Microscopic Slice Image Processing and Automatic Labeling - Biomedical Signal Analysis: Contemporary Methods and Applications - page 361

1.5

1

0.5

0

0.5

1

1.5

2

4

2 4

3

2

0

1

0

2

1

2

4

3

Figure 13.5
Data set with 120 samples after three-dimensional PCA projection (91% of the data
was retained). The dots mark the 60 samples representing cells; the crosses mark
the 60 non cell data points. The two circles indicate clusters of a k -means
application with a search for two clusters. Obviously, k -means nicely differentiates
between the cell and the non cell components.

using a linear perceptron, projection along the direction of the fifth IC,
together with a sign function, can be used in order to separate cells
and non cells. This is quite interesting because in comparison to the
supervised perceptron learning approach above, the ICA is completely
unsupervised. Only later, when comparing the ICs, do we use the prior
information on cells and non cells in order to differentiate between the
source components. The fact that the data set contains a cell/non cell
independent component was already indicated by the k -means cluster
analysis from figure 13.5, where we saw that the data set clusters into
the cell and non cell components. If we perform PCA to decorrelate
the data, we can also identify a cell/non cell component; however, its
crosscorrelation with the correct classification function is 5% lower than
the ICA result. This confirms that higher-order correlations improve