X-ray Detection & Noise Flashcards
xray Absorbed energy is either
(1) directly producing an image (analogue)
(2) processed electronically (digital)
xray Detector Characteristics
⚫ Spatial resolution
⚫ Spectral resolution
⚫ Dynamic range
⚫ Noise
⚫ Read-out time
⚫ Quantum efficiency
⚫ Field of view
Spatial Resolution types
𝑃𝑜𝑖𝑛𝑡 𝑠𝑝𝑟𝑒𝑎𝑑 𝑓𝑢𝑛𝑐𝑡𝑖𝑜𝑛 (𝑃𝑆𝐹): 𝑛𝑜𝑡 𝑝𝑟𝑎𝑐𝑡𝑖𝑐𝑎𝑙 𝑡𝑜 𝑚𝑒𝑎𝑠𝑢𝑟𝑒 →𝑙𝑜𝑤 𝑓𝑙𝑢𝑥
𝐿𝑖𝑛𝑒 𝑠𝑝𝑟𝑒𝑎𝑑 𝑓𝑢𝑛𝑐𝑡𝑖𝑜𝑛 (𝐿𝑆𝐹): 𝑑𝑒𝑡𝑒𝑐𝑡𝑜𝑟 𝑟𝑒𝑠𝑝𝑜𝑛𝑠𝑒 𝑡𝑜 𝑎 1𝐷 𝑙𝑖𝑛𝑒 𝑖𝑙𝑙𝑢𝑚𝑖𝑛𝑎𝑡𝑖𝑜𝑛
𝐸𝑑𝑔𝑒 𝑠𝑝𝑟𝑒𝑎𝑑 𝑓𝑢𝑛𝑐𝑡𝑖𝑜𝑛 (𝐸𝑆𝐹): derivative of ESF = LSF
1𝐷 𝐹𝑜𝑢𝑟𝑖𝑒𝑟 𝑡𝑟𝑎𝑛𝑠𝑓𝑜𝑟𝑚 𝑜𝑓 𝐿𝑆𝐹 𝑔𝑖𝑣𝑒𝑠 𝑡ℎ𝑒
𝑀𝑇𝐹 (𝑀𝑜𝑑𝑢𝑙𝑎𝑡𝑖𝑜𝑛 𝑇𝑟𝑎𝑛𝑠𝑓𝑒𝑟 𝐹𝑢𝑛𝑐𝑡𝑖𝑜𝑛)
Dynamic Range, refers to the range
between the minimum and maximum detectable signal levels. It indicates the ability of a system or device to capture and represent a wide range of signal values accurately.
Dark Current
Noise without input (illumination)
e.g. thermal fluctuations of el.-holes-pairs
Read-out Noise:
Errors during read-out procecure
dark current is dependent on ………., while read-out noise only depends on the ……….
dark current is dependent on exposure time, while read-out noise only depends on the read-out speed or ‘frame rate‘
‚Frame rate
Exposure time + read-out time
CCD (Charge-Coupled Device)
transition from film-based photography to digital imaging.
A CCD chip is a silicon-based semiconductor device that can convert incoming light into an electrical charge
Digital X-ray detector types
Conventional detectors
e.g. CCD, Flatpanel
Photon-counting detectors
Spectral detectors
CONVENTIONAL TECHNOLOGY FLOW
X-RAY
SCINTILLATOR (X-RAYS ARE CONVERTED TO LIGHT)
PHOTODIODE (LIGHT IS CONVERTED TO ELECTRICAL SIGNALS)
ELECTRIC CIRCUIT
PHOTON-COUNTING TECHNOLOGY flow
X-RAY
DIRECT CONVERSION (X-RAY IS CONVERTED TO
ELECTRICAL SIGNALS)
ELECTRIC CIRCUIT
PULSE HEIGHT ANALYZER
general definition of noise
any unwanted signal (image)
we will use the following definition of noise
random, uncorrelated signal (image)
Probability Density Function (PDF)
𝑃 (𝑎 < 𝑥 < 𝑏) = integral from 𝑎 to 𝑏 (𝑃(𝑥) 𝑑𝑥)
Expectation Value
𝐸 [𝑥 ] = integral from −∞ to∞ (𝑥 ∙ 𝑃(𝑥) 𝑑𝑥 ) = 𝜇
variance
𝑉𝑎𝑟[ 𝑥 ]= 𝐸 [ (𝑥 − 𝐸[ 𝑥 ]) ^2]
Skewness (symmetry)
𝑆 [𝑥] =𝐸 [(𝑥 − 𝐸[ 𝑥 ]) ^3]
kurtosis (peakedness)
higher central moments
𝐾 [𝑥] =𝐸 [(𝑥 − 𝐸[ 𝑥 ]) ^4]
Central Limit Theorem
sum of independent and identically distributed variables converge towards the normal distribution
Poisson distribution:
Probability mass function
expectation
variance
occurence
𝑃 (𝑥 = 𝑘) =(𝜆^k exp(−𝜆))/ 𝑘!
𝐸[ 𝑥 ]= 𝜆
𝑉𝑎𝑟 [𝑥 ]= 𝜆
𝑝ℎ𝑜𝑡𝑜𝑛 𝑐𝑜𝑢𝑛𝑡𝑖𝑛𝑔 𝑜𝑟 𝑒𝑚𝑎𝑖𝑙 𝑐𝑜𝑢𝑛𝑡𝑖𝑛𝑔
other distributions
Wrapped normal distribution
Rice distribution
Lorentz distribution
Gamma distribution
Detector noise (for CCDs)
− shot noise (photon statistics, Poisson)
− dark current (thermal electronic fluctuations in semiconductor, Poisson)
− readout noise (fluctuations during amplification and digitization, Gaussian)
dark frame measures ……,
bright frame measures …….
dark frame measures detector noise, hot pixels, dead pixels
bright frame measures gain differences and imperfections (dust, etc)
contrast
𝐶 = |𝑆 (𝑥2) − 𝑆(𝑥1)|
absolute difference of signal intensities in two different pixel locations
Visibility (Michelson contrast)
𝑉𝑖𝑠𝑖𝑏𝑖𝑙𝑖𝑡𝑦 = (|𝑆 (𝑥2) − 𝑆(𝑥1)|)/ (𝑆 (𝑥2) + 𝑆(𝑥1))
Contrast-to-noise (CNR)
𝐶𝑁𝑅 = |𝑆 (𝑥2) − 𝑆(𝑥1)|/ (sqrt(𝜎(𝑥)^2+ 𝜎(𝑥(1) ^2))
Noise power spectrum (NPS)
Power spectrum of pure noise image
𝑛 (𝑥, 𝑦) ⇔ 𝑁(𝑢, 𝑣)
𝑁𝑃𝑆 = 𝐸 [𝑁(𝑢, 𝑣)]^2
NPS Connection to auto-correlation
inv fourier of abs(N(u,v)^2) = autocorrelation
Weiner-Khinchin theorem
𝑎𝑢𝑡𝑜 𝑐𝑜𝑟𝑟𝑒𝑙𝑎𝑡𝑖𝑜𝑛 𝑜𝑓 𝑛𝑜𝑖𝑠𝑒 ⇒ 𝑛𝑜𝑖𝑠𝑒 𝑠𝑝𝑒𝑐𝑡𝑟𝑎𝑙 𝑑𝑒𝑛𝑠𝑖𝑡𝑦
red noise
white noise
blue noise
low freq high power
constant power
high freq high power
White noise in spatial domain means
white noise in frequency domain
White noise is perfectly
uncorrelated (Auto-correlation of white
noise is the delta function)
Noise reduction by averaging requirement
additive noise, zero mean
Denoising by linear filtering
Noise reduction
possible, but at
cost of sharpness (blurred)
Median filtering Good for
getting rid of ‘outliers’
Median filtering Less sensitive to
Less sensitive to outliers in pixel ensemble, better edge preservation
Bilateral filter similar to
Gaussian filter, which is a weighted average of
intensity of adjacent pixels with a decreasing weight with spatial distance
Bilateral filter takes into account the
variation of intensities to preserve edges
Bilateral Filter The kernel shape depends on
The kernel shape depends on the image content
Describe the way indirect detection works
X-Ray Photons
Scintillator
Visible Light Photons
Photodiode
Read-Out Circuit