Isolation

• For the biological subject, against electric shock from the amplifier’s power source(s)–or, how to not kill your patient while taking measurements!

• For the signal-to-noise ratio (SNR): isolation keeps the (60-Hz mains and other) noise out of the sensor pickup–and out of the amplifier input.

One simple way to isolate the input is to use an LED/phototransistor pair:

The sensor is connected to an LED, which outputs light of intensity proportional to signal voltage.

The light from the LED falls on a photodiode or phototransistor, which is biased in such a way that current only flows when light hits the device, and the current (and thus measured voltage) is proportional to light intensity.

Phototransistor output is to the amplifier.

The power circuitry for the amplifier is completely isolated from the sensor (electrodes etc.) Power supplies for the sensor use a transformer or battery, and the “ground” for the sensor side floats relative to the amplifier side (which should have ground connected to earth eventually.)

These devices are manufactured in one piece, so the frequency of light output from the LED is matched to the ideal absorption frequency of the phototransistor. However, there must be no electrical connection between the two sides of the circuit. More exactly, the impedance between the two circuits should be near a Teraohm ().

Single-Ended vs. Differential Input

• Referenced Single-Ended: two leads of sensor are + signal and earth (computer) ground. Note: to avoid ground loops, it is best to avoid connecting sensor earth to computer earth.

• Non-Referenced Single-Ended: two leads of sensor are + signal and – signal, which is connected through a bias resistor to earth ground.

• Differential: three sensor leads are available: + and – signal and (earth) ground.

The 4-BNC -to- DAQ-card lead you’ve probably used in a lab before is referenced single-ended: each of 4 sensors has available a signal wire, and the 4 sensors share a common ground wire. If the DAQ card is configured in Differential mode using this input lead, the signal gradually rises to a saturated level, as the DAQ card assumes a signal on the – differential input, which is in reality floating.

Problems with Single-Ended Input:

• Cross-talk: Because multiple signals share a ground wire, they also share some portion of the signal, so if only one input is connected that signal shows up, somewhat weakened, on the second input trace.

• Lack of isolation: using the ground wire as the negative signal lead means the subject and the amplifier are strongly coupled–and strong electric signals from the amplifier may travel to the subject. Also, stray noise (such as from the power mains) easily couples into the input.

Differential Input:

I’m talking a lot about DAQ cards, these are very important when working with A/D systems and analog and digital filters.  If you are not familiar with these, go do a little light reading.  Hopefully this is making enough sense to get the concepts across.  The DAQ card comes with a finite number (16, in our case) of analog input ports (channels), each with paired input pins: a signal (ACH#) and a ground (AIGND). All the analog input ground pins are tied together on the DAQ card. In differential mode, the analog input ports are paired, so channel 1 uses AICH0 as +, AICH8 as -, and the ground wires for the pair are tied together as the reference ground. Thus a 16-channel DAQ card has only 8 channels in differential mode.

• Because no channels share leads, cross-talk is reduced. The DAQ card has high common-mode rejection, so if the – input leads are tied to ground (simplest configuration), channels will not interfere because the cross-talk is common to both + and – inputs.

• The ground lead cannot be isolated from the input electrodes without causing the DAQ circuitry to saturate, so the DAQ card itself does not provide isolation.

Wiring a DAQ card in Differential Mode:

The figure is adapted from Fig. 4.6, p. 4-15, of the NI 6024E (DAQ card) User Manual, and illustrates the appropriate connections for differential input to the card. Recommended values of R+ and R- depend on the impedance Rs (in series with Vs) and coupling of the source signal Vs:

R+ and R- provide bias current return paths. Bias currents result from not-quite infinite input impedance to the DAQ, and if not balanced the noise they represent will not be common to both + and – inputs, and thus won’t be rejected. However, R+ and R- load down the source with an equivalent 2R+, which will decrease gain if R+, R- are too low. If R+, R- are too large, they will produce a DC offset at the DAQ input.

Bandwidth

Bandwidth is a critical parameter in determining the type of amplifier needed. Often a bioamp has adjustable bandwidth; typical applications have both low-pass and high-pass adjustable filters.

Typical Bandwidths of Biological Signals: (Table from Webster Fig 6.16, p. 259.)

(Electro-oculogram: signal from electrodes placed either side or above and below the eye. Linearly proportional to angle of gaze; DC signal)

Noise Reduction

Can be either or both analog and/or digital. If digital, it is done by post-processing collected data. If analog, wiring of the printed-circuit board, handling of wires on the bench, and internal circuitry are all details to consider carefully.

Protective Shielding

For equipment, against transient large signal sources–includes grounding the outer case of the equipment and surge protection, as well as isolation.

Power Supplies

AC (mains) with rectification and usually transformer isolation. Subject to power failures, risk of electrocution.
Battery: provides its own isolation. Limited lifetime and undesirable behavior just before failure.

Specialized Amplifiers

Lock-In Amplifier Block Diagram

Lock-in Amplifier: (www.lockin.de) If the signal source is mostly at a single frequency but is very weak and/or subject to a great deal of noise, a lock-in amplifier can be used to extract the signal. In biological cases the single-frequency property of the signal is most-often externally generated, by applying a single-frequency excitation in one way or another.

Theory:

Signal source S(t) = A cos(ω1t + θ1) + B cos(ω2t + θ2)

Reference signal R(t) = C cos(ω1t)

Product S(t)R(t) = AC cos(ω1t) cos(ω1t + θ1) + other terms in cos(ω1t) and cos(ω2t)

= AC/2 (cos(2ω1t + θ1) + cos(-θ1) ) + other terms in cos(ω1t) and cos(ω2t)

Integrating over an even number of cycles of the reference signal reduces all terms in ω1 to zero, so the displayed signal is proportional only to the amplitude of the component of the signal source at the reference frequency.

To get the reference frequency into the signal source various means are employed: one may have to excite the subject (nerve, membrane…) at the reference frequency, or one may already know the source has a dominant resonance; alternatively one may look at each of several spectral components in the source signal piecemeal, by tuning the reference frequency.

Pre-Amplifier:  high input impedance, moderate gain, high CMRR. Often differential input and isolation. May include DC offset control, gain control switches, and/or calibration signal.

Chopper-stabilized Amplifier:  removes thermal DC drift from a (very-low-frequency or DC) signal by using negative feedback and chopping the low-frequency signal at a frequency above the amplifier’s high-pass limit (effectively, this is frequency modulation). The signal can be reconstructed by demodulation after amplification; noise signals at both high frequencies and those below the chopper frequency are rejected.

This post was made using some old class notes, a DAQ card user manual and just some good old knowledge.  I know this is not a traditional engineersphere.com lesson, it is more of an ‘informative read’ for the avid electrical engineer interested in biomedical applications.  Enjoy

Amplifiers

Definition (for Bioinstrumentation): Circuit that makes a small signal, usually voltage but occasionally current or power, big enough to do something useful–including excite an output mechanism.

Common uses:

• Biological measurements of small signals
• Audio engineering: a large current is needed to drive speakers
• Wireless communications: far from originating antenna, signal is very weak and must be
amplified to be useful.

Secondary applications:

Many amplifiers are also filters, preferentially amplifying some frequencies over others

• Don’t want to amplify noise along with signal
• Only interested in low- or high-frequency portion of signal
• Active filter provides amplification as added bonus

General Amplifier Characteristics

Common Mode Rejection Ratio (CMRR)–ratio (usually in dB) of the amplifier’s common-mode gain to its differential-mode gain. Common-mode signals are input signals common to both + and – inputs and are usually unwanted noise–60-Hz, thermal, etc; differential signals are applied to only one input.

Gain–voltage out over voltage in, or current out over current in. May be given in dB.  Bioamp requirement: often adjustable; should be 1000 or greater, should be calibrated.

Input Impedance–what the input source sees as its load working into the amplifier: if the entire amplifier circuit were modelled as a resistor, what would be the value of the resistor? Typical bioamp: Rin = 10MΩ–signal source need not provide much current.

Output Impedance–same as input impedance but from the output end: model the entire amplifier as a source, and this is its internal impedance. Bioamp requirement: Ro << Rload.

Frequency Response–over what range of frequencies is the gain constant? Graphically illustrated with a Bode plot of gain vs. frequency.

DC offset–usually an amplifier has an operator-adjustable DC offset knob, to null out any offset associated with non-ideal amplifier or sensor behavior. A DC offset signal results in an incorrect reading unless removed or filtered out (high-pass filter).

Operational Amplifier:

basis for most instrumentation-related amplifiers, cheap, readily available, easy to work with. “Operational” = good for mathematical operations (+, -, log, …)

Meanings and advantages:

• equal input voltages –> within the limits of external power supplies, an op amp outputs whatever current is needed to drive the two input voltages equal. Result is that the output voltage follows the input, scaled by a large gain.
• infinite input resistance means the op amp never loads down the source, even if the source cannot supply much power.
• zero output resistance means the op amp is an ideal voltage source, with output voltage independent of whatever load impedance it must work into.
• infinite open-loop gain means the amplification properties of a circuit containing an op amp are independent of the op amp internal properties.

Carr and Brown go through several common op-amp configurations and show how to derive their voltage gains. Suffice it for now to know that if you want to build an amplifier, an op amp is a good place to start.

Common op-amp circuit configurations:

• Inverting and non-inverting amplifiers
• Summing and difference amplifiers
• Integrating and differentiating amplifiers
• Log and anti-log amplifiers
• Instrumentation amplifier
• Low-pass filter
• High-pass filter
• Band-pass and notch filters
• Buffer (voltage follower, or unity-gain buffer)

Op Amp Equivalent Circuit

This schematic illustrates the important properties of the op amp, and of any amplifier. It can also make it easier to understand circuit operation.

Note the open-circuit inputs– Rin = infinity. The
output voltage supply is a dependent voltage
source. Also, since the gain A is infinite, v2 – v1
must be zero to get a finite output.

Difference (Differential) Amplifier

Example: derive the gain relationship for the basic differential amplifier shown, assuming U1 is ideal and Vin = V2 – V1.

To get equal gain of both V1 and V2, set R2/R1 = R4/R3. Then Vo = R2/R1(V2-V1).

To get a high gain, R2 >> R1, but to get high input impedance R1 (and/or R3) should be large, making R2 and R4 even larger…Result: high gain and high input impedance are difficult to achieve together.

Instrumentation Amplifier

A difference amp with input buffer/gain stages to increase input impedance and gain. To analyze, realize that the same current must flow in R5, R6 and R5 (since no current flows into the op amps). Set R1=R3, R2 = R4; then Vo = G1* (v3(U1) -v2(U1)), where G1 = R2/R1 = gain of second (differential) stage.

Gain of input stage is 1 + 2*R5/R6 = G2. Overall gain is G1*G2. Making R6 a potentiometer allows compensation for inequalities in the two R5s, as well as for variable gain of the entire circuit.

Overall Gain: A practical difference amp can have a gain of 100, so it is not hard to get an overall gain of 10,000 from an instrumentation amp.
Input Impedance: equal to that of the op amps U1 and U2–very large. Use FET-based amps for extremely high input impedance
Output Impedance: close to that of the op amp U1–very small: the amp will provide whatever current is needed to maintain the output voltage regardless of load impedance.

Equal resistors: in practice one cannot buy matched discrete resistors; however it is fairly easy to manufacture them within an integrated circuit. Monolithic diff-amps are available.

Non-idealities of amplifiers

Gain:  TANSTAAFL–you cannot have gain without a power supply to provide it. Real gain is limited by the external power  supplies (+/- 12 or 15 V, for op amp circuits) Exceeding the limits of the power supply results in Saturation, or “hitting the rail”.

Output impedance: a zero output impedance means the circuit will provide whatever current is needed to maintain the requested output voltage. Practically, however, an op amp can only provide some 20mA, meaning RO is negligible only for RL>>15V/20mA = 750 Ω.

Frequency dependence: to avoid oscillation or saturation, circuitry must often be added that limits the bandwidth of an amplifier.
• To keep DC offset signals (from polarizing electrodes, for example) out of the amplifier, a high-pass filter is used to cut off DC (and lower-frequency ac) signals.
• If the load to be driven contains substantial capacitance, the current output limit again becomes a problem, limiting gain at high frequencies, where capacitors look like shorts.

Input bias current: real op amps do have non-zero input currents, which produce voltage drops at the input–another source of DC offset. This source can be minimized by using FET op amps.

Impedance Bridge

Often the measurand is the relation between voltage and current (one applied, the other a response) rather than a biologically generated source. An example in Carr and Brown uses a wire heated by an applied current as an airflow sensor:  air flow from a breathing patient cools the wire, changing its resistance. Similarly, a voltage applied to a membrane induces a current flow; the ratio of voltage to current is a resistance. Such relations are best measured using a Bridge, and if the bridge is made solely of resistors it is called a Wheatstone Bridge.

Usually drawn as a diamond, this configuration of resistors is “balanced” when V+ – V- = 0. If Rtest then varies a little, a differential amplifier across V+ and V- will register a potential difference proportional to
the change in Rtest.

The impedances can have capacitance and/or inductance associated with them, in which case the bridge can measure both energy storage and  resistive loss in an element.

A return path to ground for (DC) bias currents is automatically provided by this circuit to prevent saturation.

Well there you have it, a few common amplifier configurations and some useful terms pertaining to them.  Remember important concepts such as amplifier saturation, Input Impedance, Output Impedance, and Gain.  A solid understanding of these concepts is sure to impress somebody!  Amplifiers  part II will continue to elaborate on more fun amplifier concepts.

References: References: Carr and Brown ch. 7; Webster chs. 3, 6; Neamen, Electronic Circuit Analysis and
Design (McGraw Hill, 2001) ch. 9

Basic Image Modification

While image processing operations can be performed in hardware (e.g., in cases where speed is paramount), many image processing operations are performed in software. This section discusses some of the basic processing operations performed on different types of images, including image arithmetic, point operations, and geometric operations. Many of these operations are driven by the need to isolate regions of an image or to improve the appearance of an image.
Like more advanced image modification techniques, these basic operations are rarely used alone. Rather, the individual operations are cascaded together to produce an overall desired effect.

Image Arithmetic

Image arithmetic represents a broad category of algorithms that use pixel-based processes for combining or extracting image information. For example, if images P1 and P2 contain pixels that are referenced by a row integer i and column integer j, Table 1
illustrates some arithmetic operations that can be performed on these images. In the table, the symbol C represents a constant value by which an image can be adjusted or scaled. Note that both images must be the same size for these operations to apply. These operations are primarily used as sub-steps in more complex image processing operations, rather than being useful on their own.

Table 1. Expressions for some common pixel-based image arithmetic algorithms.

Figure 13. Example of an image that has been brightened by simple pixel-based addition

Boolean operations are pixel-based, logical operations performed on sets of images (see Figure 14). Note that these operations are usually applied to two-color images (the pixel values are either “0” or “1”). A common example of a boolean operation is the use of NAND to identify objects that have moved between images (AND will yield the intersection of two images, denoting the stationary objects).

Figure 14. Boolean operations performed on images

Point Operations

Point operations, like image arithmetic, are pixel-based. However, these operations perform a mapping between image pixel intensity values and their representations on the new image. The most common point operations are thresholding, contrast stretching, and histogram equalization.

Thresholding

When an algorithm thresholds an image, the pixels whose intensities are above and below the threshold are assigned binary values. This technique is often used as an image preprocessing step before the image is passed to other processing algorithms. An example of thresholding is shown in Figure 15. In this figure, a slice of brain tissue (containing nervous cells and glia cells) is thresholded and then connected-component labeled so that the number of cells in the image can be counted.

Figure 15. Illustration of how thresholding and connected-component labeling can be used for cell counting applications [http://www.dai.ed.ac.uk/HIPR2/threshld.htm].

Histogram Equalization

Histogram equalization reassigns the intensity values of pixels in the input image so that the output image contains a uniform distribution of intensities (i.e., a flat histogram). This is illustrated in Figure 16 (compare this figure to the image/histogram in Figure 10).

Figure 16. An example of histogram equalization for magnifying the level of detail inportions of a gray-scale image.

The next post, Biomedical Image Processing – V, will discuss contrast stretching, Image Denoising and Enhancement, and digital filters for noise reduction.  References for this post are listed in Biomedical Image Processing – I.

DICOM Image Standard

Medical specialists have been slow to adopt widely accepted standards for image/film storage, display, and transmission. However, one standard has been adopted with reasonable success in the radiology community: DICOM (Digital Imaging and Communication in Medicine) has been progressively developed since 1983 by ACR/NEMA (America College of Radiologists and the National Electrical Manufacturers’ Association). DICOM defines a standard for the exchange and storage of medical images from various imaging modalities, including MRI, CT, and ultrasound. It focuses on the communication interface between a host computer and the scanner, but it also defines file format standards to which a system must adhere in order to be considered DICOM-compliant. Some examples of DICOM images are depicted in Figure 7.

Good starter links:

http://www.scispy.com/Standards/DICOM.html

ftp://ftp.philips.com/pub/ms/dicom/DICOM_Information/CookBook.pdf

Figure 7. Radiological images stored in the DICOM standard image format [Vepro
Computersysteme GmbH, Cardio Viewing Station, Version 4.41].

Image Analysis

Image analysis encapsulates a set of basic image processing operations whose purpose is to query, but not alter, an image. These operations include

• intensity histogram generation,
• information classification, and
• connected components labeling.

Intensity Histogram

An intensity histogram describes the distribution of pixel intensity information within an image. This information is displayed as the number of counts associated with each intensity level (see Figure 9). Histograms can also be ascertained for color images, where distributions are displayed for the individual red/green/blue values.

Figure 9. Depiction of a histogram for an 8-bit gray scale image (256 intensity levels).

Figure 10. Histogram: 256-level gray scale image (X-ray of a human abdomen)
[http://www.medphys.ucl.ac.uk/research/borg/research/NIR_topics/imaging_exp.htm].

Classification

Image classification includes a broad set of algorithms for identifying portions of an image that are related to one another. This is very closely related to segmentation, which physically separates these regions from other regions. For example, in a fluorescence image of cells, a researcher might be interested in an image processing algorithm that localizes cell nuclei (see Figure 11) for counting purposes.

Figure 11. Results of an algorithm that attempts to classify and segment portions of a
fluorescence image that correspond to cell nuclei.

Connected Components Labeling

Connected components labeling is a process whereby an algorithm scans an image and groups sets of pixels based on common features (such as pixel intensity). Once these pixel elements are grouped, they are all assigned the same value and labeled as a region. This process is illustrated in Figure 12. Connected component labeling is different from classification in that the algorithms make no judgments as to which components exhibit similar properties.

Figure 12. Example of component labeling based on nearest neighbor analysis
[http://www.dai.ed.ac.uk/HIPR2/label.htm].

Image Properties

Once an image is stored in digital format, it can be described by a number of different parameters. Some of the relevant parameters are briefly discussed here. The traditional convention for an image coordinate system is depicted in Figure 4.

Figure 4. General convention for image coordinate systems.

While biomedical images are generally viewed in 2D, it is sometimes helpful to view gray-scale (monochrome) images in perspective, with the third axis being brightness. This is illustrated with “pseudoimage” data in Figure 5.

Figure 5. Pseudoimage data viewed in 2D and perspective modes. Data values (either a “0″ or a “2″) are placed within a 31-row by 24-column matrix.

Images can be described by a large number of different parameters. Some of these are listed here.

• Pixels (a.k.a. picture elements, pels, image elements) – Individual rectangular
  elements that comprise an image. The term voxel describes a pixel’s 3D analog.
• Gray levels – An 8-bit, gray-scale image with 1024 × 1024 pixels requires a
  megabyte of storage.
• Color depth – Usually reported as powers of 2, this can range from 2 colors up to 32-
  bit color (a.k.a. True Color). Colors are often defined using RGB (red-green-blue) or
  HSB (hue-saturation-brightness) combinations (see Figure 6).
• Aspect ratio – The scaling ratio between the x and y axes.
• Contrast – The relationship between the brightest and dimmest pixel in the image.
• Histogram – A binned representation of the gray levels, colors, or brightness levels
  in the image.

Other notes:

• Because of the optimization properties of 2D Fourier transforms, it is often
  advantageous to select image sizes whose number of rows and columns are both
  powers of 2 (e.g., 512 x 512 array with 128 gray levels – comparable to a
  monochrome TV image)
• In motion videos, images are displayed at a rate of 30 frames per second

RGB

Primary colors: red, green, blue
Secondary colors: yellow = red + green, cyan = green +
blue, magenta = blue + red.

White = red + green + blue

Black = no light.
[http://www.cecs.csulb.edu/~jewett/colors/rgb.html]

HUE: actual color
Measured in angular degrees around the cone starting and ending at red = 0 or 360 (so yellow = 60, green = 120, etc.).

SATURATION: purity of the color
Measured in percent from the center of the cone (0) to the surface (100). At 0% saturation, hue is meaningless

BRIGHTNESS: measured in percent from black (0) to white (100). At 0% brightness, both hue and saturation are meaningless.
[http://www.cecs.csulb.edu/~jewett/colors/hsb.html]

Figure 6. RGB and HSB color descriptions.

The next Biomedical Image Processing lesson will discuss image analysis, classification, and component labeling.  Happy Holidays!

-Jeff

• Image creation: generation of digital images and their resulting properties.
• Image modification: operations that can be performed on digital images including denoising, enhancement, and compression.
• Image analysis: techniques for extraction of biomedical data from images, including feature recognition.

The issues and techniques discussed here are general: they apply to images acquired using various methodologies, including X-ray radiography, X-ray computed tomography, MRI, ultrasound, and PET. Note that while image processing can be performed on
analog images (e.g., by way of optical Fourier techniques), this overview concentrates on the processing of digital biomedical images.

Digital Image Creation

Image Generation

Analog images, having continuously varying color or gray-scale representations, can be stored as digital images for use in computer-based systems. The rules that apply to the digitization of one-dimensional (1D) signals (e.g., voltage versus time) also apply to the digitization of spatial images (e.g., gray level versus position): the images must be sampled and quantized with enough fidelity that they truly represent their analog image counterparts. This, of course, depends on the application for which the images will be utilized. Scanning, the initial step in the image digitization process, includes the division of the picture into a number of small regions called picture elements, or pixels (see Figure 1).

The scanning operation represents the image by a grid consisting of m rows × n columns. For example, in the image shown in Figure 1, m = 288 and n = 432, for a total of 124,416 pixels. Each pixel is addressed by its (m,n) coordinate within the matrix. The scanning operation is accompanied by a sampling operation, which transforms the representative brightness of each pixel into an analog voltage level or other measurable signal. In the case of an X-ray radiograph, this operation might be performed by a photomultiplier tube.
A good example of a common line scan sensor is a flatbed scanner, which uses an array of discrete silicon imaging elements, called photosites, to produce voltage output signals that are proportional to the intensity of the incoming light at each linear location. These solid state devices can be shuttered at high speeds (e.g., 1/10,000 of a second), and their precision typically ranges from 256 to 4096 elements [Gonzalez]. Solid state area sensors also exist, containing grid sizes from 32 × 32 pixels up to 1280 × 1024 pixels.
Finally, each pixel in the grid is assigned an integer that corresponds to its brightness level. This quantization operation is similar to the quantization performed on 1D signal data. For example, in an 8-bit gray-scale image, the integer 0 may represent black, the integer 255 may represent white, and the integers in between will represent various gray scales that correspond to image intensity. The general convention in gray-scale images is that larger integers represent brighter pixel values (or pixels exhibiting greater signal
strength).

Figure 1. Eight-bit, gray-scale digital image consisting of individual square pixels, or
picture elements.

As illustrated in Figure 2, two-dimensional (2D) spatial images must obey sampling constraints that are similar to those imposed on 1D, time-based signals. When 1D signals are digitized with an analog-to-digital (A/D) converter, they must be sampled at a frequency greater than or equal to twice the highest frequency component in the signal. Likewise, 2D images must be sampled at a spatial frequency greater than or equal to twice the highest spatial frequency component in the image. Otherwise, these undersampled images will display the type of aliasing depicted in Figure 2. Note that to redisplay a digitally stored image, a D/A conversion is often required (e.g., for a cathode ray tube in a television set).

Figure 2. Simple illustration of aliasing in a digital image line [after Seeram, p. 55].

Some images are not produced on native rectangular grids. B-mode ultrasound images, for example, are often reconstructed from fan-shaped beams or radially-distributed ultrasonic reflectance data (see Figure 3). Production of a 2D image requires that these radial line data be interpolated. These interpolation schemes can affect the quality of the overall images, especially if the individual line scans are separated by large angles.

Figure 3. Examples of rectangular images constructed from non-rectangular imaging
modalities [Fetal Image: http://www.biosound.com/image5.html; OCT/IVUS: CLEO
’96 Proceedings, p. 55]

Why Digitize an Image?

The reasons for digitizing 2D image data are very similar to the reasons for digitizing 1D data, revolving primarily around (1) the ability to transmit image data “noise free” and (2) the potential for processing these images with computational tools. These tools
provide assistance in the following areas:

• Image Enhancement/Restoration: image improvement, possibly through the
  reduction or removal of noise, artifact, or unnecessarily fine detail (e.g.,
  processing (1) degraded images of unrecoverable objects or (2) images from
  experiments that are too expensive to duplicate)
• Image Analysis: extraction of information with/without interpretation
• Pattern Recognition: structures and patterns are “seen” and recognized
• Image Detection: identification of certain shapes, contours, or textures while
  disregarding other image features
• Geometric Transformation: rotation and/or scaling of the image
• Data Compression: Reduction in image size for storage or transmission

Part II of Biomedical Image Processing will discuss image properties and DICOM Image Standards.  This information was compiled by Steve Warren of Kansas State University using the following references:

[1] Seeram, Euclid. Computed Tomography: Physical Principles, Clinical
Applications, and Quality Control, W.B. Saunders, Philadelphia, © 1994, ISBN 0-
7216-6710-4
[2] Shung, K. Kirk, Michael B. Smith, and Benjamin Tsui. Principles of Medical
Imaging, Academic Press, San Diego, © 1992, ISBN 0-12-640970-6
[3] Rosenfeld, Azriel and Avinish C. Kak. Digital Picture Processing, Second Edition,
Volume 1, Academic Press, San Diego, ©1982, ISBN0-12-597301-2.
[4] Rosenfeld, Azriel and Avinish C. Kak. Digital Picture Processing, Second Edition,
Volume 2, Academic Press, San Diego, ©1982, ISBN 0-12-597301-2.
[5] Gonzalez, Rafael C. and Richard E. Woods. Digital Image Processing, Addison-
Wesley, Reading, MA, ©1993, ISBN 0-201-50803-6.
[6] Teuber, Jan. Digital Image Processing, Prentice Hall, New York, ©1989, ISBN 0-
13-213364-4.

The majority of physical processes are affected, to one degree or another, by temperature.  Whether these changes are exploited as a technique for temperature transduction or used to identify temperature induced artifacts, the measurement of temperature is of fundamental importance.  Some physical effects used as the basis for temperature transduction are listed below:

1. Thermal expansion
2. Thermocapacitive effect
3. Thermochemical effect
4. Thermoelectric effect
5. Thermoresistive effect
6. Pyroelectric effect
7. Radiation effect

Thermal expansion is the basis for function of mercury or alcohol thermometers while thermocouples employ the thermoelectric effect.  Like the resistance temperature detector (RTD), the thermistor is also a temperature sensistive resistor.  Thermistor is an acronym for thermally sensitive resistors.  While the thermocouple is the most versatile temperature transducer and the Platinum RTD is the most table, the word that best describes the thermistor is sensitive.

Of the three major categories of sensors, the thermistor exhibits by far the largest parameter change with temperature.   Thermistors are generally composed of semiconductor materials.  Although positive temperature coefficient units are available, most thermistors have a negative temperature coefficient (TC); that is, their resistance decreases with increasing temperature.  The negative T.C. can be as large as several percent per degree Celsius, allowing the thermistor circuit to detect minute changes in temperature, which could not be observed with an RTD or thermocouple circuit.  The price we pay for this increased sensitivity is loss of linearity.  The thermistor is an extremely non-linear device that is highly dependent upon process parameters.  Consequently, manufacturers have not standardized thermistor curves to the extent that RTD and thermocouple curves have been standardized.

An individual thermistor curve can be very closely approximated through the use of the Steinhart-Hart Equation:

Here,

= temperature in degrees kelvin for i = 1,…,n (number of data points),

= resistance of the thermistor (ohms) for i = 1,…,n, and

A, B, C = Curve-fitting coefficients.

The choice of temperature transduction technique depends on a variety of factors, including temperature range, cost, size, weight, power consumption, sensitivity, aging effects, etc.  Another prime consideration that affects a choice is the range over which measurements are to be made and the precision of the readings.  Other considerations that may not be so obvious include selection of a thermal mass to be compatible with the bandwidth requirements of the anticipated temperature variations.