Desktop version

Home arrow Engineering arrow Cellular Image Classification

Source

Experimental Results on the ICIP2013 Training Dataset

In this section, we evaluate the AdaCoDT method on the ICIP2013 training dataset by dividing the dataset into training set (6842 cells) and test sets (6754 cells). The classification performance of different methods at the cell level are shown in Table 7.5. Our proposed AdaCoDT method achieves the best cell-level performance again. Although the AdaCoDT method almost achieves the same classification performance with the LSC-SIFT method, it is worth noting that the size of codebook for the BoW framework is 1024 while the number of GMM components for the AdaCoDT method is only 128. With the same codebook size (for the AdaCoDT method, the number of GMM components can be seen as the codebook size), the AdaCoDT method significantly outperforms the BoW framework. Table7.6 shows the confusion matrix of the AdaCoDT method at the cell level. Homogeneous pattern gets the highest classification accuracy rate of 89.5 %, followed by nuclear membrane 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 large speckles throughout the interphase nuclei. The classification accuracy for golgi pattern is only 43.6 %. The characteristic of golgi pattern is similar to that of nucleolar pattern. On the other hand, there are only 362 golgi pattern cells from 4 slide images in the ICIP2013 training dataset, which influences its classification performance. Table7.7 illustrates the confusion matrix at the image level. The AdaCoDT method obtains the classification accuracy of 87.8 % at image level, which means that 36 slide images are correctly identified while there are 41 slide images in the test set. Homogeneous, nucleolar and nuclear membrane patterns particularly obtain 100 % image level accuracy. It is evident that golgi pattern is wrongly classified as nucleolar, which is very common at the cell level.

Table 7.5 Classification performance on the ICIP2013 training dataset

Algorithm

Average accuracy (%)

Sensitivity (%)

AdaCoDT

75.8

72.9

LSC-SIFT

75.6

72.7

LSC-CoDT

70.6

69.8

FK-SIFT

69.7

68.3

CoALBP [25]

67.1

65.8

RIC-LBP [20]

66.4

64.4

LBP [1]

60.7

54.5

Table 7.6 The confusion matrix for the ceil level classification on the ICIP2013 training dataset

ho (%)

sp (%)

nu (%)

ce (%)

nm (%)

go (%)

ho

89.5

3.4

4.1

0.3

2.4

0.3

sp

11.6

66.7

5.6

15.1

0.7

0.3

nu

0.8

8.1

74.2

11.7

2.6

2.6

ce

0.5

22.2

2.4

74.7

0.0

0.2

nm

1.1

2.4

1.1

0.4

88.7

6.3

go

6.6

5.0

38.2

0.8

5.8

43.6

Table 7.7 The confusion matrix for the image level classification on the ICIP2013 training dataset

ho (%)

sp (%)

nu (%)

ce (%)

nm (%)

go (%)

ho

100.0

0.0

0.0

0.0

0.0

0.0

sp

12.5

75.0

0.0

12.5

0.0

0.0

nu

0.0

0.0

100.0

0.0

0.0

0.0

ce

0.0

25.0

0.0

75.0

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
Found a mistake? Please highlight the word and press Shift + Enter  
< Prev   CONTENTS   Next >

Related topics