Computerized Densitometry
and Color Coding of [C]
Deoxyglucose Autoradiographs

Charles Goochee, BS, Wayne Rasband, MS, and Louis Sokoloff, MD

 

 

A computerized image processing system has been developed for quantitative analyses of autoradiographs obtained
with the ['*C]deoxyglucose method. The system is composed of standard, commercially available components and
includes a scanning microdensitometer, computer, image memory and display system, and monochrome and color
monitors. The associated computer programs are written in PASCAL. Autoradiographs are automatically scanned,
and the optical density of each spot is digitized at a maximum resolution of 65,536 readings per 6.4 x 6.4 mm area
and stored in memory. Images can be reconstructed from the data in memory, displayed on the monitors, and
utilized for microdensitometric analyses or manipulated for image enhancement, enlargement, or weighted av-
eraging of selected regions. The digitized data can also be utilized to solve the operational equation of the
['*C]deoxyglucose method, and color-coded images of autoradiographs can be reconstructed so that each color
represents a narrow range of the rate of glucose utilization. By means of this system, it is possible to generate
quantitative metabolic maps that display the distribution of actual rates of local glucose utilization throughout the

entire central nervous system in regions as small as 100 jm or less.

Goochee C, Rasband W, Sokoloff L: Computerized densitometry and color coding of ['*C]deoxyglucose

autoradiographs. Ann Neurol 7:359-—370, 1980

 

 

A recent useful application of quantitative au-
toradiography has been the radioactive deoxyglucose
method for determination of local cerebral metabolic
rate [19]. This method, which employs tracer
amounts of 2-deoxy-D-[“C]glucose ({*C]DG) to
trace glucose metabolism through the hexokinase-
catalyzed phosphorylation step, makes possible
quantitative determination of the rates of glucose
utilization simultaneously in all the macroscopic
structural and functional components of the central
nervous system in laboratory animals in either nor-
mal conscious of altered physiological states. The
quantification of local cerebral glucose utilization is
based on a kinetic model that incorporates the known
biochemical properties of 2-deoxyglucose in brain
tissue, the kinetics of the exchange of 2-de-
oxyglucose and glucose between plasma and brain,
and the kinetics of their phosphorylation by
hexokinase. Localization is achieved by measurement
of tissue concentrations of carbon 14 in specific re-
gions of brain by a quantitative autoradiographic
technique. The method has been applied to a variety
of experimental conditions in the rat [4, 18, 19] and
monkey [6, 8, 9], and studies are in progress to adapt
it to the dog (Duffy T: unpublished data, 1979) and

cat (Magnes J, Kennedy C, Miyaoka M, et al: unpub-
lished data, 1979).

The amount of information stored in autoradio-
graphs of brain obtained with the ['*C]DG method is
immense. The method is presently capable of mea-
suring glucose utilization in regions as small as 100
pm. This level of resolution and the great heteroge-
neity of cerebral metabolic activity throughout the
brain are clearly visible in autoradiographs. Because
of the enormous complexity, it has been impractica-
ble to recapture all the information present in au-
toradiographs, combine it with the quantitative po-
tential of the [C]DG method, and reconstruct
highly detailed, spatially precise, quantitatively cor-
rect metabolic maps demonstrating the actual rates
of glucose utilization throughout the tissues of the
brain. Because of the practical limits of conven-
tional manual densitometry, it has been necessary
to subdivide the brain arbitrarily into discrete regions
of metabolic activity that correspond more or less to
traditional neuroanatomically defined entities and to
determine an average value of glucose utilization
based on an arbitrary finite number of densitometric
measurements in each of these regions. In view of the
great heterogeneity of metabolic activity within even

 

From the Laboratory of Cerebral Metabolism and the Research
Services Branch, National Institute of Mental Health, Bethesda,
MD.

Reprinted from Annals of Neurology, Vol 7, No 4, April 1980. Published by Lite, Brown and Company, Boston, MA. Copyright © 1980 by the American Neurological A

Accepted for publication Sept 5, 1979.

Address reprint requests to Mr Goochee, NIMH, 9000 Rockville
Pike, Bldg 36, Room 1A-27, Bethesda, MD 20205.

359
; all rights

Ne part of this repring may be repeidueed!in nay emn or by any cheated or sarchenied ince iehading worrnatcs meme. an retried poeta with te poblcher? etane neniace

d
well-defined anatomical structures, an inordinate
number of densitometric readings has often been
necessary to avoid sampling errors and to obtain a
reliable estimate of the properly weighted mean
value for metabolic rate within each structure. Data
so obtained have generally been presented in large
tables consisting of long but limited lists of named
structures and their corresponding rates of glucose
utilization. Such presentation of data ignores or
obscures much of the information present in the au-
toradiographs, e.g., information about structures not
included in the lists or about heterogeneity within
each of the structures.

The present report describes a system for
computer-assisted image processing of autoradio-
graphs that greatly expands the analytical scope of the
{4C]DG method. This system permits quantitative
presentation of local rates of glucose utilization in the
pictorial form of autoradiographs. Through color
coding, the autoradiographs are transformed into
representations of actual rates of local glucose utili-
zation, and the heterogeneity and complexity in the
distribution of local metabolic activities are fully re-
tained and visualized. Furthermore, metabolic maps
from different autoradiographs and from different
animals can be normalized through the assignment of
designated colors to specific rates of glucose utiliza-
tion, thus permitting direct visual but quantitative
comparison of autoradiographs from different ex-
periments. Computer programs have also been de-
veloped which permit use of this system for high-
resolution densitometry. It is presently possible to
determine the rates of glucose utilization in struc-
tures as small as approximately 100 ym in width, and
greater resolution may be possible in the near future.
Other programming features permit determination
of the average rate of glucose utilization for the brain
as a whole, properly weighted for the relative masses
of all its component parts.

Methods
Basic Principles of the [*C\DG Method

Before the design and capabilities of the computerized
system are discussed, a brief summary of the [“C]DG
method will be presented. A detailed description of the
theory and procedures of this method has recently been
published [19].

The deoxyglucose method is based on accumulation of
2-deoxy-D-['*C]glucose-6-phosphate in tissue following an
intravenous pulse of 2-deoxy-D-[“C]glucose. Deoxyglu-
cose is transported across the blood-brain barrier by the
same carrier that transports glucose, and it competes with
glucose for hexokinase, the enzyme which phosphorylates
both to their respective phosphates. The rate of deoxyglu-
cose phosphorylation is quantitatively related to the rate of
glucose phosphorylation, depending on their relative con-
centrations in the precursor pools and the kinetic prop-

360 Annals of Neurology Vol 7 No 4 April 1980

erties of hexokinase with respect to the two substrates. In a
steady state of glucose metabolism, the net rate of glucose
phosphorylation equals the rate of glucose utilization. On
the basis of these kinetic principles, an operational equa-
tion has been derived that expresses the rate of glucose
utilization in terms of defined constants and measurable
variables, e.g., the final tissue concentration of '*C, the time
courses of the arterial plasma glucose and [*C]DG con-
centrations, the rate constants for the transport of [*C]DG
across the blood-brain barrier and its phosphorylation, and
a combination of constants, the so-called lumped constant,
consisting of the ratios of the distribution volumes and the
Michaelis-Menten kinetic constants of hexokinase for
deoxyglucose and glucose [19].

The experimental period is initiated by an intravenous
pulse of ['*C]DG. Timed arterial blood samples are with-
drawn during the ensuing period, and the plasma is
analyzed for ['*C]DG and glucose concentrations. After 45
minutes the animal is decapitated, and the brain is removed
and frozen in Freon XII chilled to —60°C with liquid nitro-
gen.

Brain sections 20 wm thick are prepared with a cryostat
(American Optical Co, Buffalo, NY) maintained at —21° to
—22°C. The brain sections are picked up on a cover glass,
dried on a hot plate at 60° to 70°C, and then placed se-
quentially in an x-ray cassette with Kodak SB-5 x-ray film
(Eastman Kodak Co, Rochester, NY). Exposure time is
generally five to six days. Calibrated ['*C]methyl methac-
rylate standards are included with the brain sections during
autoradiographic exposure; these are used to obtain a cali-
bration curve for each film for conversion from optical den-
sity to tissue ‘'C concentration.

The Computer-assisted Image Processing System

The computerized system developed to process the au-
toradiographic data includes a rotating-drum scanning den-
sitometer, a computer, a disk storage system, an image
display system, monochrome and color monitors, joystick
controls, a video terminal, and a line printer (Fig 1).

The rotating-drum scanning densitometer (Model P-
1000, Optronics International, Chelmsford, MA) is used to
convert the photometric data from the autoradiographs
into digital form. The film containing the autoradiographic
images of the brain sections and the C plastic standards is
positioned on the scanner drum. Under computer control a
digitizing scan can be performed on any area of the film
from a maximum size of 23 x 23 cm toa minimum of 6.4 X
6.4 mm. Aperture sizes of 25, 50, and 100 um are possible
for both the incident and collecting objectives. At its high-
est resolution, 40 lines per millimeter, the scanner permits
access of 8,000 x 8,000 (i.e., 64,000,000) distinct data
points per 20 < 20 cm of film area (1,600 data points/mm?).
The currently employed image processing system, how-
ever, is capable of presenting simultaneously only a 256 x
256 array of data points, i.e., 65,536 points. If the area of
the film to be scanned contains more than 256 x 256 po-
tential readings, then readings are selected from the area at
equal intervals to reduce the total number to the 65,536
permissible limit.

The scanner provides discrimination of 256 density
levels (8 binary digits, or bits, per reading) over the optical
 

DISK
STORAGE

 

 

 

 

 

 

SCANNING
DENSITOMETER

COMPUTER

 

 

 

 

 

 

 

VIDEO
TERMINAL

 

 

 

 

SssSS=3 i 1

 

[A/D CONVERTER |

IMAGE DISPLAY MEMORY |

 

 

t

 

 

 

 

 

 

 

|

|

| |

JOY STICK | TRANSFORM
CONTROLS | TION TABL |

| |

; [MonocHRomE| [ coLor | |

| | MONITOR MONITOR | |

 

 

 

IMAGE DISPLAY SYSTEM

Fig 1. Components of the computer-assisted image processing
system. The identities and sources of all components are
Specified in the text.

density ranges of 0 to 2 or 0 to 3. According to the man-
ufacturer’s specifications, density readings are reproducible
to 2 parts in 256, and positional accuracy is rated at +2 um
per centimeter. The time required for a complete 256 x
256 scan ranges from 30 seconds to 2 minutes.

The 256 X 256 array of optical density readings is pro-
cessed through a digital computer (Model PDP 11/34,
Digital Equipment Corp, Maynard, MA) and is stored in
the image memory of the image display system (Model
ID-2000 Image-Display System, DeAnza Systems, San
Jose, CA). The image display system includes a mono-
chrome image transformation table which serves to convert
the digital data stored in the image display memory into
electrical signals that determine the intensity of the beam
on the screen of the cathode ray tube of a monochrome
monitor, and also a color intensity transformation table
which controls the conversion of data into the combination
of red, green, and blue signals necessary to create the ap-
propriate color image on the face of the color monitor. The
monitors are continually refreshed from the image display
memory via the transformation tables at a frequency of
30 times per second, thus creating a 256 X 256 element
display on a Model 8025 high-resolution, 19 inch color
monitor (Aydin Controls, Fort Washington, PA) and a
Model BH high-resolution 13 inch monochrome monitor
(Ball/Miratel, St. Paul, MN). The image display memory
provides 12 bits for each of the 256 x 256 data points, the
lower 8 containing the value for optical density, the upper
4 bits providing the opportunity for information coding,
graphic overlays, area outlines, and other functions.

A random-access disk storage system (Model PM-DS/
11B, 5.0 Mbyte Cartridge Disk System, Plessey Microsys-
tems, Irvine, CA) permits storage and later recall of any
256 X 256 display. Approximately eight seconds is re-
quired for either operation.

Primary user—system interaction is accomplished
through a video terminal (Model VT52 terminal, Digital
Equipment Corp, Maynard MA). Two joysticks (i.e., con-
trol levers that serve as manual controls for bidimensional
potentiometers; Model 525 x-y Potentiometer Joy Sticks,
Measurement Systems, Norwalk, CT) and the requisite
analog-to-digital (A/D) converter (Model AR11 Analog
Real-time Subsystem, Digital Equipment Corp, Maynard,
MA) permit additional user interaction with the image pro-
cessing system. For example, the joysticks can be used to
generate and adjust the size and position of a rectangle
around a region of the display for more detailed analyses of
the data within it.

Software developed to coordinate this system was writ-
ten in PASCAL, a computer language which greatly facili-
tates the construction of large computer programs (OMSI
PASCAL-1 compiler, Oregon Minicomputer Software,
Portland, OR) [5, 7].

Results

Computer-assisted Image Enhancement

Image enhancement refers to manipulation of an
image to present the observer with additional infor-
mation which was less obvious in the preenhanced
image. A number of image enhancement techniques
have been reported [1, 2]. Four of these tech-
niques—contrast stretching, intensity-window slic-
ing, pseudocolor coding, and digital zooming—
have been adapted for our purposes.

MONOCHROME IMAGE ENHANCEMENT. Contrast
stretching and intensity-window slicing are methods
for improving the appearance of monochrome im-
ages. These techniques affect the way in which the
optical density readings stored in the image display
memory are converted to shades of gray in the image
presented on the monochrome CRT monitor. The
“gamma” function for a monochrome image display

Goochee et al: Image Processing of Autoradiographs 361
UNENHANCED IMAGE
A

oO

ONTRAST-STRETCHING
B

|
|
I
\
| |
a_ jb

INTENSITY-WINDOW
SLICING

LUMINOUS INTENSITY

Cc

 

2.0 0.0
OPTICAL DENSITY

Fig 2. Methods to manipulate digitized image data for pur-
poses of enhancement of monochromatic cathode ray tube (CRT)
images. The graphs illustrate computer-generated relationships
between image luminous intensity and optical density. (A)
Normal relationship between image displayed on monochrome
CRT and optical density data derived from original source (see
Fig 3A). (B) Enhancement of contrast in the selected range of
optical density (as, for example, between a and b in this illus-
tration) (see Fig 3B). (C) Method of highlighting specific re-
gions within a selected range of optical density (as, for example,
between aand bin this illustration). Structures having optical
densities within this range ave displayed with high intensity
on the face of the CRT (see Fig 3C).

system represents the relationship between the opti-
cal density readings which exist in the image display
memory and the luminous intensities which are pro-
duced on the screen in correspondence with those
numbers. A linear gamma function, such as that
shown in Figure 2A, represents the matching of film
optical density to monochrome image luminous in-
tensity that most faithfully reproduces the image ob-
served on the original autoradiograph.

Contrast stretching and intensity-window slicing
involve the production of nonlinear gamma functions
according to specific criteria. Again, the sole object
of these techniques is to improve the image appear-
ance in terms of human viewing.

A typical gamma function for a contrast-stretched

362 Annals of Neurology Vol 7 No 4 April 1980

image is shown in Figure 2B. The range of optical
densities between points a and 4 has been assigned a
broad range of luminous intensities. This technique is
valuable for enhancing an image that contains data
over only a limited range of optical densities. Com-
pare Figure 3A with its contrast-stretched counter-
part in Figure 3B.

The gamma function for an image that has been
intensity-window sliced (Fig 2C) indicates that a se-
lected range of optical densities (between points
and 4) will be presented in the image with substan-
tially higher luminous intensities than optical den-
sities above and below the limits of this range. The
technique has proved valuable in accentuating edges
and contours in autoradiographic images. Compare
Figure 3A with its intensity-window sliced counter-
part in Figure 3C. The intensity-window slicing tech-
nique has also proved useful in identifying threshold
levels of optical density and glucose utilization for
quantitative purposes; this topic is discussed in a later
section.

PSEUDOCOLOR CODING. Pseudocolor coding is a
technique for transforming a monochrome film
image into an enhanced color image. In essence, this
technique involves division of a broad range of opti-
cal densities into subranges of optical density that are
each assigned a distinct color for presentation.
Pseudocolor transformation takes advantage of the
human visual system’s high sensitivity to color varia-
tion as compared with its more limited ability to re-
solve shades of gray. If employed merely to represent
the optical densities in color, it serves mainly to
beautify autoradiographs and adds relatively little of
scientific value, though it may help to identify the po-
sition and extent of specific structures. The pseu-
docolor transformation can, however, be used to add
a third dimension to the initially two-dimensional
display by converting the qualitative monochromatic
autoradiograph into a quantitative color-coded
image. The computer program includes a routine to
calculate a least-squares best-fitting cubic equation
from the optical densities and concentration values of
the “C plastic standards. A cubic equation was em-
pirically found to be the polynomial equation of low-
est order which closely approximates the relationship
between the 'C concentrations of the standards and
their optical densities. This equation represents a
calibration curve which is used to convert optical
density for the autoradiograph into equivalent tissue
4C concentration. On keyboard command, the pro-
gram utilizes this equation to construct and display,
adjacent to the image of the brain section, a color
scale that encompasses the full range of colors pres-
ent in the image of the brain section and is numerically
calibrated at the boundaries of each of the colors in
 

 

 

 

Fig 3. Reconstructed monochromatic images displayed on a
CRT from a (“CDG autoradiograph of a section of rat brain
processed as described in Figure 2. The images in A, B, and C
illustrate the results of image enhancement obtained by appli-
cation of the methods illustrated in Figure 2A, B, and C, re-
Spectively.

units of local tissue '*C concentration. The program
also contains the routines to solve the operational
equation of the deoxyglucose method [19]. If the
lumped constant, rate constants, and time courses of
the arterial plasma [!*C]DG and glucose concen-
trations are entered, then the program utilizes this
equation as well, and constructs and displays a similar
color scale but now calibrated in units of local cere-
bral glucose utilization. The autoradiographs are thus
transformed into quantitative color-coded maps of
the local rates of cerebral glucose utilization distrib-
uted anatomically throughout the brain section.

Because the scanning microdensitometer can dis-
criminate and the DeAnza image display system can
store 256 gray levels, it is possible to assign an indi-
vidual color for presentation to each of the 256 den-
sity levels stored in the image display memory. In
practice, however, color schemes in which the full
range of values is divided among 12 to 20 colors have
proved to be most useful in quantification of the
deoxyglucose autoradiographs. One such 20-color
scheme, a pseudospectral arrangement, is presented
in Figures 4, 5, and 6. A 12-color arrangement suit-
able for individuals with red-green color blindness is
shown in Figure 7.

The construction of each of these color schemes
was based on three criteria:

1. The scheme should contain enough colors to
allow for division of the full range of values in the
autoradiographic data into sufficiently narrow
ranges of glucose utilization to resolve relatively
small differences in rate of glucose utilization.

2. The scheme should be “logical” (i.e., easily inter-
pretable by the observer). The logical basis of the
20-color scheme in Figures 4, 5, and 6 is primarily
wavelength (i.e., hue), with intensity (i.e., bright-
ness) as a secondary factor. The basis of the 12-
color scheme (Fig 7) is primarily intensity, with
hue as a secondary factor.

3. The scheme should not contain discontinuities in
hue or intensity that could lead to misinterpreta-
tion of results.

The division of the full range of optical densities in
the autoradiograph into intervals represented by the
assigned colors is arbitrary. Colors can be assigned to
equal intervals of the full range. Alternatively, they
can be assigned to intervals that increase in some
continuous way, for example, geometrically or
logarithmically. For example, proportionally nar-
rower ranges of glucose utilization can be assigned to
colors representing lower levels of glucose utiliza-
tion, thus providing greater sensitivity in the lower
ranges (see Figs 4, 5, and 6).

Goochee et al: Image Processing of Autoradiographs 363
 

Fig 4. Pseudocolor reconstructions of
[4C1DG autoradiographs of sections of
the striate cortex from three rhesus

NS 6 Wescc7"1 monkeys: (top) both eyes open; (middle)
GLUCOSE UTD. both eyes patched; (bottom) right eye
patched. The right hemisphere of the
brain is on the right in these illus-
trations. (This illustration has been
previously published in monochrome in
Kennedy et al [8].)

BOW BtS 2% ft

 

364 Annals of Neurology Vol 7 No 4 April 1980

Fig 5. Pseudocolor reconstructions of [''C\DG autoradiographs
of brain sections taken at the level of the medial geniculate
body from rats representative of three age groups: (left) young
adult, 4 to 6 months old; (middle) middle-aged, 14 to 16
months old; (right) aged, 26 to 36 months old. Note the gener-
alized reductions in glucose utilization throughout the brain
with age and reductions in the effective size and rate of glucose
utilization in layer IV of the parietal and auditory cortices.
(Illustration taken from unpublished data of C. B. Smith.)

x
at

 

 

Fig 6. Pseudocolor reconstructions of
[4C\DG autoradiographs from a nor-
mal conscious rat (left column) and a
rat treated with phenoxybenzamine, 21
mglkg (right column), taken at the
level of the paraventricular
hypothalamic nucleus: (top) scans of
full sections with resolution of 100
data points/mm?; (middle) scans at
increased resolution of 400 data
points/mm?; (bottom) scans at highest
level of resolution, 1,600 data points/
mm”. Note the prominence and detail
observed in structures as small as the
paraventricular nucleus, which is met-
abolically activated in the
bhenoxybenzamine-treated animals by
the associated hypertension. (Illustra-
tion taken from unpublished data of
H. E. Savaki.)

Fig 7. Alternative schema of pseudocolor
coding. The autoradiographs recon-
structed are the same as those in Figure
6, but the color scheme is based primar-
tly on intensity, with hue as a second-
ary factor (see text).

Goochee et al: Image Processing of Autoradiographs 365
 

 

366 Annals of Neurology Vol 7 No 4 April 1980
DIGITAL ZOOMING AND RESCANNING AT IN-
CREASED RESOLUTION. Digital zooming involves
magnification of an area of interest in the image dis-
played on the monochrome and color monitors. Any
128 x 128 area of the 256 x 256 image display
memory can be expanded to fill the entire 256 x 256
memory array, each of the 128 x 128 elements in the
original data becoming four elements in the new 256
x 256 image. This magnification process can be re-
peated as many times as desired. The zooming pro-
cess is most often utilized to magnify an area already
scanned at highest resolution (1,600 points/mm’) to
bring out even more detail. Figure 8B was produced
by rescanning at highest resolution a 6.4 x 6.4 mm
area (the area bordered by the rectangle) of the sec-
tion represented in Figure 8A. Figure 8C was pro-
duced by digital zooming of a 3.2 X 3.2 mm area
(bordered by rectangle) of Figure 8B. The distinction
between rescanning and digital zooming is important.
Rescanning involves the collection of 256 X 256 new
data points from a smaller area of the film with an
associated increase in resolution. Digital zooming in-
volves magnification of a portion of existing data
stored in the image display memory and therefore re-
sults in no further increase in resolution.

Computer-assisted High-Resolution

Densitometric Analysis

The capability of the image processing system to scan
and store 256 X 256 individual readings from an area
as small as 6.4 x 6.4 mm, and subsequently to display
these data as a 20 X 20 cm image on the mono-
chrome and color monitors, provides resolution and

 

Fig 8. Examples of features of the image processing system
useful for microdensttometric analysis of [4C|DG au-
toradiographs. (A) Scan (100 data points/mm?) of a full sec-
tion from the brain of a phenoxybenzamine-treated (42 mg/kg)
vat taken at the level of the supraoptic nucleus. Computer-
generated rectangle encompasses area of interest, including
supraoptic nucleus, for rescanning at higher resolution. (B)
Scan of region of autoradiograph containing the supraoptic
nucleus at highest resolution (1,600 data points/mm?).
Computer-generated rectangle encompasses region selected for
image magnification. (C) Image magnification of area in the
rectangle in B obtained by 2X zooming. (D} Operator-
controlled computer-generated outline of area of interest, 1.¢.,
supraoptic nucleus, for computation of average optical density,
14C concentration, and glucose utilization within the desig-
nated area (see text). (E) Operator-controlled computer-
generated rectangle for designation of area of interest selected
for quantitative microdensitometric analysts (see text). (F)
Combination of rectangular outlining and intensity-window
slicing for defining regions in which average optical density,
M4C concentration, and rate of glucose utilization are computed
(see text).

quantification of brain structures at least as small as
100 xm in width.

The precision of the measurement of optical den-
sity is primarily dependent on the character of the
film (e.g., grain size and uniformity of emulsion) and
the settings of apertures on the scanner. Analyses of
replicate measurements of optical densities in au-
toradiographic images of the ['*C]methyl methacry-
late standards, uniform sources of approximately 10
to 20 mm”, yielded coefficients of variation of +12%
for Kodak SB-5 x-ray film and +8% for the finer
grained Kodak MR-1 mammography film when
scanned with apertures of 25 wm, and coefficients of
variation of +6% and +5%, respectively, when
scanned with apertures of 100 ym. These values
for the coefficients of variation were determined for
images with an optical density of approximately 1.0.
The signal-to-noise ratio of the instrument is also a
factor, however, and an inverse relationship there-
fore exists between the coefficient of variation and
the level of optical density being measured.

A factor that influences the accuracy of den-
sitometric measurements is the so-called halo effect
associated with “C film autoradiography of tissue
sections 20 wm thick. The influence of this factor has
been evaluated by analysis of autoradiographs of sec-
tions of a rat brain that was itself unlabeled but con-
tained a plug of uniformly labeled brain tissue in-
serted into it before sectioning. The results indicate
that the halo effect diminishes semilogarithmically
with distance from the true border, with a half-
distance of 50 (SD +10) um. Mainly for this reason,
we doubt that the quantitative resolution of the
method is finer than 100 wm.

The optical density data resident in the image dis-
play memory can be utilized to compute the average
optical density for any selected portion of the image.
If the appropriate experimental constants have been
entered, the data can be converted to mean tissue
radioactive tracer concentration (in nCi/gm tissue) or
to mean cerebral glucose utilization (in #mol/100
gm/min) for the selected portion of the image. The
mean density, concentration, or rate of glucose utili-
zation for a particular brain structure can be deter-
mined by any one of the three following procedures:

1. A selected portion of the image display corre-
sponding to a given cerebral structure is surrounded
by a visible outline constructed under joystick con-
trol (Fig 8D). The mean value within the outlined
area is then automatically computed and is displayed
numerically, along with the number and standard de-
viation, for the number of elements within the out-
line that were averaged.

2. A series of small rectangular areas are measured
and collected from a selected structure and then

Goochee et al: Image Processing of Autoradiographs 367
averaged by the computer (Fig 8E). The size of the
rectangle and the number of readings collected are at
the operator’s discretion.

3. The rectangular outlining and intensity-window
slicing features have been combined to produce the
third alternative procedure. A selected region of the
display is outlined by a rectangle generated and posi-
tioned under joystick control. A particular range or
window is specified for the values of optical density,
tadioactive tracer concentration, or glucose utiliza-
tion to be analyzed. The upper and lower limits of the
window are under the operator’s control. An average
is then computed for all the individual readings
within the outlined area that fall within the specified
range. The readings which fall within that range
ate continuously highlighted on the monochrome
monitor by means of the intensity window-slicing
technique previously discussed (Fig 8F).

To illustrate the capability of the system for quan-
titative densitometry, consider that the density, tis-
sue concentration, or rate of glucose utilization in a 1
x 1 mm area on the autoradiograph can be computed
as the mean of 1,600 distinct readings, 1 for each 25
x 25 wm region within that area. This capability al-
lows determination of true weighted averages for
each structure and resolves the problem encountered
with manual densitometry of achieving adequate
sampling of readings from any given structure.

Because the scanner takes readings at equally
spaced intervals, the number of readings sampled
from each individual structure is directly propor-
tional to the area of its representation on the au-
toradiograph. The mean of all the readings over an
area is therefore weighted for the sizes of all the
structures represented in that area. This capability for
weighted averaging, together with the speed with
which the computer can access data, permits rapid
determination of the weighted mean glucose utiliza-
tion for an entire brain section as well as for indi-
vidual regions within the section. From mean values
of glucose utilization and the number of data points
contributing to each mean collected from serial sec-
tions throughout the brain, it is also possible to com-
pute from the tens of thousands of data points the
average rate of glucose utilization in the brain as a
whole, properly weighted for the masses of its com-
ponent parts.

Discussion

Computer-assisted image processing was developed
in the early 1960s during the National Aeronautics
and Space Administration’s early unmanned space
program. Expansion of the technology of image pro-
cessing was subsequently stimulated by requirements
of the intelligence community for analysis of pictorial

368 Annals of Neurology Vol 7 No 4 April 1980

data [11]. Continued advances in computer technol-
ogy have only recently reduced the barriers of cost
and processing time sufficiently to encourage prolif-
eration of image processing systems within the sci-
entific community at large. A number of reports on
the design and programming of image processing
systems have been published and are helpful for
guidance in the development of image processing
techniques for specific uses [1—3, 11-13, 15].

The application of image processing techniques to
the [“C]DG method evolved from a need to com-
bine more effectively the method’s capabilities of
quantitatively determining and anatomically localiz-
ing the rates of glucose utilization throughout the
central nervous system. Autoradiography is an inte-
gral part of the method, and the autoradiographs ob-
tained with it often present a striking illustration of
the pattern of distribution of metabolic activities
throughout the brain. Their interpretation, however,
is severely limited. First of all, they are not fully
quantitative; the shades of gray in a ['*C]DG au-
toradiograph depict only the relative rates of glucose
utilization in the structural components of the brain.
Autoradiographs from different experiments cannot
therefore be directly compared, except to search for
differences in the distribution of optical densities
among the various cerebral structures. It is then often
uncertain whether areas of apparent increase in rela-
tive density indicate heightened glucose utilization in
that area or decreased utilization in other regions.
Such questions can be resolved only by full
quantification. Quantification has thus far required
manual densitometry, which involves the measure-
ment of optical densities in selected structures of the
brain.

Experience with the method has uncovered a
number of problems. Enormous numbers of readings
must be made to ensure that sampling of each struc-
ture is adequate to provide true representative values
for each region. It has been found, however, that
anatomical structures are usually far from homoge-
neous. If this heterogeneity is ignored by sampling
the entire structure, potentially valuable information
is lost. If the structures are further subdivided ac-
cording to heterogeneity, then enormous, tedious,
and cumbersome tables of data are generated. Fi-
nally, structures of interest are often not visualized,
and therefore not measured, because of their small
size or the limited ability of the human visual system
to discriminate shades of gray.

The computer-assisted image processing system
described in this report largely resolves these prob-
lems. It scans and digitizes autoradiographs and re-
constructs them on a color monitor in pseudocolor,
which adds to the spatial distribution a third dimen-
sion representing the actual rate of glucose utiliza-
tion. The rate of glucose utilization can then be
localized visually directly in each locus of the brain
with a spatial resolution of at least 100 zm. Present
equipment is capable of even finer spatial resolution.
It can scan with an aperture of 25 wm and can ac-
cumulate and store 256 X 256 data points in a 6.4 x
6.4 mm area. The 100 um limit on resolution is
therefore not in the image processing system; it is a
result of the grain size of the film and the halo effect
associated with 'C autoradiography. Similarly, use of
color coding enhances the resolution of differences
in rates of glucose utilization because of the ability of
the human visual system to discriminate colors and
tints. Reconstructed color-coded autoradiographs
greatly increase perception of the marked heteroge-
neity of metabolic activities throughout the brain.

These color-coded metabolic maps of the brain are
comparable to the maps of blood flow used by Las-
sen, Ingvar, and Skinhgj [10] to visualize rates of
blood flow in the human cerebral cortex. Although
their technique does not require the use of image
processing, rates of blood flow computed from
changes in counting rates measured with their bat-
teries of fixed radiation detectors are coded into
color and displayed spatially on a computer-con-
structed outline of the human cerebral cortex.

The benefits of image processing are not confined
to the advantages of color coding. At the very least it
provides rapid, semiautomated, comprehensive mi-
crodensitometry. Contrast enhancement makes it
possible to visualize structures not obvious in au-
toradiographs because of insufficient differences in
optical density between the structure and surround-
ing regions. Image enhancement by digital zooming
or rescanning of smaller regions makes it possible
to visualize microheterogeneity within even small
structures, and measurements can be made in far
smaller regions than is possible with manual den-
sitometry. The capability for weighted averaging of
local cerebral glucose utilization is also advantageous.
Although the deoxyglucose method was designed
specifically for measurement of glucose utilization in
discrete regions of the brain, there are occasions
when it is desirable to know also the average glucose
utilization of the brain as a whole. It has hitherto
been impossible to determine overall cerebral glu-
cose utilization from local rates measured with the
deoxyglucose technique because the overall average
must be properly weighted for the relative masses of
the individual structures. The weighting factors are
generally unknown. The properties of image pro-
cessing are such that when an area is averaged, the
weighting factors are automatically incorporated be-
cause the number of readings per structure is directly
proportional to the size of the structure. For exam-
ple, values of local cerebral glucose utilization in the

conscious monkey have recently been reported [9].
Preliminary studies in these monkeys by the com-
puter-assisted weighted averaging technique thus far
indicate a value of 33 smol/100 gm/min for aver-
age glucose utilization by the brain as a whole, a value
very close to those obtained previously by global
techniques in very lightly anesthetized monkeys [16]
or normal conscious young men [17].

The image processing system used in the present
study represents our initial effort, and a relatively
economical one, to apply image processing tech-
niques to the mapping of local cerebral glucose utili-
zation. Equipment and technology exist, however, to
expand greatly the scope of the technique. Scanning
microdensitometers are available that can scan with
greater resolution at higher speeds. There are image
memory and display systems that allow storage and
display of 512 x 512 data elements rather than the
256 X 256 used here, thereby producing finer detail
and resolution. With computer systems of greater
capacity and speed, it should be possible to develop
programs that reconstruct maps of the distribution of
metabolic activity three dimensionally in the entire
brain from data obtained by scanning of serial sec-
tions of the entire brain. Efforts in this direction are
presently in progress in this laboratory.

Recent developments in computerized emission
tomography have provided the means to apply the
2-deoxyglucose method to humans. A_ positron-
emitting derivative of deoxyglucose, 2—[**F]fluoro-
2-deoxy-D-glucose, has been synthesized and found
to retain the necessary biochemical properties of
2-deoxyglucose [14]. The resolution thus far ob-
tained in human studies is considerably below that
which has been obtained with autoradiography [14],
but it can be expected to improve as a result of tech-
nological advances. Theoretical limitations inherent
in positron-emission tomography, however, preclude
the achievement of resolution equal to that possible
with autoradiography. The image processing tech-
niques described in this report can readily be adapted
to tomographic data and, when combined with the
[!8F]fluorodeoxyglucose procedure, make it possible
to obtain quantitative metabolic maps of glucose
utilization throughout the human brain.

 

The authors express their appreciation to Ruth Bower for her
editorial assistance in the preparation of this manuscript and to
J. D. Brown for his monochrome and color photography.

 

References
1. Andrews H: Monochrome digital image enhancement. Appl
Optics 15:495-503, 1976
2. Booth JM, Schroeder JB: Design considerations for digital
image processing systems. Computer 10:15-—20, 1977

Goochee et al: Image Processing of Autoradiographs 369
10.

. Grogono P: Programming

. Briggs SJ: Procedures for Determining Digital CRT Displays’

“Best” Gamma Function. Seattle, Boeing Aerospace Com-
pany, Document D180-15861-1, January, 1978

. Collins RC, Kennedy C, Sokoloff L, et al: Metabolic anatomy

of focal motor seizures. Arch Neurol 33:536-542, 1976
in Pascal. Reading, MA,
Addison-Wesley, 1978

. Jarvis CD, Mishkin M, Shinohara M, et al: Mapping the pri-

mate visual system with the ['*C]2-deoxyglucose technique.
Neurosci Abstr 4:632, 1978

. Jensen K, Wirth N: Pascal User Manual and Report. New

York, Springer-Verlag, 1974

. Kennedy C, Des Rosiers MH, Sakurada O, et al: Metabolic

mapping of the primary visual system of the monkey by means
of the autoradiographic ['*C]deoxyglucose technique. Proc
Natl Acad Sci USA 73:4230-4234, 1976

. Kennedy C, Sakurada O, Shinohara M, et al: Local cerebral

glucose utilization in the normal conscious Macaque monkey.
Ann Neurol 4:293-301, 1978
Lassen NA, Ingvar DH, Skinhgj E: Brain function and blood
flow. Sci Am 239:62-71, 1978

11. Lyons EC: Digital image processing. Computer 10:12-14,
1977
370. Annals of Neurology Vol 7 No 4 April 1980

13.

14.

15.

16.

17.

18.

19.

. Masters CT: ID-2000 Series Image Display System Pro-

gramming Manual. Santa Clara, CA, DeAnza Systems, 1977
Pratt WK: Digital Image Processing. New York, Wiley In-
terscience, 1978

Reivich M, Kuhl D, Wolf A, et al: The ['*F]fuorodeoxyglu-
cose method for the measurement of local cerebral glucose
utilization in man. Circ Res 44:127—137, 1979

Rosenfeld A, Kak AC: Digital Picture Processing. New York,
Academic, 1976

Schmidt CF, Kety SS, Pennes HH: The gaseous metabolism
of the brain of the monkey. Am J Physiol 143:33—52, 1945
Sokoloff L: Circulation and energy metabolism of the brain,
in Siegel GJ, Albers RW, Katzman R, et al (eds): Basic
Neurochemistry. Second edition. Boston, Little, Brown,
1976, pp 388-413

Sokoloff L: Relation between physiological function and
energy metabolism in the central nervous system. J Neu-
rochem 29:13—26, 1978

Sokoloff L, Reivich M, Kennedy C, et al: The ['4C]deoxyglu-
cose method for the measurement of local cerebral glucose
utilization: theory, procedure, and normal values in the con-
scious and anesthetized albino rat. J Neurochem 28:897-916,
1977