- Email: [email protected]
Assessment of gastric mucosal ulceration by computerized image processing
JOURNAL
OF SURGICAL
RESEARCH
Assessment
42,271-283
(1987)
of Gastric Mucosal Ulceration Processing’
HAROLDV.GASKILL
by Computerized
111,M.D. ANDBARRYA.
Image
LEVINE,M.D.~
Department of Surgery, Audie L. Murphy Memorial Veterans Hospital and the University of Texas Health Science Center, San Antonio, Texas Submitted for publication December 2, 1985 The absence of a rapid, objective, and reproducible method for assessing mucosal ulceration has long been a frustration to research in the field of gastric physiology. This study compared assessment of mucosal injury by computerized image processing with values obtained by the shed microsphere technique. An ex vivo gastric chamber model based on miniature swine was used. Five chambers were subjected to hemorrhagic shock and acid-bile solution and five chambers were maintained in normotension and exposed to normal saline (controls). After 3 hr, mucosal injury was assessed by both techniques. The chambers exposed to shock and acid-bile all developed visible ulceration ranging from 1.8 to 99.7 cm2 by computerized image processing. These values correlated well with the results obtained by the shed microsphere technique (23 to 4 19 mg, r = 0.99, Pi .05). No ulceration developed in the control chambers. Implementation of computerized image processing as well as its limitations is discussed. 0 1987 Academic Press Inc.
One of the major unanswered questions in gastric pathophysiology is the exact mechanism leading to gastric mucosal ulceration. Although a variety of factors have been implicated, their relative importance in the etiology of these lesions is unknown. A major frustration for investigators in this field has been the lack of any objective and reproducible method for quantifying mucosal injury. This paper presents a new technique based on the use of computerized image processing: a rapid, objective, and reproducible method for assessing mucosal ulceration. METHODS
Ten miniature swine (University of Texas Systems Animal Facility, Bastrop, TX) weighing lo- 15 kg each were fasted for 48 hr (water allowed ad libitum) prior to study. * This work was supported in part by the Veterans Administration Research Medical Service. ’ To whom reprint requests should be addressed: Department of Surgery, The University of Texas Health Science Center, 7703 Floyd Curl Drive, San Antonio, Texas 78284-7842.
Each animal was anesthetized with intravenous chloralose ( 100 mg/kg/iv) and mechanically ventilated with supplemental oxygen via a tracheostomy. The P&O2 was maintained between 30 and 40 Torr by adjusting tidal volume and/or respiratory rate. Using a cervical incision, a PE-240 catheter was introduced into the right common carotid artery and advanced retrograde into the left cardiac ventricle for injection of 1.0-2 pm radiolabeled microspheres (New England Nuclear, Boston, MA). The right femoral artery was cannulated with a PE-200 catheter for measurement of systemic arterial pressure and withdrawal of the reference sample during injection of radiolabeled microspheres (8.86 ml/mm). A number 5-French Swan-Ganz thermodilation catheter was flow directed into the pulmonary artery for measurement of cardiac output. Each animal underwent celiotomy through a 10 cm midline incision. The greater curve of the stomach was isolated with Kocher clamps and a vascularized pedicle of full thickness stomach was incorporated as the bottom of a circular plexiglass 277
0022-4804187 $1 SO Copyright 8 1987 by Academic Press, Inc. AI1 tight.9 of reproduction in any form reserved.
278
JOURNAL
OF SURGICAL
RESEARCH:
chamber. The chamber was suspended just above the abdomen by an aluminum frame apparatus and the remainder of the abdomen was covered with a thermostatically controlled heating blanket. The chambers were then loaded with 150 ml of either acid-bile solution (N = 5) or normal saline (N = 5). The acid bile solution was prepared by adding 5 ml of bile aspirated from the animals gallbladder to 95 ml of warm water. Then, 100 ml of 280 mM HCl was added slowly to the bile solution. This resulted in a final concentration of 2.5% bile in 140 mM HCl. After a 30-min stabilization period lo6 15 pm spheres labeled with 46Sc were injected into the left cardiac ventricular catheter. These spheres entered the systemic circulation and were partially distributed with a similar proportion of the cardic output to the gastric mucosa. The animals treated with acid/bile were then bled to a mean arterial pressure of 50 mm Hg and maintained at that level for 3 hr. The saline treated animals were maintained in normotension (systolic blood pressure greater than 100 mm Hg) for 3 hr. After 3 hr the fluid from the chamber was drained and saved and the mucosa was also photographed with color transparency film. The mucosa was then rinsed with saline and gently brushed with a solution of N-acetyl cysteine to remove adherent clots and mucous. This material was combined with the chamber contents in an 8 X 12 in. Pyrex baking dish lined with heavy-duty waxed paper. The mucosa/submucosa was sharply divided from the muscularis/serosa, blotted dry, and weighed. The tissue was then divided into portions of approximately 1 g and packed into glass tubes to a height of 2 cm. The dish containing the gastric contents was then evaporated to dryness in an oven at 70°C. The waxed paper containing the dried material was cut into strips 2 cm wide. These strips were rolled up and placed in glass tubes thus maintaining a sample geometry similar to that of the gastric tissue. For each experiment, tubes representing background, gastric tissue, and gastric contents were counted for 5 min each during a single day.
VOL. 42, NO. 3, MARCH
COMPUTERIZED
IMAGE
1987
PROCESSING
The color transparencies of the gastric mucosal were mounted on a light table designed to provide uniform light transmission at its surface. The image was then recorded through a red filter by a monochrome video camera (EYECAM), connected to an image digitizer/processor (Grinnell Systems) and host (Digital Equipment Co. VAX). A portion of ulcerated mucosa as it is input through the video monitor is seen in Fig. 1. Since areas of ulceration are ssociated with hemorrhage into ulcer beds, these areas are stained dark black by the formation of acid hematin. The digitized images are then analyzed by a threshold function which converts black and dark-grey portions to pure black while lighter grey and white portions are converted to pure white. With the grey scale divided into 256 portions (0 = pure black, 255 = pure white), a threshold of 30 consistently identifies ulcerated areas. This image can also be displayed on the monitor as seen in Fig. 2. A trackball and software are used to designate the area of interest on the video image and the computer then calculates the area of ulcerated mucosa as a percentage of the total mucosa analyzed. This percentage is multiplied by the total area of the floor of the chamber yielding the area of ulceration in square centimeters. SHED
MICROSPHERE
TECHNIQUE
The shed microsphere technique served as the basis for comparison to computerized image processing. This technique has been reported previously [7, 81. In summary, radiolabeled microspheres injected into the left cardiac ventricle at a baseline period enter the circulation and are uniformly distributed throughout the gastric mucosa. When ulceration occurs these spheres are shed into the lumen along with the mucosa. Thus, the number of spheres recovered from the lumen is proportional to the volume of microsphere bearing mucosa shed. The weight of the shed mucosa is calculated by dividing the total radioactivitv in the intact mucosa bv the total
FIG. 1. Portion of ulcerated mucosa photographed through a red filter and ready for viewing by the video camera.
FIG. 2. Digitized image from Fig. 1. 279
280
JOURNAL
OF SURGICAL
RESEARCH:
weight of the intact mucosa to determine the radioactivity per unit weight. Once the total radioactivity recovered from the lumen has been determined, it is possible to calculate the weight of the ulcerated tissue from the equation: weight of ulcerated tissue =
weight of intact mucosa radioactivity of intact mucosa X radioactivity
from the lumen.
Accuracy of this method depends on reliable recovery of shed microspheres from the chamber. In pilot studies with this model in our laboratory, a known number of spheres placed in the chamber were recovered at a greater than 86% rate in each instance. For each animal the area of ulceration, as determined by computerized image processing, was paired with the weight of shed mucosa as determined by the shed microsphere technique. Linear regression of the resulting data was performed and the correlation coefficient calculated. RESULTS
After 3 hr, gastric mucosa in all five chambers exposed to normal saline alone under normotensive conditions were without visible ulceration. By contrast, the gastric mucosa in chambers exposed to acid/bile and systemic hypotension all developed visible ulceration. When the color photographs from these ulcerated stomachs were analyzed by computerized image processing the calculated area of ulceration ranged from 1.8 to 99.7 cm2. For each stomach, the area of ulceration in square centimeters as determined by computerized image processing was compared to the weight of shed mucosa in milligrams as determined by the shed microsphere technique. In each of the acid/bile/ hypotension animals, radioactivity correlating with greater than 460 spheres recovered from the gastric chamber was counted-assuring a high degree of confidence. The values derived by this latter
VOL. 42, NO. 3, MARCH
1987
method ranged from 23 to 4 19 mg. The comparison of damage estimated by the two methods showed a high degree of correlation (r = 0.99, P < 0.01) (Fig. 3) Although there was no visible ulceration on the stomachs exposed to normal saline alone some radioactivity was recovered from the chambers. However, the radioactivity counted in each of the five animals represents less than 200 spheres recovered and may well represent background alone. Therefore, shed tissue weights derived from this data in the saline group cannot be viewed as conclusive. Since these weights were negligible, the question remains moot. The calculated weight of shed mucosa for these stomachs averaged 12.1 f 4.8 mg versus an average of 143.6 f 7 1.3 mg for the stomachs subjected to acid/bile and systemic hypotension (mean + SEM, P < 0.01). DISCUSSION
Investigators studying the pathophysiology of stress-mediated gastric mucosal ulceration have long been frustrated by the absence of an objective and reproducible method of quantifying mucosal injury. The most common method used is scoring by a blinded observer using an ad hoc system. The report of Desiderato (Ref. [6]) is typical. They calculated the ulcer score by giving lesions under 1 mm a value of 1, those between
FIG. 3. Gastric mucosal ulceration in square centimepared to milligrams of ulcerated mucosa determined by the shed microsphere technique, (r = 0.99, P -e 0.01).
GASIULL
AND
LEVINE:
COMPUTERIZED
1 and 2.9 mm a value of 2 between 3 and 4.9 a value of 3, and those over 5 mm a value of 4. Only those lesions which extended to or through the submucosa of the glandular stomach were counted as ulcers. In rare cases of disagreement between judges, the mean of their two values was computed. These authors also calculated the percentage of subjects per group exhibiting ulceration and the number of ulcers per subject. By comparison, Guth and Gati (Ref. [9]) used a similar system. In their study the stomachs were opened along the greater curvature, washed, and mucosal lesions measured and scored as follows: petechiae, 1; and erosions < 1 mm, 2; l-2 mm, 3; 2-4 mm, 4; and >4 mm, 5. Although these systems are clearly workable, minor differences make comparison of results between studies difficult. In addition, there is the possibility that significant findings may be elucidated by one scale but not another. A later publication by Grossman et al. (Ref. [ 181) addressed some of these issues by defining the ulcer index. Lesions in the stomach and small intestine were measured under a dissecting microscope with a l-mm2 grid eyepiece (X 10). The ulcer index was expressed as total of the area in square millimeters of the individual lesions. Another approach to assessing mucosal injury is based on grading the severity of selected lesions. This is done by examining histologic sections with light or electron microscopy (Refs. [ 10, 11, 201). The disadvantages of this technique are several. First, it is impractical to sample and grade every ulceration in the specimen of mucosa. Thus, an arbitrary system must be devised to determine which lesions should be examined. Second, any injury seen in the sections must ultimately be quantified by a system with the same limitations of those used for intact mucosa. Many investigators have used metabolic parameters such as transmucosal electrical potential difference and electrolyte flux to assess mucosal injury (Refs. [3, 5, 14, 16, 171). The advantages of these techniques are many. They are based on well established
IMAGE
PROCESSING
281
methods and provide highly objective and quantifiable results. They are also well suited to continuous or multiple measurements throughout an experiment. Unfortunately, these methods fail to address the focal nature of mucosal ulceration. The values obtained reflect the average metabolic status of the entire mucosal surface. Since ulceration typically involves only a few percent of the surface area, significant metabolic alterations associated with these lesions may be obscured. Other methods used for assessing the metabolic status of the mucosa include measurement of acid secretion (or back diffusion), tissue gas tension, and blood flow (Refs. [ 13, 191). However, most techniques for measuring these parameters yield only an average value for the entire mucosal surface. As mentioned above, such techniques are not well suited to studies of focal lesions. Although electrical potential difference and tissue gas tension can be measured in areas as small as a single cell (Refs. [ 1, 13]), the area to be evaluated must be selected prior to actual ulceration. Furthermore, these techniques are inherently limited to sampling only a very small portion of the total mucosa. Thus, changes preceding ulceration are detected only by chance. Since all of these methods require contact with or direct puncture of the mucosal surface the possibility of altering mucosal metabolism at the area under study cannot be excluded. One very promising technique has recently been reported by Ritchie et al. (Ref. [2]). Bathing fluid from an ex vivo gastric chamber is sampled and assayed for mucosal efflux of DNA. Since DNA is located primarily in the nucleus of the gastric mucosal cells, accumulation of DNA in the gastric lumen is roughly proportional to cellular loss. One of the advantages of this system is that it should be able to detect even minimal mucosal injury. The most important drawback of this method is that most common methods of assaying DNA are not accurate when the samples are contaminated with gastric secretions-notably sialic acid (Ref.
282
JOURNAL
OF
SURGICAL
RESEARCH:
[4]). Accordingly, techniques requiring the use of expensive spectrofluorometric equipment are necessary. Additional errors may be caused by destruction of the DNA by DNase, an enzyme ubiquitous in animal tissues. Finally, the processing of tissue and fluids for DNA measurement involves multiple, time consuming steps. Thus, it is not ideally suited to the processing of large numbers of samples. Another method which has been used to evaluate mucosal architecture is microangiography using agents such as stained hemoglobin or fine-grain barium sulfate (Refs. [ 12, 131). These techniques reveal changes in the microvascular structure of the tissue which may be associated with the mucosal stress response-including ulceration. The primary disadvantage of microangiography is that only one assessment is possible for each section of tissue. Furthermore, changes observed are not readily quantifiable by any well standardized and objective technique. The technique reported herein, computerized image processing, has many advantages when compared to these other methods. First, it is highly objective. The major subjective determination is the image density threshold between ulcerated mucosa and nonulcerated mucosa. In practice, however, the density gradient between acid hematinstained ulceration and normal or ischemic mucosa is fairly steep. Thus, large changes in the threshold value result in relatively small changes in the area of ulceration perceived by the system. The gradient can be increased even more by using a red filter between the input video camera and the transparency of the mucosa to be analyzed. As with any experimental technique, consistency and attention to detail are necessary for reproducible results. Exposure conditions must be carefully controlled. Uniform biplane illumination is necessary to avoid shadowing by mucosal folds. Shadows may be interpreted by the system as ulceration, thus overestimating the area of injury. When data from several experiments are to be compared, it is important to record film
VOL.
42, NO.
3, MARCH
1987
type, shutter speed, F-stop, and the distance from camera to mucosa. In practice, these factors are easily controlled by mounting the camera and lights in a frame above the operating table. There are some disadvantages to computerized image processing. Cost is certainly a significant factor. The computer and video hardware described under Methods section has been in use for several years and cost in excess of $100,000 at the time it was originally purchased. Today, all of the necessary equipment would cost less than $lO,OOOincluding the computer. Investigators with access to a computer could set up a workable system for three to five thousand dollars. While this is certainly more expensive than viewing and grading the injury by traditional methods, it is not unreasonable. One serious limitation of computerized image processing is that it cannot assess the depth of the lesions. A lesion involving only the superficial mucosa is interpreted exactly the same as a lesion penetrating to the muscularis. In addition, lesions which do not bleed will be missed entirely. Since this system is based on the detection of the black acid hematin residue from mucosal hemorrhage, care is required to avoid disruption of this material. Although it is possible to recover stomachs from intact animals, this may result in disruption of the clots. The ideal preparation is the isolated mucosal chamber model described by Moody et al. (Ref. [ 151). This model facilitates consistent photographic technique as well as serial photographs during the experiment. Another limitation is the uneven nature of the gastric mucosa. Lesions which are on the side of a mucosal fold will be viewed at an angle by the camera, thus reducing the apparent area of the ulcer. Although this effect can be reduced by mounting the mucosa carefully in the chamber, it cannot be eliminated entirely. In summary, computerized image processing has many advantages over traditional methods of assessing mucosal ulceration. Multiple mesurements can be made during
GASKILL
AND LEVINE:
COMPUTERIZED
an experiment. The data produced is consistent and highly reproducible. The necessary equipment is available at reasonable cost. However, it is best suited for use in open systems such as the isolated gastric chamber. REFERENCES 1. Bowen, J. C., and Gang, D. K. Effect of graded mechanical ischemia on oxygen tension and electrical potential in the canine gastric mucosa. Castroenterology 73: 84, 1977. 2. Carter, K. J., Farley, P. C., and Ritchie, W. P. Effect of topi.cal bile acids on gastric surface epithelial cells. Surgery 96: 196, 1984. 3. Chaudhury, T. K., and Jacobson, E. D. Prostaglandin cytoprotection of gastric mucosa. GastroenteroIogy 74: 59, 1978. 4. Croft, D. N., and Lubran, M. The estimation of deoxyribonucleic acid in the presence of sialic acid: Application to analysis of human gastric washings. B&hem. J. 95: 612, 1965. 5. Davenport, H. W. Gastric mucosal injury by fatty and acetylsalicylic acids. Gastroenterology 46: 245, 1964. 6. Desiderato, O., MacKinnon, J. R., and Hissom, H. Development of gastric ulcers in rats following stress termination. J. Comp. Physiol. Psychol. 87: 208, 1974. 7. Gaskill, H. V., III, Sirinek, K. R., and Levine, B. A. 16,16-Dimethyl prostaglandin E2 reverses focal mucosal &hernia associated with stress ulcers. J. Surg. Res. 37: 83, 1984. 8. Gaskill, H. V., III, Sirinek, K. R., and Levine, B. A. Focal mucosal &hernia is a precursor to stress-mediated gastric mucosal ulceration. Surg. Gastroenterol. 2: 225, 1983. 9. Gati, T., and Guth, P. H. Mucosal lesions due to gastric distension in the rat. Dig. Dis. Sci. 22: 1083, 1977. 10. Goodman, A. A., and Osborne, M. P. An experimental model and clinical definition of stress ulceration. Surg. Gynecol. Obstet. 134: 563, 1972.
IMAGE
PROCESSING
283
11. Guth, P. H., Paulsen, G., and Foroozan, P. Experi-
mental chronic gastric ulcer due to &hernia in rats. 12 Dig. Dis. Sci. 20: 824, 1975. Guth, P. H., and Hall, P. Microcirculatory and mast ’ cell changes in restraint-induced gastric ulcer. Gustroenteroiogy 50: 562, 1966. 13. Kivilaakso, E., Ahonen, J., Aronsen, K.-F., Hockerstedt, K., Kalima, T., Lempinen, M., Suoranta, H., and Vemerson, E. Gastric blood flow, tissue gas tension and microvascular changes during hemorrhage-induced stress ulceration in the pig. Amer. J. Surg. 143: 322, 1982. 14. Konturek, S. J., Tasler, J., Obtulowicz, W., Kwiecien, N., Sito, E., and Oleksy, J. Effect of methylated prostaglandin E2 analogue on canine and human gastric mucosal barrier. Gastroenterol. Clin. Biol. 2: 177, 1978. 15. Moody, F. G., Cheung, L. Y., Simons, M. A., and Zalewsky, C. Stress and the acute gastric mucosal lesion. Dig. Dis. Sci. 21: 148, 1976. 16. Ritchie, W. P., Jr., Cherry, K. J., Jr., and Gibb, A. Influence of methylprednisolone sodium succinate on bile-acid-induced acute gastric mucosal damage. Surgery 84: 283, 1978. 17. Ritchie, W. P., Jr., Shearbum, E. W., III, and Nading, A. Relationship of transmural electrical potential difference to changes in gastric mucosal permeability to H+ and blood flow. Amer. J. Surg. 135: 110, 1978. 18. Satoh, H., Guth, P. H., and Grossman, M. I. Role of bacteria in gastric ulceration produced by indomethacin in the rat: cytoprotective action of antibiotics. Gastroenterology 84: 483, 1983. 19. Shirazi, S. S., Mueller, T. M., and Hardy, B. M. Canine gastric acid secretion and blood flow measurement in hemorrhagic shock. Gastroenterology 73: 75, 1977. 20. Svanes, K., Ito, S., Takeuchi, K., and Silen, W. Restitution of the surface epithelium of the in vitro frog gastric mucosa after damage with hyperosmolar sodium chloride. Morphologic and physiologic characteristics. Gastroenterology 82: 1409, 1982.
OF SURGICAL
RESEARCH
Assessment
42,271-283
(1987)
of Gastric Mucosal Ulceration Processing’
HAROLDV.GASKILL
by Computerized
111,M.D. ANDBARRYA.
Image
LEVINE,M.D.~
Department of Surgery, Audie L. Murphy Memorial Veterans Hospital and the University of Texas Health Science Center, San Antonio, Texas Submitted for publication December 2, 1985 The absence of a rapid, objective, and reproducible method for assessing mucosal ulceration has long been a frustration to research in the field of gastric physiology. This study compared assessment of mucosal injury by computerized image processing with values obtained by the shed microsphere technique. An ex vivo gastric chamber model based on miniature swine was used. Five chambers were subjected to hemorrhagic shock and acid-bile solution and five chambers were maintained in normotension and exposed to normal saline (controls). After 3 hr, mucosal injury was assessed by both techniques. The chambers exposed to shock and acid-bile all developed visible ulceration ranging from 1.8 to 99.7 cm2 by computerized image processing. These values correlated well with the results obtained by the shed microsphere technique (23 to 4 19 mg, r = 0.99, Pi .05). No ulceration developed in the control chambers. Implementation of computerized image processing as well as its limitations is discussed. 0 1987 Academic Press Inc.
One of the major unanswered questions in gastric pathophysiology is the exact mechanism leading to gastric mucosal ulceration. Although a variety of factors have been implicated, their relative importance in the etiology of these lesions is unknown. A major frustration for investigators in this field has been the lack of any objective and reproducible method for quantifying mucosal injury. This paper presents a new technique based on the use of computerized image processing: a rapid, objective, and reproducible method for assessing mucosal ulceration. METHODS
Ten miniature swine (University of Texas Systems Animal Facility, Bastrop, TX) weighing lo- 15 kg each were fasted for 48 hr (water allowed ad libitum) prior to study. * This work was supported in part by the Veterans Administration Research Medical Service. ’ To whom reprint requests should be addressed: Department of Surgery, The University of Texas Health Science Center, 7703 Floyd Curl Drive, San Antonio, Texas 78284-7842.
Each animal was anesthetized with intravenous chloralose ( 100 mg/kg/iv) and mechanically ventilated with supplemental oxygen via a tracheostomy. The P&O2 was maintained between 30 and 40 Torr by adjusting tidal volume and/or respiratory rate. Using a cervical incision, a PE-240 catheter was introduced into the right common carotid artery and advanced retrograde into the left cardiac ventricle for injection of 1.0-2 pm radiolabeled microspheres (New England Nuclear, Boston, MA). The right femoral artery was cannulated with a PE-200 catheter for measurement of systemic arterial pressure and withdrawal of the reference sample during injection of radiolabeled microspheres (8.86 ml/mm). A number 5-French Swan-Ganz thermodilation catheter was flow directed into the pulmonary artery for measurement of cardiac output. Each animal underwent celiotomy through a 10 cm midline incision. The greater curve of the stomach was isolated with Kocher clamps and a vascularized pedicle of full thickness stomach was incorporated as the bottom of a circular plexiglass 277
0022-4804187 $1 SO Copyright 8 1987 by Academic Press, Inc. AI1 tight.9 of reproduction in any form reserved.
278
JOURNAL
OF SURGICAL
RESEARCH:
chamber. The chamber was suspended just above the abdomen by an aluminum frame apparatus and the remainder of the abdomen was covered with a thermostatically controlled heating blanket. The chambers were then loaded with 150 ml of either acid-bile solution (N = 5) or normal saline (N = 5). The acid bile solution was prepared by adding 5 ml of bile aspirated from the animals gallbladder to 95 ml of warm water. Then, 100 ml of 280 mM HCl was added slowly to the bile solution. This resulted in a final concentration of 2.5% bile in 140 mM HCl. After a 30-min stabilization period lo6 15 pm spheres labeled with 46Sc were injected into the left cardiac ventricular catheter. These spheres entered the systemic circulation and were partially distributed with a similar proportion of the cardic output to the gastric mucosa. The animals treated with acid/bile were then bled to a mean arterial pressure of 50 mm Hg and maintained at that level for 3 hr. The saline treated animals were maintained in normotension (systolic blood pressure greater than 100 mm Hg) for 3 hr. After 3 hr the fluid from the chamber was drained and saved and the mucosa was also photographed with color transparency film. The mucosa was then rinsed with saline and gently brushed with a solution of N-acetyl cysteine to remove adherent clots and mucous. This material was combined with the chamber contents in an 8 X 12 in. Pyrex baking dish lined with heavy-duty waxed paper. The mucosa/submucosa was sharply divided from the muscularis/serosa, blotted dry, and weighed. The tissue was then divided into portions of approximately 1 g and packed into glass tubes to a height of 2 cm. The dish containing the gastric contents was then evaporated to dryness in an oven at 70°C. The waxed paper containing the dried material was cut into strips 2 cm wide. These strips were rolled up and placed in glass tubes thus maintaining a sample geometry similar to that of the gastric tissue. For each experiment, tubes representing background, gastric tissue, and gastric contents were counted for 5 min each during a single day.
VOL. 42, NO. 3, MARCH
COMPUTERIZED
IMAGE
1987
PROCESSING
The color transparencies of the gastric mucosal were mounted on a light table designed to provide uniform light transmission at its surface. The image was then recorded through a red filter by a monochrome video camera (EYECAM), connected to an image digitizer/processor (Grinnell Systems) and host (Digital Equipment Co. VAX). A portion of ulcerated mucosa as it is input through the video monitor is seen in Fig. 1. Since areas of ulceration are ssociated with hemorrhage into ulcer beds, these areas are stained dark black by the formation of acid hematin. The digitized images are then analyzed by a threshold function which converts black and dark-grey portions to pure black while lighter grey and white portions are converted to pure white. With the grey scale divided into 256 portions (0 = pure black, 255 = pure white), a threshold of 30 consistently identifies ulcerated areas. This image can also be displayed on the monitor as seen in Fig. 2. A trackball and software are used to designate the area of interest on the video image and the computer then calculates the area of ulcerated mucosa as a percentage of the total mucosa analyzed. This percentage is multiplied by the total area of the floor of the chamber yielding the area of ulceration in square centimeters. SHED
MICROSPHERE
TECHNIQUE
The shed microsphere technique served as the basis for comparison to computerized image processing. This technique has been reported previously [7, 81. In summary, radiolabeled microspheres injected into the left cardiac ventricle at a baseline period enter the circulation and are uniformly distributed throughout the gastric mucosa. When ulceration occurs these spheres are shed into the lumen along with the mucosa. Thus, the number of spheres recovered from the lumen is proportional to the volume of microsphere bearing mucosa shed. The weight of the shed mucosa is calculated by dividing the total radioactivitv in the intact mucosa bv the total
FIG. 1. Portion of ulcerated mucosa photographed through a red filter and ready for viewing by the video camera.
FIG. 2. Digitized image from Fig. 1. 279
280
JOURNAL
OF SURGICAL
RESEARCH:
weight of the intact mucosa to determine the radioactivity per unit weight. Once the total radioactivity recovered from the lumen has been determined, it is possible to calculate the weight of the ulcerated tissue from the equation: weight of ulcerated tissue =
weight of intact mucosa radioactivity of intact mucosa X radioactivity
from the lumen.
Accuracy of this method depends on reliable recovery of shed microspheres from the chamber. In pilot studies with this model in our laboratory, a known number of spheres placed in the chamber were recovered at a greater than 86% rate in each instance. For each animal the area of ulceration, as determined by computerized image processing, was paired with the weight of shed mucosa as determined by the shed microsphere technique. Linear regression of the resulting data was performed and the correlation coefficient calculated. RESULTS
After 3 hr, gastric mucosa in all five chambers exposed to normal saline alone under normotensive conditions were without visible ulceration. By contrast, the gastric mucosa in chambers exposed to acid/bile and systemic hypotension all developed visible ulceration. When the color photographs from these ulcerated stomachs were analyzed by computerized image processing the calculated area of ulceration ranged from 1.8 to 99.7 cm2. For each stomach, the area of ulceration in square centimeters as determined by computerized image processing was compared to the weight of shed mucosa in milligrams as determined by the shed microsphere technique. In each of the acid/bile/ hypotension animals, radioactivity correlating with greater than 460 spheres recovered from the gastric chamber was counted-assuring a high degree of confidence. The values derived by this latter
VOL. 42, NO. 3, MARCH
1987
method ranged from 23 to 4 19 mg. The comparison of damage estimated by the two methods showed a high degree of correlation (r = 0.99, P < 0.01) (Fig. 3) Although there was no visible ulceration on the stomachs exposed to normal saline alone some radioactivity was recovered from the chambers. However, the radioactivity counted in each of the five animals represents less than 200 spheres recovered and may well represent background alone. Therefore, shed tissue weights derived from this data in the saline group cannot be viewed as conclusive. Since these weights were negligible, the question remains moot. The calculated weight of shed mucosa for these stomachs averaged 12.1 f 4.8 mg versus an average of 143.6 f 7 1.3 mg for the stomachs subjected to acid/bile and systemic hypotension (mean + SEM, P < 0.01). DISCUSSION
Investigators studying the pathophysiology of stress-mediated gastric mucosal ulceration have long been frustrated by the absence of an objective and reproducible method of quantifying mucosal injury. The most common method used is scoring by a blinded observer using an ad hoc system. The report of Desiderato (Ref. [6]) is typical. They calculated the ulcer score by giving lesions under 1 mm a value of 1, those between
FIG. 3. Gastric mucosal ulceration in square centimepared to milligrams of ulcerated mucosa determined by the shed microsphere technique, (r = 0.99, P -e 0.01).
GASIULL
AND
LEVINE:
COMPUTERIZED
1 and 2.9 mm a value of 2 between 3 and 4.9 a value of 3, and those over 5 mm a value of 4. Only those lesions which extended to or through the submucosa of the glandular stomach were counted as ulcers. In rare cases of disagreement between judges, the mean of their two values was computed. These authors also calculated the percentage of subjects per group exhibiting ulceration and the number of ulcers per subject. By comparison, Guth and Gati (Ref. [9]) used a similar system. In their study the stomachs were opened along the greater curvature, washed, and mucosal lesions measured and scored as follows: petechiae, 1; and erosions < 1 mm, 2; l-2 mm, 3; 2-4 mm, 4; and >4 mm, 5. Although these systems are clearly workable, minor differences make comparison of results between studies difficult. In addition, there is the possibility that significant findings may be elucidated by one scale but not another. A later publication by Grossman et al. (Ref. [ 181) addressed some of these issues by defining the ulcer index. Lesions in the stomach and small intestine were measured under a dissecting microscope with a l-mm2 grid eyepiece (X 10). The ulcer index was expressed as total of the area in square millimeters of the individual lesions. Another approach to assessing mucosal injury is based on grading the severity of selected lesions. This is done by examining histologic sections with light or electron microscopy (Refs. [ 10, 11, 201). The disadvantages of this technique are several. First, it is impractical to sample and grade every ulceration in the specimen of mucosa. Thus, an arbitrary system must be devised to determine which lesions should be examined. Second, any injury seen in the sections must ultimately be quantified by a system with the same limitations of those used for intact mucosa. Many investigators have used metabolic parameters such as transmucosal electrical potential difference and electrolyte flux to assess mucosal injury (Refs. [3, 5, 14, 16, 171). The advantages of these techniques are many. They are based on well established
IMAGE
PROCESSING
281
methods and provide highly objective and quantifiable results. They are also well suited to continuous or multiple measurements throughout an experiment. Unfortunately, these methods fail to address the focal nature of mucosal ulceration. The values obtained reflect the average metabolic status of the entire mucosal surface. Since ulceration typically involves only a few percent of the surface area, significant metabolic alterations associated with these lesions may be obscured. Other methods used for assessing the metabolic status of the mucosa include measurement of acid secretion (or back diffusion), tissue gas tension, and blood flow (Refs. [ 13, 191). However, most techniques for measuring these parameters yield only an average value for the entire mucosal surface. As mentioned above, such techniques are not well suited to studies of focal lesions. Although electrical potential difference and tissue gas tension can be measured in areas as small as a single cell (Refs. [ 1, 13]), the area to be evaluated must be selected prior to actual ulceration. Furthermore, these techniques are inherently limited to sampling only a very small portion of the total mucosa. Thus, changes preceding ulceration are detected only by chance. Since all of these methods require contact with or direct puncture of the mucosal surface the possibility of altering mucosal metabolism at the area under study cannot be excluded. One very promising technique has recently been reported by Ritchie et al. (Ref. [2]). Bathing fluid from an ex vivo gastric chamber is sampled and assayed for mucosal efflux of DNA. Since DNA is located primarily in the nucleus of the gastric mucosal cells, accumulation of DNA in the gastric lumen is roughly proportional to cellular loss. One of the advantages of this system is that it should be able to detect even minimal mucosal injury. The most important drawback of this method is that most common methods of assaying DNA are not accurate when the samples are contaminated with gastric secretions-notably sialic acid (Ref.
282
JOURNAL
OF
SURGICAL
RESEARCH:
[4]). Accordingly, techniques requiring the use of expensive spectrofluorometric equipment are necessary. Additional errors may be caused by destruction of the DNA by DNase, an enzyme ubiquitous in animal tissues. Finally, the processing of tissue and fluids for DNA measurement involves multiple, time consuming steps. Thus, it is not ideally suited to the processing of large numbers of samples. Another method which has been used to evaluate mucosal architecture is microangiography using agents such as stained hemoglobin or fine-grain barium sulfate (Refs. [ 12, 131). These techniques reveal changes in the microvascular structure of the tissue which may be associated with the mucosal stress response-including ulceration. The primary disadvantage of microangiography is that only one assessment is possible for each section of tissue. Furthermore, changes observed are not readily quantifiable by any well standardized and objective technique. The technique reported herein, computerized image processing, has many advantages when compared to these other methods. First, it is highly objective. The major subjective determination is the image density threshold between ulcerated mucosa and nonulcerated mucosa. In practice, however, the density gradient between acid hematinstained ulceration and normal or ischemic mucosa is fairly steep. Thus, large changes in the threshold value result in relatively small changes in the area of ulceration perceived by the system. The gradient can be increased even more by using a red filter between the input video camera and the transparency of the mucosa to be analyzed. As with any experimental technique, consistency and attention to detail are necessary for reproducible results. Exposure conditions must be carefully controlled. Uniform biplane illumination is necessary to avoid shadowing by mucosal folds. Shadows may be interpreted by the system as ulceration, thus overestimating the area of injury. When data from several experiments are to be compared, it is important to record film
VOL.
42, NO.
3, MARCH
1987
type, shutter speed, F-stop, and the distance from camera to mucosa. In practice, these factors are easily controlled by mounting the camera and lights in a frame above the operating table. There are some disadvantages to computerized image processing. Cost is certainly a significant factor. The computer and video hardware described under Methods section has been in use for several years and cost in excess of $100,000 at the time it was originally purchased. Today, all of the necessary equipment would cost less than $lO,OOOincluding the computer. Investigators with access to a computer could set up a workable system for three to five thousand dollars. While this is certainly more expensive than viewing and grading the injury by traditional methods, it is not unreasonable. One serious limitation of computerized image processing is that it cannot assess the depth of the lesions. A lesion involving only the superficial mucosa is interpreted exactly the same as a lesion penetrating to the muscularis. In addition, lesions which do not bleed will be missed entirely. Since this system is based on the detection of the black acid hematin residue from mucosal hemorrhage, care is required to avoid disruption of this material. Although it is possible to recover stomachs from intact animals, this may result in disruption of the clots. The ideal preparation is the isolated mucosal chamber model described by Moody et al. (Ref. [ 151). This model facilitates consistent photographic technique as well as serial photographs during the experiment. Another limitation is the uneven nature of the gastric mucosa. Lesions which are on the side of a mucosal fold will be viewed at an angle by the camera, thus reducing the apparent area of the ulcer. Although this effect can be reduced by mounting the mucosa carefully in the chamber, it cannot be eliminated entirely. In summary, computerized image processing has many advantages over traditional methods of assessing mucosal ulceration. Multiple mesurements can be made during
GASKILL
AND LEVINE:
COMPUTERIZED
an experiment. The data produced is consistent and highly reproducible. The necessary equipment is available at reasonable cost. However, it is best suited for use in open systems such as the isolated gastric chamber. REFERENCES 1. Bowen, J. C., and Gang, D. K. Effect of graded mechanical ischemia on oxygen tension and electrical potential in the canine gastric mucosa. Castroenterology 73: 84, 1977. 2. Carter, K. J., Farley, P. C., and Ritchie, W. P. Effect of topi.cal bile acids on gastric surface epithelial cells. Surgery 96: 196, 1984. 3. Chaudhury, T. K., and Jacobson, E. D. Prostaglandin cytoprotection of gastric mucosa. GastroenteroIogy 74: 59, 1978. 4. Croft, D. N., and Lubran, M. The estimation of deoxyribonucleic acid in the presence of sialic acid: Application to analysis of human gastric washings. B&hem. J. 95: 612, 1965. 5. Davenport, H. W. Gastric mucosal injury by fatty and acetylsalicylic acids. Gastroenterology 46: 245, 1964. 6. Desiderato, O., MacKinnon, J. R., and Hissom, H. Development of gastric ulcers in rats following stress termination. J. Comp. Physiol. Psychol. 87: 208, 1974. 7. Gaskill, H. V., III, Sirinek, K. R., and Levine, B. A. 16,16-Dimethyl prostaglandin E2 reverses focal mucosal &hernia associated with stress ulcers. J. Surg. Res. 37: 83, 1984. 8. Gaskill, H. V., III, Sirinek, K. R., and Levine, B. A. Focal mucosal &hernia is a precursor to stress-mediated gastric mucosal ulceration. Surg. Gastroenterol. 2: 225, 1983. 9. Gati, T., and Guth, P. H. Mucosal lesions due to gastric distension in the rat. Dig. Dis. Sci. 22: 1083, 1977. 10. Goodman, A. A., and Osborne, M. P. An experimental model and clinical definition of stress ulceration. Surg. Gynecol. Obstet. 134: 563, 1972.
IMAGE
PROCESSING
283
11. Guth, P. H., Paulsen, G., and Foroozan, P. Experi-
mental chronic gastric ulcer due to &hernia in rats. 12 Dig. Dis. Sci. 20: 824, 1975. Guth, P. H., and Hall, P. Microcirculatory and mast ’ cell changes in restraint-induced gastric ulcer. Gustroenteroiogy 50: 562, 1966. 13. Kivilaakso, E., Ahonen, J., Aronsen, K.-F., Hockerstedt, K., Kalima, T., Lempinen, M., Suoranta, H., and Vemerson, E. Gastric blood flow, tissue gas tension and microvascular changes during hemorrhage-induced stress ulceration in the pig. Amer. J. Surg. 143: 322, 1982. 14. Konturek, S. J., Tasler, J., Obtulowicz, W., Kwiecien, N., Sito, E., and Oleksy, J. Effect of methylated prostaglandin E2 analogue on canine and human gastric mucosal barrier. Gastroenterol. Clin. Biol. 2: 177, 1978. 15. Moody, F. G., Cheung, L. Y., Simons, M. A., and Zalewsky, C. Stress and the acute gastric mucosal lesion. Dig. Dis. Sci. 21: 148, 1976. 16. Ritchie, W. P., Jr., Cherry, K. J., Jr., and Gibb, A. Influence of methylprednisolone sodium succinate on bile-acid-induced acute gastric mucosal damage. Surgery 84: 283, 1978. 17. Ritchie, W. P., Jr., Shearbum, E. W., III, and Nading, A. Relationship of transmural electrical potential difference to changes in gastric mucosal permeability to H+ and blood flow. Amer. J. Surg. 135: 110, 1978. 18. Satoh, H., Guth, P. H., and Grossman, M. I. Role of bacteria in gastric ulceration produced by indomethacin in the rat: cytoprotective action of antibiotics. Gastroenterology 84: 483, 1983. 19. Shirazi, S. S., Mueller, T. M., and Hardy, B. M. Canine gastric acid secretion and blood flow measurement in hemorrhagic shock. Gastroenterology 73: 75, 1977. 20. Svanes, K., Ito, S., Takeuchi, K., and Silen, W. Restitution of the surface epithelium of the in vitro frog gastric mucosa after damage with hyperosmolar sodium chloride. Morphologic and physiologic characteristics. Gastroenterology 82: 1409, 1982.