Desktop version

Home arrow Engineering arrow Cellular Image Classification

Experimental Results on the ICIP2013 Training Dataset

We evaluate our algorithm on the ICIP2013 training dataset following the way similar to experimental protocol of the ICIP’13 contest. The parameters of each algorithm

Table 6.5 Classification performance on the ICIP2013 training dataset

Algorithm

Cell level accuracy

(%)

Sensitivity (%)

Image level accuracy

(%)

Our algorithm

77.1

74.4

87.8

SIFT-BoW

75.6

72.7

87.8

PLTP-SRI

74.6

73.6

87.8

CoALBP

67.1

65.8

75.6

RIC-LBP

66.4

64.4

75.6

CLBP

63.5

61.2

73.2

LBP

60.7

54.5

65.9

Table 6.6 The confusion matrix for the cell level classification on the ICIP2013 training dataset

ho (%)

sp (%)

nu (%)

ce (%)

nm (%)

go (%)

ho

78.1

14.1

5.1

0.0

2.6

0.1

sp

4.7

68.8

13.4

12.2

0.7

0.2

nu

1.5

5.2

80.6

7.3

2.8

2.6

ce

0.7

16.1

1.9

78.8

1.1

1.4

nm

3.2

1.9

0.9

0.7

90.1

3.2

go

6.1

0.8

37.8

1.1

3.9

50.3

are the same with those for ICPR2012 dataset as shown in Table 6.1. The classification performance of different algorithms are shown in Table 6.5. Our proposed algorithm achieves the best cell level performance again. Table6.6 shows the confusion matrix of our proposed algorithm at the cell level. Nuclear membrane pattern gets the highest classification accuracy rate of 90.1 %, followed by nucleolar as they have distinguished characteristic compared with other patterns. Speckled and golgi patterns are hard to be recognized by comparison. Speckled pattern is easily to be categorized into centromere pattern, as they both have a uniform diffuse fluorescence of the entire interphase nuclei. It is sometimes mis-classified into nucleolar pattern, because some nucleolar patterns which are characterized by weakly clustered small granules are similar to speckled pattern. Golgi pattern is often mistaken for nucleolar pattern, because some golgi pattern have large speckles within the nucleoli while some only have several cluster of irregular granules, which is just similar to nucleolar pattern. On the other hand, golgi pattern only contains cells from 4 slide images, which influences its classification performance. Table6.7 illustrates the confusion matrix at the image level. The proposed algorithm obtains the classification accuracy of 87.8 % at image level, which indicates that 36 images are correctly identified while there are 41 images in the test set. Both nucleolar and nuclear membrane patterns particularly obtain 100 % image level accuracy. The mistakes happened in the image level classification are just similar to that in the cell level classification.

Table 6.7 The confusion matrix for the image level classification on the ICIP2013 training dataset

ho (%)

sp (%)

nu (%)

ce (%)

nm (%)

go (%)

ho

87.5

12.5

0.0

0.0

0.0

0.0

sp

0.0

75.0

12.5

12.5

0.0

0.0

nu

0.0

0.0

100.0

0.0

0.0

0.0

ce

0.0

12.5

0.0

87.5

0.0

0.0

nm

0.0

0.0

0.0

0.0

100.0

0.0

go

0.0

0.0

50.0

0.0

0.0

50.0

 
Source
< Prev   CONTENTS   Source   Next >

Related topics