Desktop version

Home arrow Engineering arrow Cellular Image Classification

Experimental Results on the ICPR2012 Dataset

We first test performance of the proposed LLDC method on the ICPR2012 dataset following the experimental protocol of the HEp-2 cells classification contest by dividing the cell images into a training set and a test set. The subdivision is performed while maintaining approximately the same image pattern distribution over the two sets [10]. To validate the efficiency of our proposed method for staining pattern classification, we compare four different image representation: the original SIFT based BoW image representation (LLC/LSC-sift), the distance vector based image representation (LLC/LSC-(sift+dv)), our proposed image representation using local distance vector (LLC/LSC-ldv) and the proposed concatenated image representation (LLC/LSC-(sift+ldv)). Table 5.1 gives the comparison results of the classification performance at the cell level and at the image level. It can be observed that, the proposed LLDC method outperforms all the other methods. It is worth noting that the LLDC method outperforms CoALBP [11] which is the winner of the contest with 70.4% of classification accuracy and 68.4% of sensitivity (at the cell level). Furthermore, the performance obtained by LLC/LSC-(sift+ldv) is better than that obtained by LLC/LSC-sift and LLC/LSC-ldv respectively. In particular, the classification performance achieved by LSC-(sift+ldv) is better than that achieved by LLC-(sift+ldv).

Table 5.2 shows the confusion matrix at the cell level by the proposed LLDC method using the LSC strategy on the concatenated image representation. The entry (corresponding to row i and column j) in the confusion matrix represents the percentage of cells from class i assigned to class j. It is obvious that cytoplasmic, centromere and homogeneous patterns are classified more accurately than the others. Particularly, cytoplasmic can achieve 100 % of classification accuracy. Compared to the cytoplasmic pattern with distinguishable shape and centromere pattern with clear fluorescent dots, fine speckled pattern and homogeneous pattern has similar characteristic and is hard to find discriminative features to separate.

To evaluate the classification performance at the image level, we report the corresponding confusion matrix in Table 5.3. Similarly, the table represents the percentage

Algorithm

Cell level

classification accuracy

(%)

Cell level sensitivity (%)

Image level classification accuracy

(%)

LLC-(sift+ldv)a

70.9

71.6

78.6

LLC-(sift+dv)b

67.7

69.2

78.6

LLC-ldvc

67.4

68.9

78.6

LLC-siftd

66.4

68.1

78.6

LSC-(sift+ldv)d

71.7

72.9

85.7

LSC-(sift+dv)f

69.3

70.6

78.6

LSC-ldvg

69.1

70.6

78.6

LSC-sifth

68.9

70.6

78.6

a The LLDC method based on the concatenated representations of LLC-sift and LLC-ldv bThe LDC method based on the concatenated representations of LLC-sift and LLC-dv cThe LLC method based on local distance vector dThe LLC method based on SIFT features

eThe LLDC method based on the concatenated representations of LSC-sift and LSC-ldv fThe LDC method based on the concatenated representations of LSC-sift and LSC-dv gThe LSC method based on local distance vector hThe LSC method based on SIFT features

Table 5.2 The confusion matrix for the cell level classification on the ICPR2012 dataset

ce (%)

cs (%)

cy (%)

fs (%)

ho (%)

nu (%)

ce

84.2

3.4

0.6

0.0

1.3

10.5

cs

6.9

72.3

3.8

12.1

4.9

0.0

cy

0.0

0.0

100.0

0.0

0.0

0.0

fs

22.8

1.7

2.6

39.1

32.9

0.9

ho

6.7

4.5

0.0

11.1

76.0

1.7

nu

20.1

0.0

5.6

0.7

7.9

65.7

Table 5.3 The confusion matrix for the image level classification on the ICPR2012 dataset

ce (%)

cs (%)

cy (%)

fs (%)

ho (%)

nu (%)

ce

100.0

0.0

0.0

0.0

0.0

0.0

cs

0.0

66.7

0.0

33.3

0.0

0.0

cy

0.0

0.0

100.0

0.0

0.0

0.0

fs

0.0

0.0

0.0

50.0

50.0

0.0

ho

0.0

0.0

0.0

0.0

100.0

0.0

nu

0.0

0.0

0.0

0.0

0.0

100.0

of images of class i identified to class j with respect to the total number of images in the test set. Our proposed LLDC method obtains 85.7 % of the image-level classification accuracy. Centromere, cytoplasmic, homogeneous, and nucleolar patterns achieve

Algorithm

Cell level

classification accuracy

(%)

Cell level sensitivity

(%)

Image level classification accuracy

(%)

LLC-(sift+ldv)

79.1

76.2

90.2

LLC-(sift+dv)

76.1

73.6

87.8

LLC-ldv

75.1

72.3

87.8

LLC-sift

75.8

73.3

87.8

LSC-(sift+ldv)

77.6

74.5

90.2

LSC-(sift+dv)

75.3

72.8

87.8

LSC-ldv

74.6

72.1

87.8

LSC-sift

75.6

72.7

87.8

100 % of classification accuracy. The most frequent mistake is existed between fine speckled and homogeneous pattern, which is the common mistake at the cell level.

 
Source
< Prev   CONTENTS   Source   Next >

Related topics