Desktop version

Home arrow Engineering arrow Cellular Image Classification

Experimental Results on the ICPR2012 Dataset

We firstly test performance of our automatic staining pattern classification system following the experimental protocol of the ICPR’12 HEp-2 cells classification contest by dividing the ICPR2012 dataset into one training set and one test set. A linear SVM is trained by 10-fold cross validation based on training set. Table 6.2 shows the classification performance of each algorithm. Our proposed algorithm outperforms all the other algorithms at the cell level, achieving 75.9 % of classification rate on individual cells. It is worth noting that our proposed PLTP-SRI feature and the combined feature of PLTP-SRI and SIFT-BoW can respectively outperform CoALBP [1] which is the winner of the contest with accuracy of 68.7 %.

Table 6.3 illustrates the confusion matrices presenting the classification performance for each staining pattern at the cell level. It is obvious that cytoplasmic, centromere and homogeneous patterns are classified more accurately than the other patterns. More particularly, cytoplasmic can achieve 100 % of classification accuracy. However, the fine speckled pattern is hard to categorize, with 37.7 % of classification accuracy. It is easily mis-classified as homogeneous pattern. We find that the differences between fine speckled and homogeneous are subtle, therefore it is difficult to find discriminative features to separate them.

With respect to the classification performance at the image level, we use the majority voting scheme. As can be seen in the Table 6.2, our proposed algorithm and CoALBP achieves the highest classification accuracy of 85.7 %. Table 6.4 shows the corresponding confusion matrix. We can see that the element in the confusion matrix

Table 6.2 Classification performance on the ICPR2012 dataset

Algorithm

Cell level accuracy (%)

Sensitivity (%)

Image level accuracy (%)

Our algorithm

75.9

76.5

85.7

PLTP-SRI

70.2

71.2

78.6

CoALBP [13]

68.7

70.4

85.7

SIFT-BoW

68.9

70.6

78.6

RIC-LBP [2]

67.5

67.6

71.4

CLBP

61.2

61.1

71.4

LBP

58.9

59.2

64.3

Table 6.3 The confusion matrix for the cell level classification on the ICPR2012 dataset

ce (%)

cs (%)

cy (%)

fs (%)

ho (%)

nu (%)

ce

92.0

0.0

0.0

0.7

0.0

7.3

cs

7.9

70.3

1.0

19.8

1.0

0.0

cy

0.0

0.0

100.0

0.0

0.0

0.0

fs

29.0

1.8

0.9

37.7

30.6

0.0

ho

6.7

2.2

0.0

7.2

82.2

1.7

nu

16.5

2.2

0.0

0.0

4.3

77.0

Table 6.4 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

50.0

0.0

0.0

50.0

0.0

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

at the image level is similar to that of the confusion matrix at the cell level. It is obviously that one coarse speckled image is wrongly classified into fine speckled class while one fine speckled image is identified as centromere by mistake.

 
Source
< Prev   CONTENTS   Source   Next >

Related topics