Page 105

proefschrift_Schols_SLV

Multispectral characterization of tissues encountered during laparoscopic colorectal surgery Supplementary section TPCR9 Total Principal Component Regression (TPCR) is a mathematical method that can be applied to automatically learn specific features in measured patterns (sometimes referred to as ‘fingerprints’). These features correspond to particular types of objects. In the reported investigation the object types are different tissues and the measured patterns are the respective optical reflectance spectra. TPCR can essentially be seen as a classification method. The TPCR method consists of two phases, the training phase and the classification phase. During the training phase the method is used to calculate a classifier that can be used during the subsequent classification phase on a measured spectrum to determine the sample tissue type. In the following paragraph, the training phase is explained in more detail.   Consider a number of measured tissue samples of different types. A matrix X can be formed where each column contains the optical spectrum values at different selected wavelengths. In addition a matrix Y is formed where each row represents a different tissue type and each column contains the corresponding tissue type indication, 1 for positive and 0 for negative. For illustration purposes consider the following example. Suppose 5 samples are measured (5 columns of X) and each at 8 different light wavelengths (8 rows of X):   00... 1100 00... 1122 33..3. 0000 33... 2233 22... 9999 00... 4400 00... 4433 33... 5566 33... 5500 33... 2200        00... 7700 00... 7777 22..21 2211 2.34 22..3344 33..3.01 0011 00..0.83 8833 0.80 00..8800 1.80 11..8800 11..1.79 7799 11..1.87 8877 00..0.91 9911 00..0.92 9922 11..1.01 0011 11..1.00 0000 00..0.99 9999 00..0.88 8888 00..0.80 8800 00..0.98 9988 11..1.00 0000 00..0.97 9977 00..0.30 3300 00..0.22 2222 00..0.10 1100 00..0.20 2200 00..0.19 1199 1 1 0 0 0                                         The label matrix Y contains the tissue types. Each type is assigned an input label based on the visual judgment of the surgeon who indicates the tissue type for each measurement; this is sometimes referred to as the “ground truth”. In this example there are 2 tissue types (2 rows of Y) and 5 measurements, of which some will be type 1 and some will be type 2. In this instance the first 2 measurements represent the first 103  0.01 0.00 0.21 0.33 0.20        0 0 1 1 1 Y       00..0011 00..0000 00..2211 00..3333 00..2200 XXXX   


proefschrift_Schols_SLV
To see the actual publication please follow the link above