Apparatus and method for the detection and classification of articles using flow cytometry techniques

• US 4,661,913 A
• Filed: 09/11/1984
• Issued: 04/28/1987
• Est. Priority Date: 09/11/1984
• Status: Expired due to Term

• Current Assignee: Becton, Dickinson & Co
• Original Assignee: Becton, Dickinson & Co
• Inventors: Wu, Hai-Ping, Sage, Burton H. Jr., Adrion, Robert F.
• Primary Examiner(s): Krass, Errol A.
• Assistant Examiner(s): Teska, Kevin J.
• Application Number:
  US06/649,449
• Time in Patent Office: 959 Days
• Field of Search: 364/497-500, 364/550, 364/554, 364/555, 356/39, 356/433, 356/436, 356/441, 356/442, 382/6, 382/36, 436/63, 422/68, 250/461.2
• US Class Current: 702/19
• CPC Class Codes: G01N 15/1456 without spatial resolution of the texture or inner structure of the particle, e.g. processing of pulse signals
• CPC Class Codes: G06F 18/40 Software arrangements specially adapted for pattern recognition, e.g. user interfaces or toolboxes therefor
• CPC Class Codes: G06V 20/698 Matching; Classification

Abstract

This invention concerns a flow apparatus and method for the detection of particles in a sample. Particles are moved, substantially one at a time, in a fluid flow stream. An incident beam of illumination is provided so as to be directed at the particles in the flow stream. Data associated with each moving particle as it passes through the beam of illumination is detected. A class of particles is established, these particles having common characteristics based on the data detected from such class of particles. The data is then stored. Such stored data is compared to data detected from sample particles of an unknown class. A determination is then made that particles from the unknown class belong to the established class as a result of matching respective data.

Claims

1. A flow cytometry apparatus for the detection of biological particles of interest from a sample of unknown biological particles comprising:

means for moving particles from a sample of biological particles of an unknown class, substantially one at a time, in a fluid flow stream;

means for providing an incident beam of illumination directed at the particles in said flow stream;

means for detecting light-related data associated with each moving particle as the particle passes through said beam of illumination;

means for storing data representing a known class of particles having common characteristics based on light data collected from such known class of particles;

means for applying such stored data to light data detected from sample particles of said unknown class in order to match patterns between the data of the known and unknown particles; and

means for recognizing that particles from said unknown class belong to said known class as a result of matching patterns of respective light-related data.

2. The apparatus of claim 1 wherein said means for detecting light-related data includes means for detecting a plurality of different light signals.

3. The apparatus of claim 2 wherein said means for detecting includes means for detecting light scattered by the particles and fluorescence emitted from the particles passing through said beam of illumination.

4. The apparatus of claim 3 wherein said means for detecting detects said light scatter and fluorescence signals simultaneously.

5. The apparatus of claim 3 wherein said means for detecting includes means for detecting light scattered in at least two different directions and for detecting fluorescence emitted by particles at a minimum of two different wavelengths.

6. The apparatus of claim 1 wherein said means for storing includes an electrical circuit with memory means, said circuit including means for converting light signals into electrical signals.

7. The apparatus of claim 1 wherein said means for applying data to match patterns includes decision function means operative to partition a space for measuring light data into a region which contains the sample particle data belonging to said known class.

8. The apparatus of claim 7 wherein said decision function means includes means for transforming the data of said measurement space in order to cluster the sample particle data into data points representing elements of said known class.

9. The apparatus of claim 8 wherein said clustering transformation means is capable of transforming vectors relating to pattern data of said measurement space to a new space according to the following relationship:

space="preserve" listing-type="equation">x'"'"'=Wx

wherein;

x is the original vector of the measurement space

x'"'"' is the transformed vector

W is the transformation matrix comprised of variable weighting coefficients, w_ij.

10. The apparatus of claim 9 wherein said clustering transformation means is capable of minimizing the intraset distance between data points of the known class.

11. The apparatus of claim 10 wherein the transformation means is capable of relying upon the following relationship in minimizing intraset distance for data points in the transformed space:

##EQU7## where;

D² is the mean square intraset distance between two points in a multidimensional space

σ

_k is the unbiased sample variance of the components along the x direction w_kk is the variable weighting factor, and for minimizing D², ##EQU8##

12. The apparatus of claim 7 wherein in said decision function means further includes means for extracting features from said stored data of the known particles for the discrimination of data related to said unknown particles.

13. The apparatus of claim 12 wherein said means for extracting is capable of extracting a plurality of features of said stored data, each feature being a linear combination of all of the original stored data measurements.

14. The apparatus of claim 13 wherein said means for extracting is capable of extracting uncorrelated features of said stored data when the measurements of said features are correlated in nature, whereby the mathematical form of the function which measures the correlation among the features is a covariance matrix.

15. The apparatus of claim 14 wherein said means for extracting is capable of transforming the stored data of the original measurement space to another representation space and of determining a subspace for the class which substantially preserves the data available in the original measurement space, according the the following mathematical relationship:

space="preserve" listing-type="equation">C'"'"'=WCW.sup.T

where,

C'"'"' is the transformed covariance matrix

C is the original covariance matrix

W is the transformation matrix comprised of variable weighing coefficients, w_ij

W^T is the transposed form of the transformation matrix, W.

16. The apparatus of claim 15 wherein said means for extracting is capable of decoupling the covariance of the component features by relying on a modal matrix which transforms the original measurement space into a new space wherein the covariance matrix is diagonal and the coordinates of the new space are uncorrelated.

17. The apparatus of claim 16 wherein said means for recognizing includes means for defining a region for said known class as a hypersphere with a specified radius in the new space, said recognizing means further capable of designating that an unknown particle belongs to said known class if its distance from the cluster center is less than the radius of said hypersphere.

18. An apparatus for the detection of particles in a sample comprising:

sensing means, including a sensing zone, for sensing characteristic features of particles;

means for moving particles of an unknown class, substantially one at a time, in a fluid flow stream through said sensing zone;

means for detecting data associated with the characteristic features of each moving particle as the particle passes through said sensing zone;

means for storing data from a known class of particles having common characteristics in accordance with a pattern of data collected from such known class of particles; and

means for recognizing that sample particles from said unknown class belong to said known class responsive to the recognition of said pattern by data detected from said sample particles.

19. A flow cytometry apparatus for the detection of biological particles of interest from a heterogeneous population of biological particles comprising:

means for moving particles, substantially one at a time, in a liquid flow stream;

means for providing an incident beam of illumination directed at the particles in said flow stream;

means for simultaneously detecting light scattered by and fluorescence emitted from each moving particle as the particle passes through said beam of illumination;

means for storing data representing a known class of particles having common characteristics based on scattered light and fluorescence data collected from such known class of particles;

means for transforming light data detected from sample particles of an unknown class in order to cluster the data points so transformed;

means for extracting features from said stored data for the discrimination of data related to said unknown particles; and

means associated with said transforming and said extracting means for determining that particles from said unknown class belong to said known class as a result of matching patterns of respective light-related data.

20. A flow cytometry apparatus for the detection and classification of biological particles of interest from a sample of an unknown class of biological particles comprising:

means for moving particles from a sample of unknown biological particles, substantially one at a time, in a fluid flow stream;

means for providing an incident beam of illumination directed at the particles in said flow stream;

means for detecting light-related data associated with each moving particle as the particle passes through said beam of illumination;

means for storing data representing a known class of particles having common characteristics based on said light data collected from such known class of particles;

means for separating data from said known class into a plurality of subclasses thereof, each subclass having its light-related data stored separately;

means for applying such stored data to light data detected from sample particles of said unknown class in order to match patterns between the data of the known and unknown particles; and

means for recognizing that particles from said unknown class belong to said known class as a result of matching patterns of respective light-related data and for identifying the subclasses to which the unknown particles belong.

21. The apparatus of claim 20 wherein said means for detecting is capable of numerically characterizing said subclasses by vectorial representation in a space.

22. The apparatus of claim 21 wherein said means for recognizing is capable of mapping particles into points and particle subclasses into clusters in a space of limited dimensionality.

23. The apparatus of claim 22 wherein said means for recognizing performs said mapping by providing a two-dimensional presentation of four-dimensional data.

24. The apparatus of claim 23 wherein said means for recognizing is capable of extracting from said stored four-dimensional data those features that are most significant for classification purposes, and to linearly transform the representation vectors of the original clusters into a new coordinate system wherein the coordinate variables are mutually uncorrelated and the data from the original clusters are concentrated in the first few axes of the new coordinate system.

25. The apparatus of claim 24 wherein said means for recognizing extracts said features by performing an eigenvector analysis of a covariance matrix associated with data of said original clusters and utilizing the two largest eigenvalues of the covariance matrix as the basis for the new two-dimensional space.

26. The apparatus of claim 20 wherein said means for recognizing is capable of identifying subclasses by utilization of statistical decision means.

27. The apparatus of claim 26 wherein said statistical decision means is capable of partitioning the original data storage region into a plurality of mutually exclusive and exhaustive subregions corresponding to the number of subclasses and assigning a particle to one of said subclasses if its light-related data give the largest probability in the associated subregion.

28. The apparatus of claim 20 wherein said means for recognizing is capable of identifying subclasses by distinguishing particles of interest from background particles by reliance on the data representing structure of the particles to delimit the boundary between signal and noise.

29. The apparatus of claim 28 wherein means is included to cluster data representing subclasses of particles by encoding a set of numbers for each subclass of particles produced by two mathematical transformations, said sets each being further functionally represented by a center and a radius in the transformed space, wherein an unknown particle is identified as belonging to a known subclass if its data vector falls into the specific encoded set of numbers.

30. The apparatus of claim 20 wherein said means for recognizing is capable of identifying subclasses by distinguishing clusters of particles of interest from other clusters or from background particles by reliance on the data representing structure of the particles to delimit the boundary between clusters.

31. The apparatus of claim 30 wherein means is included to distinguish clusters by utilizing a histogram for data representation wherein points of data in the histogram are mapped into a new form represented by large numbers of the frequency axis thereof, the remaining points being assigned to surrounding area with small values for frequency counts.

32. The apparatus of claim 31 wherein said histogram is capable of identifying a cluster of interest in response to the application of a frequency threshold to the histogram data such that all frequency counts greater than the threshold are mapped into the cluster and all other frequency counts are mapped into a background region.

33. The apparatus of claim 20 wherein said means for recognizing is capable of identifying subclasses by examining a similarity measure defined for all pairs of points in all the subclasses.

34. The apparatus of claim 30 wherein said similarity measure is a Euclidean distance between all pairs of points.

35. The apparatus of claim 34 wherein said means for recognizing calculates the distance between all pairs of data points and sorts the calculated values into ascending order to determine the nearest neighbors of all points, unknown particles being assigned to the subclass in which a majority of neighboring points occurs.

36. The apparatus of claim 35 wherein a threshold number of points are stored in said means for storing so that said threshold must be exceeded before the unknown particles are assigned to a subclass, otherwise the particles are classified as background particles.

37. An apparatus for the detection and classification of particles in a sample comprising:

sensing means, including a sensing zone, for sensing characteristic features of particles;

means for moving particles of an unknown class, substantially one at a time, in a fluid flow stream through said sensing zone;

means for detecting data associated with the characteristic features of each moving particle as the particle passes through said sensing zone;

means for storing data from a plurality of known subclasses of particles each having common characteristics in accordance with a pattern of data collected from such known subclasses of particles; and

means for recognizing that sample particles from said unknown class belong to said known subclasses responsive to the recognition of said patterns by data detected from said sample particles.

38. A method for detecting biological particles of interest from a sample of an unknown class of biological particles comprising:

moving particles from a sample of unknown biological particles, substantially one at a time, in a fluid flow stream;

providing an incident beam of illumination directed at the particles in said flow stream;

detecting light-related data associated with each moving particle as the particle passes through said beam of illumination;

storing data representing a known class of particles having common characteristics based on light data collected from such known class of particles;

applying such stored data to light data detected from sample particles of said unknown class in order to match patterns between the data of the known and unknown particles; and

recognizing that particles from said unknown class belong to said known class as a result of matching patterns of respective light-related data.

39. The method of claim 38 wherein said detecting step includes detecting a plurality of different light signals.

40. The method of claim 39 wherein said detecting step includes detecting light scatter and fluorescence signals simultaneously.

41. The method of claim 40 wherein said detecting step includes detecting light scattered in at least two different directions and detecting fluorescence emitted by particles at a minimum of two different wavelengths.

42. The method of claim 38 wherein said applying step includes partitioning a space for measuring light data into a region which contains the sample particle data belonging to said known class.

43. The method of claim 42 wherein the partitioning step further includes transforming the data of said measurement space in order to cluster the sample particle data into data points representing elements of said known class.

44. The method of claim 43 wherein said transforming step transforms vectors relating to pattern data of said measurement space to a new space according to the following relationship:

space="preserve" listing-type="equation">x'"'"'=Wx

wherein;

x is the original vector of the measurement space

x'"'"' is the transformed vector

W is the transformation matrix comprised of variable weighting coefficients, w_ij.

45. The method of claim 44 wherein said transforming step includes minimizing the intraset distance between data points of the known class.

46. The method of claim 42 wherein said partitioning step includes extracting features from said stored data of the known for the discrimination of data related to said unknown particles.

47. The method of claim 46 wherein said extracting step includes extracting a plurality of features of said stored data, each feature being a linear combination of all of the original stored data measurements.

48. The method of claim 47 wherein said extracting step includes extracting uncorrelated features of said stored data when the measurements of said features are correlated in nature, whereby the mathematical form of the function which measures the correlation among the features is a covariance matrix.

49. The method of claim 48 wherein said extracting step includes transforming the stored data of the original measurement to another representation space and determining a subspace for the class which substantially preserves the data available in the original measurement space according to the following mathematical relationship:

space="preserve" listing-type="equation">C'"'"'=WCW.sup.T

where

C'"'"' is the transformed covariance matrix

C is the original covariance matrix

W is the transformation matrix comprised of variable weighting coefficients, w_ij

W^T is the transposed form of the transformation matrix, W.

50. A method for detecting particles in a sample comprising:

moving particles, substantially one at a time, in a fluid flow stream through a sensing zone;

detecting data associated with characteristic features of each moving particle as the particle passes through said sensing zone;

storing data from a known class of particles having common characteristics in accordance with a pattern of data collected from such known class of particles; and

recognizing that sample particles from an unknown class belong to said known class responsive to the recognition of said pattern by data detected from said sample particles.

51. A method for detecting particles in a sample comprising:

storing data from a class of particles having known, common characteristics in accordance with a pattern of data collected from such class of known particles; and

recognizing that sample particles from an unknown class belong to said known class responsive to the recognition of said pattern by data detected from said unknown sample particles.

52. A method for detecting and classifying biological particles of interest from a sample of an unknown class of biological particles comprising:

moving particles from a sample of unknown biological particles, substantially one at a time, in a fluid flow stream;

providing an incident beam of illumination directed at the particles in said flow stream;

detecting light-related data associated with each moving particle as the particle passes through said beam of illumination;

storing data representing a known class of particles having common characteristics based on said light data collected from such known class of particles;

separating data from said known class into a plurality of subclasses thereof, each subclass having its light-related data stored separately;

applying such stored data to light data detected from sample particles of said unknown class in order to match patterns between the data of the known and unknown particles; and

recognizing that particles from said unknown class belong to said known class as a result of matching patterns of respective light-related data and identifying the subclasses to which the unknown particles belong.

53. The method of claim 52 wherein said recognizing step includes numerically characterizing said subclasses by vectorial representation in a space.

54. The method of claim 53 wherein said recognizing step includes mapping particles into points and particle subclasses into clusters in a space of limited dimensionality.

55. The method of claim 54 wherein said mapping is accomplished by providing a two-dimensional presentation of four-dimensional data.

56. The method of claim 55 wherein said recognizing step includes extracting from said stored four-dimensional data those features that are most significant for classification purposes, and transforming the representation vectors of the original clusters into a new coordinate system wherein the coordinate variables are mutually uncorrelated and the data from the original clusters are concentrated in the first few axes of the new coordinate system.

57. The method of claim 52 wherein said recognizing step identifies subclasses by a statistical decision step which includes partitioning the original data storage region into a plurality of mutually exclusive and exhaustive subregions corresponding to the number of subclasses and assigning a particle to one of said subclasses if its light-related data give the largest probability in the associated subregion.

58. The method of claim 57 wherein said partitioning step includes clustering data representing subclasses of particles by encoding a set of numbers for each subclass of particles produced by two mathematical transformations, said sets each being further functionally represented by a center and a radius in the transformed space, wherein an unknown particle is identified as belonging to a known subclass if its data vector falls into the specific encoded set of numbers.

59. The method of claim 52 wherein said recognizing step for identifying subclasses includes distinguishing clusters of particles of interest from other clusters or from background particles by reliance on the data representing structure of the particles to delimit the boundary between clusters.

60. The method of claim 52 wherein said recognizing step for identifying subclasses includes examining a similarity measure defined for all pairs of points in all of the subclasses.

61. The method of claim 60 wherein said examining step includes calculating the distance between all pairs of data points and sorting the calculated values into ascending order to determing the nearest neighbors of all points, unknown particles being assigned to the subclass in which a majority of neightboring points occurs.

62. The method of claim 52 wherein the particles to be detected and classified are cells.

63. The method of claim 52 wherein the particles to be detected and classified are a plurality of subclasses of bacteria.

64. The method of claim 63 wherein the bacteria to be detected and classified are derived from a sample of biological fluid.

65. The method of claim 64 wherein the biological fluid is urine.

66. The method of claim 52 wherein the particles to be detected and classified are different subclasses of leukocytes.

67. The method of claim 52 wherein said storing step includes preparing a liquid sample with a high concentration of a known type of particles having common characteristics, and collecting and storing light data from such known class of particles to establish a region having a recognizable pattern of data points.

68. A method for detecting and classifying particles in a sample comprising:

moving particles of an unknown class, substantially one at a time, in a fluid flow stream through a sensing zone;

detecting data associated with characteristic features of each moving particle as the particle passes through said sensing zone;

storing data from a plurality of subclasses of particles each having known common characteristics in accordance with patterns of data associated with such known subclasses of particles; and

recognizing that sample particles from said unknown class belong to said known subclasses responsive to the recognition of said patterns by data associated with said sample particles.

69. A method for detecting and classifying particles in a sample comprising:

storing data from a plurality of subclasses of particles each having known common characteristics in accordance with patterns of data associated with such known subclasses of particles; and

recognizing that sample particles from an unknown class belong to said known subclasses responsive to the recognition of said patterns by data associated with said unknown sample particles.