Lectures 3,4: Medical Data Processing & Analysis Flashcards
3.1 Explain the effect of background on the perception of an object
The human visual system has band-pass filter characteristics, which lead to responses that are proportional to differences between illumination levels, rather than to absolute illumination levels. That is, the human visual system tends to under/overshoot around the boundary of regions of different intensities.
For example, two squares of the same grey level value, eg 130, may be placed on two different background regions, one lighter (150), one darker (50). The lighter background will make the inner square region appear darker than the same square against the darker background, despite being of the same shade value.
– More detail:
This can be explained by simultaneous contrast:
- Lighter background: (130-150)/150 = -0.1333
- Darker background: (130-50)/50 = +1.6
Or by normal contrast:
- Lighter: (130-150)/(130+150) = -0.0714
- Darker: (130-50)/(130+50) = 0.4444
The advantage of normal contrast is that the values are limited to the range [-1,1]
A negative contrast value for the square against the lighter background indicates that it’s darker than its background, whereas the positive contrast value indicates that it’s lighter than its background.
3.2 Explain why the Just-Noticeable Difference (JND) is important in characterisation of medical image quality.
The concept of JND is important in analysing contrast, visibility, and the quality of medical images
Experiments have shown that the JND is almost constant, at approximately 0.02 or 2%, over a wide range of background intensity - Weber’s Law.
In the visualisation of medical images, features may or may not always be easily distinguished, eg in micro-calcifications in tissues. They may possess high contrast against fat and low-density tissue and be easily visible, or they may have low contrast when present with a high-density tumour - close to the JND - and be difficult to detect.
Such features present significant challenges in eg breast cancer screening, as the human visual system cannot easily recognise contrast levels close to the JND, and enhancement of the contrast and visibility of such features could assist in improving the accuracy of detecting early breast cancer.
3.3 Explain the use of a digital image Histogram in the characterisation of Image Quality and an example of its use.
The histogram provides a view of the intensity profile of an image, often displayed as a bar chart. Pixel values are partitioned and counted with the population of each partition value placed in its own bin.
Pixel intensities are plotted along the x-axis and occurrences for each intensity against the y-axis.
Histograms can be viewed as probability density functions, and allow for an assessment of intensity frequencies that may be indicative of diagnosis depending on the image context, such as in high-intensity tumour values in mammography images.
(draw example? 16 grey level histogram just like slides)
[MR Angiography (MRA) uses a contrast agent to improve contrast in vasculature; images have a dark background and tend to be of low overall contrast.]
3.4 How does Noise Impact MI quality?
Any image/pattern/signal other than that of interest could be termed as interference/artifact/noise, which could be a result of physiological processes, the instrumentation used or the environment of the experiment. Typically the human body is a full of biological processes that present sources of noise in biomedical images, eg movement:
- Respiratory or cardiovascular activity in imaging of the chest
- Peristalsis of the GIT in imaging of the abdomen
- Pulsatile movement of arteries in subtraction angiography
Common types of noise include:
- Salt-and-pepper - random occurrences of black and white pixels
- Impulsive - random occurrences of white pixels
- Gaussian - variations in intensity drawn from a Gaussian normal distribution
All of the above may have an impact on MI quality and detract from a physician’s ability to extract relevant medical information about the patient.
3.5 What is the main difference between the Digital Image Processing methods:
- Spatial domain methods
- Frequency (transform) domain methods
- Spatial domain processing techniques are based on direct manipulation of pixels in an image.
- Frequency (transform) domain processing techniques are based on modifying the Fourier (or others) transform of an image.
3.6 What are Function Curves and how are they useful in the manipulation of Medical Images?
Graphs/maps that represent and control different attributes of an image, eg brightness/colour. These attributes can be easily modified by manipulating the function curves without having to alter the image directly with a retouching tool.
Making image manipulations that involves all of the image or large portions of it are best performed with function curves.
Function curves for image manipulation are usually represented by a line that starts a the lower left corner of a square and ends at the upper right corner. The straight diagonal line represents one or several untouched attributes of the original image. Any changes to the line will result in changes to the image.
3.7 What are Grey Level Transformations (GLT) and how are they useful in the manipulation of Medical Images?
Simple image enhancement techniques that transform original pixel values “s” based on a transformation function “T”, i.e. “s=T(r)”
There are numerous types of GLTs:
- Linear - negative, identity transformations - suited for enhancing white/grey detail embedded in dark regions, especially when the black areas are dominant in size. Reverses the order of pixel intensities. Eg in mammograms, angiograms for visualisation of tumours, blood vessels (respectively)
- Logarithmic - log, inverse-log - dynamic range of an image may exceed capability of display device, eg only the brightest parts of an image like an x-ray may be visible. Log GLTs map narrow range of low input GLTs to wider output values to expand values of dark pixels in an image while compressing the higher level values.
- Power-law - n’th power and n’th root - “s = c*r^gamma” where c and gamma are positive constants - increases contrast in dark areas, decreases contrast in bright areas. Can be used when clinically relevant info is situated in dark areas like the lungs.
Also: “Window Level operation” (see 3.8) and “Pseudo-colour table transformation” (see 3.9)
3.8 Describe the GLT: Window-Level Operation (“Window-Centre Adjustment”)
An interval/window is selected in the original grey level range, determined by the window centre “l” and the window width “w”.
Contrast outside the window is lost completely, whereas the portion of the range lying inside the window is stretched to the original grey level range (“contrast stretching”)
Eg grey scale region between 0.5 & 0.75 in a mammogram can be expanded to the full [0,1] range, this GLT useful for highlighting an intensity band of interest, eg that might display a spiculated tumour more clearly.
3.9 Describe the use of colour in GLT’s
Colour is a visual feature immediately perceived when looking at an image, but it’s not normally captured in medical imaging, which are usually displayed in grayscale, or false colour (“pseudo”).
Pseudo-colour transformations are useful for analysing MI’s as humans can discern thousands of colour shades/intensities compared to about two dozen shades of grey.
Grey values are signed colour based on criterion, eg individual input brightness. “Intensity Slicing Coding” can be used to assign colours if the image can be interpreted as a 3D function (intensity versus spatial coordinates), where a different colour is assigned to each side of a plane separating the 3D function. More sophisticated colour levels and visualisation can be established with multiple slices, eg creating “rainbow” representations of medical image features (separating organs, bone, etc.)
3.10 Describe Temporal Subtraction (“Multi-image Operation”) and how it may be useful in MI
Operation that subtracts images in a pixel-wise way.
For two images I1 and I2, the difference I- is defined as:
I-(i,j) = I1(i,j) - I2(x,y)
Subtraction can be used to get rid of the background in two similar images.
Eg in angiography, two images are made - one with and one without a contrast agent injected in the blood vessels. Subtraction of these two images yields a pure image of the blood vessels because the subtraction deletes the other anatomical features (background).
3.11 Describe Temporal Average and how it may be used in MI.
Operation that adds images in a pixel-wise way.
For two images I1 and I2, the difference I+ is defined as:
I+(i,j) = I1(i,j) + I2(x,y)
If these operations yield values outside the original dynamic range, the resulting image can be brought back to that range by a linear transformation.
The average of n images is defined:
Iav(i,j) - (1/n)(I1(i,j) + … + In(i,j))
Averaging can be useful to decrease the noise in a sequence of images of a motionless object. The random noise averages out, whereas the object remains unchanged (if the images are perfectly registered).
Eg an MRI image with a low signal to noise ratio (SNR) may be improved in quality by obtaining more subsequent images of the same slice, and averaged to increase this SNR
3.12 How can Histogram Transformation be used in MI Analysis?
An image whose pixels tend to occupy the entire range of possible grey levels and tend to be distributed uniformly, will have an appearance of high contrast and exhibit a large variety of grey tones. This can be obtained through a transformation function based only on the information available in the histogram of an input image:
Histogram Equalisation - mapping to increase the contrast in an image by stretching its histogram to approximately uniformly distributed. The image that has been histogram equalised always has pixels that reach the brightest grey level.
Eg a normal abdominal image with narrow pixel distribution between 50-100 may be histogram-equalised to have pixels distributed out over the entire dynamic range such that better definition of the kidney and organs may be obtained.
What types of images are represented by the following histograms?
(see attached)
Answers:
A) Dark Image
B) Bright Image
C) Low-contrast Image
D) High-contrast Image
3.14 Describe Image Enhancement by Spatial Filtering
Spatial domain processes can be denoted by: g(x,y) = T[f(x,y)]
- f(x,y) is input image, g(x,y) is processed image, T is operator on f, defined over neighbourhood (x,y)
- Aka. pixel group processing, mask processing or filtering
- Sub-image = (spatial) filter, mask, kernal, template or window. Values in this sub-image are ‘coefficients’ (rather than pixels)
The process consists simply of moving the filter mask from point to point in an image. At each point (x,y) the response of the filter at that point is calculated using a predefined relationship.
Eg “linear spatial filtering” - response given by sum of products of the filter coefficients of the corresponding image pixels in the area spanned by the filter mask.
3.15 Describe “Image Smoothening” and why one might use it.
Smoothing filters (type of spatial filtering) used for blurring and noise reduction, eg in pre-processing steps such as removal of small details from an image prior to (large) object extraction, and ridging of small gaps in lines/curves.
Examples include
- Linear smoothing filters (eg average filters: box filter, weighted filter)
- Non-linear smoothing filters (eg median filter, max filter, min filter)
3.16 Describe Averaging Filters
Output of a smoothing, linear spatial filter is the average of the pixels contained in the neighbourhood of the filter mask.
Eg a box filter has all coefficients equal whilst a weighted filter has pixels in the neighbourhood multiplied by different coefficients (more ‘weight’ or importance to some pixels than others)
(see attached)
4.9 How may image noise detract from the ability to use a histogram for thresholding?
Large amounts of noise prevent the differentiation of histogram modes despite specific grey level features showing some visibility to the human eye.
Without additional processing there is little ability to find a suitable threshold for segmentation.
(see attached image; draw histograms in exam)
4.11 How may Illumination and Reflectance impact image thresholding?
Illumination may be present in an image as an “intensity ramp”, a directional gradient of pixel intensities. This may distort the image histogram intensity modes and thus selection of a threshold for segmentation.
- (see attached image)
Reflectance presents in images when pixel intensities may be non-uniform based on natural reflectivity variations in the surface of objects and/or background.
