Pancreatic Cancer Survival Prediction: A Survey of the State-of-the-Art

Table 5

Feature extraction advantages and disadvantages.

(a)

Method	Description (texture features)	Advantage	Disadvantage
Gabor wavelet transform [52, 55, 58]	In information theory applications, Dennis Gabor used complex functions to build wavelets forming a basis for Fourier transforms.	Multiscale robust	Incomplete cover of spectrum plane needs rotation normalization
GLCM-based method [48]	Find the frequency of a set of pixel and its spatial relationship in an image to characterize its texture.	Easy to use compact robust	High computation cost partial description of texture

(b)

Method	Description (colour features)	Advantage	Disadvantage
Scalable Colour Descriptor [48]	Defined in the hue-saturation-value (HSV) colour space with fixed colour space quantisation and uses a novel, Haar transform encoding	Compact and robust never changing and uninterrupted	Needs postprocessing for spatial information
Colour histogram [47, 48, 53]	A histogram represents the dispersion of colours in an image. It can be visualised as a graph that gives a high level of suspicion regarding the pixel value distribution	Simple to compute, easy to use and understand	High dimension sensitive to information

(c)

Method	Description (shape features)	Advantage	Disadvantage
Wavelet transform [47]	Mathematical means for performing signal analysis when the signal frequency varies over time.	Translation and scale invariant good affine transformation good noise resistance	Average occultation resistance
Zernike moments [48]	A set of rotation invariant features is introduced. They are the magnitude of a set of orthogonal complex moments of the image.	Good noise resistance	High computational complexity bad affine transform