are known, and an uncertainty when the classes to which the cases belong are not known.

6. The method of claim 1, in which the degree is determined as a score.

7. The method of claim 1, in which the comparison of one or more first values of the feature with one or more second values of the feature comprises determining differences between first values and second values.

8. The method of claim 1, in which the comparison of one or more first values of the feature with one or more second values of the feature comprises a comparison of a first value of the feature with two or more second values of the feature.

9. The method of claim 1, in which the degree is determined with respect to values of more than one such feature, the degree depending on:

(a) uncertainty about values that the more than one such feature will have depending on the classes to which cases belong, and

(b) comparison of one or more first values of the more than one such feature that are associated with cases that belong to a first class with one or more second values of the more than one such feature associated with cases belonging to a second class.

10. The method of claim 3, in which the comparison comprises a metric distance between the one or more first values of the feature and the one or more second values of the feature.

11. The method of claim 1, in which the cases comprise medical cases.

12. The method of claim 1 in which the cases comprise communication cases.

13. The method of claim 1 in which the cases comprise computer-vision cases.

14. The method of claim 1 in which the cases comprise computer-security cases.

15. The method of claim 1 in which the cases comprise voice-recognition cases.

16. The method of claim 1 in which the cases comprise genetic cases.

17. The method of claim 10, in which the cases comprise skin lesions and the classes include melanoma and non-melanoma.

18. The method of claim 1, further comprising:

determining a first degree to which a value of a first feature that characterizes cases correctly predicts classes to which a particular case belongs;

determining a second degree to which a value of a second feature that characterizes cases correctly predicts classes to which a particular case belongs; and

comparing the determined first and second degrees.

19. The method of claim 1, further including:

selecting a feature for use in classification, based at least in part on the determination of the degree.

20. The method of claim 1, in which the cases are represented by digital images and the features relate to what is portrayed in the digital images.

21. The method of claim 20, in which the values comprise measures of features that appear in the digital images.

22. The method of claim 1 in which the uncertainty is represented by an accuracy measure.

23. The method of claim 1 in which the degree is determined based on a single calculation that uses at least the factors (a) and (b).