BME 311 Lecture 2

imaging depth in the body

-confocal microscopy: 100um Resolution: 1um

-optical coherence tomography (OCT): 1 mm Resolution: 2-10um

-ultrasound: 1 cm Resolution: 150um

-high-res computed tomography (CT): 10cm Resolution: 300um

-magnetic resonance imaging: entire body Resolution: 1 mm

what is spatial resolution

-the size in the image of a “point” source of contrast

-PSF = point spread function

-measured in each dimension, can differ in x,y,z

spatial resolution provides the measurement of

how well the imaging system can distinguish small objects

imaging volume is discretized into

-pixels (2D), often square but not always

-voxels (3D), often cubic (but often called “pixel” too)

smaller resolution value is better

and we call this higher resolution

sometimes defined by pixel size, sometimes larger than pixel size

-pixel size closely related to resolution (MRI)

-pixel size is arbitrary in PET/SPECT (smaller than resolution)

-pixel size chosen should not cause loss of info in the data, but not too many to waste memory

resampling

changing voxel size and/or alignment

-done to save disk space of match with other images

-simple resampling to finer grid can create impression of better resolution

-simplistic resampling (eg. nearest neighbor) by non-integer factors can create artifacts

-easiest using integer factors = 2x, 3x

-best using interpolation (bilinear or other)

misinterpreting shape-resolution

pixel size or image appearance does not always reflect resolution

discretization issues

artifact generation: Nearest-neighbor resampling with non-integer scaling creates artificial patterns

partial volume effect

-voxel/pixel contains >1 type of tissue or signal value

-so pixel has intermediate value

what is temporal (time) resolution

-how fast an image can be taken

-how fast the imaging system creates an image

-how fast a change in the object can be detected

time resolution subtleties: meaning on multiple levels

-detection of raw signal itself (<1 nanosecond for PET)

-creation if image (variable for PET, defined by pulse sequence for MRI and ultrasound)

-acquisition of enough data to answer clinical/research question - better quality takes more time

-varies even within modality depending on pulse sequence, radiotracer

time resolution subtleties: changes during acquisition are averaged or worse

-patient motion in PET causes blurring but in MRI can cause major artifacts

-biological variation is typically averaged

sensitivity

-change in image/change in actual parameter

-higher sensitivity allows visualization of smaller differences

-for a given noise level, better contrast

-specific definition varies depending on imaging modality

what is contrast

-ability to see differences between different signals

-subject contrast (biology/chemistry)

-detector contrast (computer)

-affected by both background and noise

effects of noise

lose details

computers in bioimaging

originally: designed to perform mathematical computations and data processing quickly

now: information display, storage, networking-info transfer and communication

file size

number of voxels x memory per voxel + metadata

bit- 0 or 1

bytes-8 bits»256 different values

kilobyte(kB)-1024 bytes

megabyte(MB)-1024 kB

gigabyte(GB)-1024 MB

terabyte(TB)-1024 GB

data type of voxel value is important

-integer can be 2/4/8 bytes

-integers can be signed or unsigned

-2 bytes=65,537 values

-float can be 4/8 bytes, these are decimal numbers with exponent, and there are different formats (bit/little endian)

-no 2 byte float-sometimes float data is stored as integer*2 with an overall multiplier and offset

analog and digital data

all data originates as analog signal, must be digitized

analog transmission

-all detector signal start out as analog, but distortions possible

-inaccuracies when signal amplifies

-attenuation losses along signal wires

-electronic noise

digital transmission advantage

-immune to accumulation of errors due to signal distortion

-distortions do not have sifficient amplitude to cause 0 to be mistaken as 1

-most digital circuits do not amplify incoming data, make a fresh copy of it

-other safeguards to main information integrity-can have additional redundant information to detect errors

-eg. parity bits-additonal bit with each byte which changes if any bit changes

conversion of continuous analog data to digital (ADCs)

-only certain time points of the continuous analog signal converted=sampling

-conversion to digital=digitization or quantization

characteristics of ADC: sampling rate-number of times per second sample and digitize signal

nyquist theorem

to represent a signal of frequency, you need to sample at 2f. ex. if the sampling interval deltat is 2ms (500 Hz) then the highest frequency you can represent is fN=1/(2×0.002)=250Hz

also applies to spatial information=signal changing over distance

to display digital images from imaging modality onto video monitor

digital-to-analog converter (DAC)

disadvantages of digitization

-loss if information due to both sampling and quantization

-values of analog signal between sampling times is lost

-minimum sampling rate needed-nyquist limit

-analog signal continuous value-digital signal finite values

-N-bit signal restricted to 2^N values

storage of digital images

-digital image, usually square or rectangular array of numbers -a matrix

-pixel size=field of view (FOV) of the image/number of pixels across the image

eg. if CT has 23 cm x 23 cm FOV and images aquired in 512 pixel by 512 pixel format, pixel size is 23cm/512=0.45 mm in x and y directions

digital contrast resolution

-largest number stored in single pixel is determined by number of bits or bytes used for each pixel (i.e. gray levels)

-1 byte used = maximal number is 2⁸-1=255(8 bits)

-amount of contrast resolution provided by an imaging modality determines the number of bits per pixel needed

-ultrasound limited contrast resolution-6 or 7 bits enough - 8 bits used

-x ray CT high contrast resolution -12 bits required to represent full range of CT numbers

visual interpretation of image data

-image data is just numbers, needs to be converted to visual representation

-the way this conversion is done affects observed contrast and ability to diagnose

-human eye can detect only a limited range of brightness

-select mapping that optimizes display of important features-contrast enhancement

how?

-translation table selection

-windowing

translation table

-lookup table produces digital value indicating video intensity to the DAC

-video intensity can also be represented as different colors

windowing

-can use entire range of intensity to display portion of total range of pixel values

-windowing can be combined in translation table

processing of digital images

-image reconstruction to get initial image from raw data

-many “post-processing” techniques to apply to images

-post-processing of image generally does not unequivocally improve image quality-eg cannot resolve objects less than limits of the modality

-but can make useful tradeoffs in image quality features or combine

common processing techniques

-addition or subtraction of digital images

• images need to be in the same format, pixel size, etc

• resultant image is in the same format

• image subtraction common in angiography to remove effects of anatomic structures

spacial filtering=smoothing

-statistical nature of acquisition process can give grainy appearance-noise

-reduced by spatial filtering operation=smoothing

-each value of pixel in smoothed image is weighted average of the corresponding pixel in the unprocessed image with its neighbors

-reduces noise but also worsens spatial resolution-tradeoff but sometimes improves things like logic detectibility

quantification (quantitation)

-physical and biological meanings

-physics-how well absolute voxel value represents true value-often estimated with phantoms-can vary within each modality

-biology-values derived from raw images that represent functional processes (glucose metabolic rate)-sometimes measure reliability with test-retest” study

registration (coregistration, realignment, reslicing, spatial normalization)

-aligning 2 images of the same region of body

-equalize pixel sizes and rotate/shoft/scale

-”rigid body” if images are of same person and organ shape does not change (eg brain)

-”non rigid”, “deformable”,”warping” etc if not (etc to template)

translational methods

-readily allow transition from lab to clinic

-eg PET tracers tested in animals, then humans

for single parameter, p-value is used

-assuming groups do not differ (“null hypothesis" is true), this is the probability that your result is by chancw

MRI

-usually good resolution

-anatomy is obvious

-good contrast between soft tissue types

-low signal in cortical bone (like skull)

-usually gray scale

CT

-great resolution

-bone signal very strong, soft tissue weaker with poor contrast between soft tissue types (unless contrast agent used eg in blood)

-usually gray scale

PET/nuclear

-anatomy can be hard to discern

-resolution low, noisy appearance

-be careful-often fised (overlaid) over CT or MRI for anatomy

-usually colored scale