Neural network based automated cytological specimen classification system and method

• US 4,965,725 A
• Filed: 04/08/1988
• Issued: 10/23/1990
• Est. Priority Date: 04/08/1988
• Status: Expired due to Term

• Current Assignee: Tripath Imaging Incorporated (Becton, Dickinson & Co)
• Original Assignee: Nueromedical Systems Incorporated
• Inventors: Rutenberg, Mark R.
• Primary Examiner(s): Jablon, Clark A.
• Application Number:
  US07/179,060
• Time in Patent Office: 928 Days
• Field of Search: 382/15, 382/36, 364/513, 364/801, 364/807, 364/200 MS File, 364/900 MS File, 364/413.1, 307/201
• US Class Current: 382/224
• CPC Class Codes: G01N 15/1433 using image recognition
• CPC Class Codes: G01N 15/1468 with spatial resolution of the texture or inner structure of the particle
• CPC Class Codes: G01N 2015/1488 Methods for deciding
• CPC Class Codes: G06N 3/04 Architecture, e.g. interconnection topology
• CPC Class Codes: G06V 20/693 Acquisition
• CPC Class Codes: G06V 20/698 Matching; Classification
• 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
• CPC Class Codes: Y10S 128/925 Neural network
• CPC Class Codes: Y10S 706/924 Medical

Abstract

An automated screening system and method for cytological specimen classification in which a neural network is utilized in performance of the classification function. Also included is an automated microscope and associated image processing circuitry.

Claims

1. An automated cytological specimen classifier for classifying cells contained in a smear on a slide to identify cells that are likely to be malignant or pre-malignant, comprising:

(a) microscope means for obtaining a vie of at least part of a cytological specimen including cells and other material located generally randomly on a slide in an arrangement which can include other than a single layer of cells,

(b) camera means for creating an image of such view,

(c) image digitizing means for producing a digital representation of such image,

(d) a primary classifier means for detecting objects in a digital representation of a cytological specimen based on a detectable feature, said primary classifier means comprising a classifier means for detecting cells that are likely to be malignant or pre-malignant as well as other cells and material that initially appear to have characteristics of a malignant cell or a pre-malignant cell based on integrated optical density, and

(e) a secondary classifier for distinguishing pre-malignant and malignant cells form other cells and material among the objects detected by the primary classifier, said secondary classifier means comprising a neural computer apparatus means for effecting such distinguishing as a function of training thereof.

2. The automated classifier of claim 1 wherein said primary classifier means comprises a statistical classifier.

3. The automated classifier of claim 1 wherein such cytological specimen includes overlapping cells.

4. The automated classifier of claim 1 wherein said secondary classifier means is operable to distinguish pre-malignant and malignant cells among overlapping arrangements of cells and other material.

5. The automated classifier of claim 1, wherein said camera means is positioned to create an image of a portion of such cytological specimen from such view.

6. The automated classifier of claim 1, said neural computer apparatus means comprising an electronic neural computer.

7. The automated classifier of claim 1 wherein the primary classifier means is restricted to evaluation of the cellular nucleus while the secondary classifier means evaluates both the cellular nucleus and its surrounding cytoplasm.

8. The automated classifier of claim 1 wherein both the primary classifier means and the secondary classifier means are restricted to evaluation of the cellular nucleus.

9. The automated classifier of claim 1 further comprising means for making an additional non-neural net classification of nuclear morphological components in addition to integrated optical density, said means being coupled between the primary classifier means and the secondary classifier means.

10. The automated classifier of claim 1, said microscope means comprising an automated microscope.

11. The automated classifier of claim 1, said camera means comprising a video camera.

12. The automated classifier of claim 1, said camera means comprising a charge coupled device.

13. The automated classifier of claim 1, said primary classifier means comprising means for detection of objects in such digital representation of a cytological specimen which has a feature that exceeds a threshold level.

14. The automated classifier of claim 1, said primary classifier means comprising means for detection of objects in such digital representation of a cytological specimen which has a feature that exceeds a threshold integrated optical density.

15. The automated classifier of claim 1, said primary classifier means comprising means for detection of objects in such digital representation of a cytological specimen based on morphological criteria.

16. The automated classifier of claim 1 further comprising a neural network pre-screening classifier means for recognition and classification of general areas within the digital representation of a specimen that contain material other than a cellular monolayer prior to primary classification by said primary classifier means.

17. The automated classifier of claim 16 wherein the output form the pre-screening classifier means is utilized to exclude such areas from further analysis.

18. The automated classifier of claim 16 wherein the output form the pre-screening classifier means is utilized to modify further analysis of images found within such areas.

27. The automated classifier of claim 16, wherein the primary classifier means is restricted to evaluation of the cellular nucleus while the secondary classifier means evaluates both the cellular nucleus and its surrounding cytoplasm.

31. The automated classifier of claim 16, said camera means comprising a video camera.

38. The automated classifier of claim 16 wherein primary classifier means comprises a statistical classifier.

39. The automated classifier of claim 16 wherein such cytological specimen includes overlapping cells.

40. The automated classifier of claim 16 wherein said secondary classifier means is operable to distinguish cells of such predetermined cell type among overlapping arrangements of cells and other materials.

19. A method of classifying cytological specimens, comprising using a primary classifier apparatus primarily classifying a specimen which is generally randomly arranged and can include other than in a single layer to determine locations of interest, and secondarily classifying such locations of interest using a neural network computer apparatus.

20. The method of claim 19, wherein said primarily classifying step comprises using a video camera or charge coupled device (CCD) to obtain images of the specimen, a digitizer to digitize such images and an integrated optical density detector.

21. The method of claim 19, wherein said primary classifying comprises using a neural network computer apparatus.

22. The method of claim 19, wherein said step of using a primary classifier apparatus primarily classifying a specimen comprises using a statistical classifier.

23. The method of claim 19, wherein said step of using a primary classifier apparatus primarily classifying a specimen comprises making a classification based on morphology.

24. The method of claim 19, wherein said step of using a primary classifier apparatus primarily classifying a specimen comprises making such primary classification based on integrated optical density.

25. The method of claim 19, further comprising training such neural network computer apparatus to identify cytological specimens of interest.

26. An automated cytological specimen classifier, comprising:

(a) microscope means for obtaining a view of at least part of a cytological specimen including cells and other material located generally randomly in an arrangement which can include other than a single layer of cells,

(b) camera means for creating an image of such view,

(c) image digitizing means for producing a digital representation of such image,

(d) primary classifier means for detecting objects in a digital representation of a cytological specimen based on a detectable feature, said primary classifier means comprising a classifier for detecting cells that are likely to be of a predetermined cell type as well as other cells and material that initially appear to have characteristics of such predetermined cell type, and

(e) secondary classifier means for distinguishing cells of such predetermined cell type from other cells and material among the objects detected by said primary classifier means, said secondary classifier means comprising a neural computer apparatus means for effecting such distinguishing as a function of training thereof.

28. The automated classifier of claim 26 wherein both the primary classifier means and the secondary classifier means are restricted to evaluation of the cellular nucleus.

29. The automated classifier of claim 26, further comprising means for making an additional non-neural net classification of nuclear morphological components, said means being coupled between the primary classifier means and said secondary classifier means.

30. The automated classifier of claim 26, said microscope means comprising an automated microscope.

32. The automated classifier of claim 26, said camera means comprising a charge coupled device.

33. The automated classifier of claim 26, said primary classifier means comprising means for detection of objects in such digital representation of a cytological specimen which have a feature that exceeds a threshold level.

34. The automated classifier of claim 26, said primary classifier means comprising means for detection of objects in such digital representation of a cytological specimen which has a feature that exceeds a threshold integrated optical density.

35. The automated classifier of claim 26, said primary classifier means comprising means for detection of objects in such digital representation of a cytological specimen based on morphological criteria.

36. The automated classifier of claim 26, wherein said camera means is positioned to create an image of a portion of such cytological specimen from such view.

37. The automated classifier of claim 26, said neural computer apparatus means comprising an electronic neural computer.

41. The automated classifier of claim 26 further comprising a neural network pre-screening classifier means for identifying general areas within the digital representation of a specimen that contain material other than a cellular monolayer prior to primary classification.

42. The automated classifier of claim 41 wherein the output from the pre-screening classifier is utilized to exclude such identified areas from further analysis.

43. The automated classifier of claim 41 wherein the output from the pre-screening classifier means is utilized to modify further analysis of images found within the areas of the specimen identified by said pre-screening classifier means.