- Email: [email protected]
Animal study of renal volume measurement on abdominal CT using digital image processing
Journal of Clinical Imaging 28 (2004) 135 – 137
Animal study of renal volume measurement on abdominal CT using digital image processing Preliminary report Jong Chul Kim* Department of Diagnostic Radiology, Chungnam National University Hospital, 640 Daesa-dong, Jung-gu, Daejeon 301-721, Republic of Korea Received 28 February 2003
Abstract On abdominal computed tomography (CT) scans of two anesthetized Landrace pigs with 3-mm slice interval, kidneys were extracted by two steps of digital image processing: automatic segmentation in a single slice using the character of pixel distribution of the kidney, and removal of residual noise using batch processing with folding method. The measured renal volume by this method was compared with the actual renal volume obtained by means of three consecutive water displacement measurements on surgically removed kidneys. The mean percentage error was 2.9% between mean renal volume measured by our image processing (78.3 and 67.4 cm3 for right and left, respectively) and mean actual renal volume (80 and 70 cm3, respectively). Renal volume measurement on abdominal CT in pigs using this digital image processing was feasible and reliable with negligible error rate in comparison with actual renal volume. D 2004 Elsevier Inc. All rights reserved. Keywords: Kidney experimental studies; Kidney image processing; Kidney CT
1. Introduction Many clinicians or patients are interested in the size of the kidney and significant change of renal volume in clinical fields. Enlargement of the kidney is caused by various conditions or disorders; such as hyperplasia, associated agenesis or hypoplasia on the opposite site, compensatory hypertrophy, obstructive hydronephrosis, polycystic disease, other cystic or dysplastic disorders, neoplasm, renal vein thrombosis, acute infection, Waldenstro¨m’s macroglobulinemia, hemophilia, acute arterial infarction, and duplication of the renal pelvis [1]. Conditions that characteristically cause bilateral renal enlargement include acute glomerulonephritis, lymphoma, leukemia in children, systemic lupus crythematosus, amyloidosis, sarcoidosis, sickle cell disease, lipod nephrosis, lobar glomerulonephritis, glycogen storage disease, hereditary tyrosinemia, and total lipodystrophy [2].
To detect the change of the kidney size or volume, various cross-sectional imaging modalities such as ultrasonography, computed tomography (CT), or magnetic resonance image have been used. To our knowledge, however, there have not been many articles reporting volume determination or computation of the kidney using CT [3 –5]. In 2000, we reported a digital image processing method of renal segmentation without using contrast media on abdominal CT image [6]. To apply this method widely and conveniently in clinical fields, the accuracy and usefulness of this method in renal volume measurement should be proved in the preceding experimental animal study. This study was performed to evaluate the efficacy and accuracy of our digital image processing method in the determination of renal volume of the pigs on abdominal CT.
2. Materials and methods * Tel.: +82-42-220-7835; fax: +82-42-253-0061. E-mail address: [email protected] (J.C. Kim). 0899-7071/04/$ – see front matter D 2004 Elsevier Inc. All rights reserved. doi:10.1016/S0899-7071(03)00115-3
Two 4-month-old Landrace pigs (24 kg of body weight) were anesthetized with Rumpum (xylazine 1 mg/kg, Korea
136
J.C. Kim / Journal of Clinical Imaging 28 (2004) 135–137 Table 1 Comparison of kidney volume of two pigs: volume by digital image processing vs. actual volume
Fig. 1. Unenhanced abdominal CT scans of a 4-month-old Landrace pig (upper raw) and their extracted images (lower raw). Both kidneys on CT were extracted by two steps of image processing: automatic segmentation in a single slice using the character of pixel distribution of the kidney, and removal of residual noise using batch processing with folding method.
Pfizer, Seoul) after premedication of Combelen (acetromazine 0.2 mg/Kg, Korea Bayer, Seoul). After proper setting of anesthetized pigs on the CT scan table, unenhanced abdominal CT scan with thin sections of 3-mm collimation was performed, using CTi Standard (General Electric Medical Systems, Milwaukee, WI). The kidneys of the pigs on CT were extracted by two steps of digital image processing: (a) Single slice processing using the character of pixel distribution of the kidney, and (b) removal of residual noise using batch processing with folding method. Single slice processing was accomplished by peak search, binary image, ratio calculation, mesh generation, pixel trace, template generation, ratio extraction, and filling. Batch processing was performed through the processes of slice folding, template generation,
Site of kidney
Volume by digital image processing (cm3)
Actual volume (cm3)
Error (%)
Right Left Mean
78.2 and 78.4 (mean 78.3) 67.9 and 66.9 (mean 67.4)
79.2 and 80.4 (mean 80) 72.1 and 67.9 (mean 70)
2.1 3.7 2.8
and noise removal. Because the template scope of opening is larger than the volume of kidney, extraction method was also used to remove the noise. After obtaining extracted renal images of each slice in right and left kidneys (Fig. 1), the volume of extracted kidney was calculated by the following equation [6]. N 1 X ðððLp X Y Þ of Si þ ðLp X Y Þ of Siþ1 Þ=2Þ D i¼1
N: the number of slices including the extracted organ; Is: slice number; D: length of slice interval; Lap: the number of pixels composing extracted organ; X: horizontal length of one pixel; Y: Vertical length of one pixel. After CT scanning, the pigs were sacrificed, and both kidneys were removed surgically in animal operating room (Fig. 2). Actual renal volume of the pigs was obtained by means of three consecutive water displacement measurements on surgically removed specimen of both kidneys. Then the measured renal volume by our digital image processing method was compared with the actual renal volume of the pigs.
3. Results The results of comparison of the renal volumes determined by this method with actual renal volume of the pigs were summarized in Table 1. In our study, as a result of comparison between the mean renal volume of both kidneys in two pigs measured by digital image processing (78.3 and 67.4 cm3; right and left, respectively) and mean actual renal volume (80 and 70 cm3, respectively), the mean percentage error was 2.8%.
4. Discussion
Fig. 2. A photograph of surgically excised specimens of both kidneys in a 4-month-old Landrace pig. The actual renal volume was obtained by means of three consecutive water displacement measurements on these renal specimens.
CT potentially offers the most accurate noninvasive means of estimating in vivo volumes. A new digital image processing method of renal segmentation on unenhanced abdominal CT image was presented by us in 2000 [6]. Our method has two steps of image processing: (a) Single slice processing using the character of pixel distribution of the kidney, and (b) removal of residual noise using batch processing with folding method.
J.C. Kim / Journal of Clinical Imaging 28 (2004) 135–137
Before practical and easy application of this digital image processing in daily clinical fields, we performed experimental animal study to estimate the accuracy and usefulness of this method in renal volume measurement on abdominal CT in two 4-month old pigs. The result in our study revealed that the mean percentage error of the renal volume of both kidneys of two pigs measured by digital image processing and actual renal volume was less than 3.7%. The extraction method in our study was based on the difference of density of the other organs. Therefore, it was not easy or simple to extract only the renal parenchyma from surrounding solid organs with densities similar to those of kidneys on CT. Because of the fact that the right kidney abuts the liver and left kidney is near to the spleen on CT scan in the anterior – posterior direction, partial volume effect can also be produced. In fact, there was difficulty in detection of anatomical position for automatic segmentation and noise removal in our study, because the size of the kidney was smaller than the size of the adjacent large organs such as liver or spleen. Especially in binary image of the kidney, it was often difficult to differentiate the kidney and hole during the course of mesh image formation. Considering the relative small size of the kidney, similar density to adjacent liver or spleen, advertent overlapping of mesh and pixel, and possible partial volume effect on CT, the mean percentage error of 2.8% is considered to be tolerable and negligible. Breiman et al. [7] performed contiguous 1-cm-thick CT scanning in dog kidneys in vivo, and reported that the mean percentage error of volume calculations using the sum-of-areas technique was 3.86% for eight dog kidneys. Brenner et al. [4] reported that calculated CT volumes of cadaver kidneys and spleens were within F 10% of directly measured volumes, with accuracy affected by respiratory movement, rapid change in vivo blood volume, low CT number-gradient at the object’s periphery, observer error in cursor tracing of the desired structure, and mathematical errors inherent in Simpson’s rule. Compared with the results of above two articles, 2.8% discrepancy rate in our study was estimated to be more relevant and reliable. Lerman et al. [8] accomplished contrast enhanced cine CT scans in 14 anesthetized dogs and the volume of right kidneys was determined after boundary identification, and
137
the mean renal volumes ( F S.E. of the mean) in vivo (78.2 F 2.4 cc) were compared with those of excised postmortem subjects (66.1 F 2.2 cc) determined by fluid displacement (r = 0.86; P < .001). If we also use cine CT scan, our results would be better than before. Our study, however, has critical limitation that the sample size (i.e., two) is too small to be statistically significant in scientific thesis. Another limitation is based on the fact that the volume of surgically removed pig kidney is not exactly equal to the volume of pig kidney in vivo. The difference may be due to the blood, filtrate, and urine contents of the in vivo kidney. In conclusion, automatic segmentation and volume measurement of the kidney using our digital image processing on unenhanced abdominal CT in two pigs were feasible, reliable and useful. The renal volume determined by this digital image processing was almost as same as the actual renal volume. The method used in this study will be clinically applicable in clinical fields to determine renal volume both in normal persons and patients.
References [1] Lalli AF. Renal enlargement. Radiology 1965;84:688 – 94. [2] Thornbury JR, Weiss SL. The urinary tract. In: Amberg JR, Kelsey CA, Peters ME, Rogers LF, Sackett JF, Stropher CM, Thornbury JR, Unger JM, Weiss SL, Wicks JD, editors. Paul and Juhl’s essentials of radiologic imaging. Philadelphia (PA): JB Lippincott, 1987. pp. 581 – 2. [3] Heymsfield SB, Fulenwider T, Nordlinger B, Sones P, Kutner N. Accurate measurement of liver, kidney, and spleen, volume, and mass by computerized axial tomography. Ann Intern Med 1979;90:185 – 7. [4] Brenner DE, Whtley NO, Houk TL, Aisner J, Wiernik P, Whitley J. Volume determination in computed tomography. JAMA 1982;247: 1299 – 302. [5] Moss AA, Froedman MA, Brito AC. Determination of liver, kidney, and spleen volumes by computed tomography. J Comput Assist Tomogr 1981;5:12 – 4. [6] Kim SH, You SW, Kim SJ, Kim JC, Park JW. Segmentation of kidney without using contrast medium on abdominal CT image. 5th International Conference on Signal Processing Proceedings 2000;147 – 52. [7] Breiman RS, Beck JW, Korobkin M, Glenny R, Akwari OE, Heaston DK, Moore AV, Ram PC. Volume determination using computed tomography. Am J Roentgenol 1982;138:329 – 33. [8] Lerman LO, Bentley MD, Bell MR, Rumberger JA, Romero JC. Quantitation of the in vivo kidney volume with cine computed tomography. Invest Radiol 1990;25:1206 – 11.
Animal study of renal volume measurement on abdominal CT using digital image processing Preliminary report Jong Chul Kim* Department of Diagnostic Radiology, Chungnam National University Hospital, 640 Daesa-dong, Jung-gu, Daejeon 301-721, Republic of Korea Received 28 February 2003
Abstract On abdominal computed tomography (CT) scans of two anesthetized Landrace pigs with 3-mm slice interval, kidneys were extracted by two steps of digital image processing: automatic segmentation in a single slice using the character of pixel distribution of the kidney, and removal of residual noise using batch processing with folding method. The measured renal volume by this method was compared with the actual renal volume obtained by means of three consecutive water displacement measurements on surgically removed kidneys. The mean percentage error was 2.9% between mean renal volume measured by our image processing (78.3 and 67.4 cm3 for right and left, respectively) and mean actual renal volume (80 and 70 cm3, respectively). Renal volume measurement on abdominal CT in pigs using this digital image processing was feasible and reliable with negligible error rate in comparison with actual renal volume. D 2004 Elsevier Inc. All rights reserved. Keywords: Kidney experimental studies; Kidney image processing; Kidney CT
1. Introduction Many clinicians or patients are interested in the size of the kidney and significant change of renal volume in clinical fields. Enlargement of the kidney is caused by various conditions or disorders; such as hyperplasia, associated agenesis or hypoplasia on the opposite site, compensatory hypertrophy, obstructive hydronephrosis, polycystic disease, other cystic or dysplastic disorders, neoplasm, renal vein thrombosis, acute infection, Waldenstro¨m’s macroglobulinemia, hemophilia, acute arterial infarction, and duplication of the renal pelvis [1]. Conditions that characteristically cause bilateral renal enlargement include acute glomerulonephritis, lymphoma, leukemia in children, systemic lupus crythematosus, amyloidosis, sarcoidosis, sickle cell disease, lipod nephrosis, lobar glomerulonephritis, glycogen storage disease, hereditary tyrosinemia, and total lipodystrophy [2].
To detect the change of the kidney size or volume, various cross-sectional imaging modalities such as ultrasonography, computed tomography (CT), or magnetic resonance image have been used. To our knowledge, however, there have not been many articles reporting volume determination or computation of the kidney using CT [3 –5]. In 2000, we reported a digital image processing method of renal segmentation without using contrast media on abdominal CT image [6]. To apply this method widely and conveniently in clinical fields, the accuracy and usefulness of this method in renal volume measurement should be proved in the preceding experimental animal study. This study was performed to evaluate the efficacy and accuracy of our digital image processing method in the determination of renal volume of the pigs on abdominal CT.
2. Materials and methods * Tel.: +82-42-220-7835; fax: +82-42-253-0061. E-mail address: [email protected] (J.C. Kim). 0899-7071/04/$ – see front matter D 2004 Elsevier Inc. All rights reserved. doi:10.1016/S0899-7071(03)00115-3
Two 4-month-old Landrace pigs (24 kg of body weight) were anesthetized with Rumpum (xylazine 1 mg/kg, Korea
136
J.C. Kim / Journal of Clinical Imaging 28 (2004) 135–137 Table 1 Comparison of kidney volume of two pigs: volume by digital image processing vs. actual volume
Fig. 1. Unenhanced abdominal CT scans of a 4-month-old Landrace pig (upper raw) and their extracted images (lower raw). Both kidneys on CT were extracted by two steps of image processing: automatic segmentation in a single slice using the character of pixel distribution of the kidney, and removal of residual noise using batch processing with folding method.
Pfizer, Seoul) after premedication of Combelen (acetromazine 0.2 mg/Kg, Korea Bayer, Seoul). After proper setting of anesthetized pigs on the CT scan table, unenhanced abdominal CT scan with thin sections of 3-mm collimation was performed, using CTi Standard (General Electric Medical Systems, Milwaukee, WI). The kidneys of the pigs on CT were extracted by two steps of digital image processing: (a) Single slice processing using the character of pixel distribution of the kidney, and (b) removal of residual noise using batch processing with folding method. Single slice processing was accomplished by peak search, binary image, ratio calculation, mesh generation, pixel trace, template generation, ratio extraction, and filling. Batch processing was performed through the processes of slice folding, template generation,
Site of kidney
Volume by digital image processing (cm3)
Actual volume (cm3)
Error (%)
Right Left Mean
78.2 and 78.4 (mean 78.3) 67.9 and 66.9 (mean 67.4)
79.2 and 80.4 (mean 80) 72.1 and 67.9 (mean 70)
2.1 3.7 2.8
and noise removal. Because the template scope of opening is larger than the volume of kidney, extraction method was also used to remove the noise. After obtaining extracted renal images of each slice in right and left kidneys (Fig. 1), the volume of extracted kidney was calculated by the following equation [6]. N 1 X ðððLp X Y Þ of Si þ ðLp X Y Þ of Siþ1 Þ=2Þ D i¼1
N: the number of slices including the extracted organ; Is: slice number; D: length of slice interval; Lap: the number of pixels composing extracted organ; X: horizontal length of one pixel; Y: Vertical length of one pixel. After CT scanning, the pigs were sacrificed, and both kidneys were removed surgically in animal operating room (Fig. 2). Actual renal volume of the pigs was obtained by means of three consecutive water displacement measurements on surgically removed specimen of both kidneys. Then the measured renal volume by our digital image processing method was compared with the actual renal volume of the pigs.
3. Results The results of comparison of the renal volumes determined by this method with actual renal volume of the pigs were summarized in Table 1. In our study, as a result of comparison between the mean renal volume of both kidneys in two pigs measured by digital image processing (78.3 and 67.4 cm3; right and left, respectively) and mean actual renal volume (80 and 70 cm3, respectively), the mean percentage error was 2.8%.
4. Discussion
Fig. 2. A photograph of surgically excised specimens of both kidneys in a 4-month-old Landrace pig. The actual renal volume was obtained by means of three consecutive water displacement measurements on these renal specimens.
CT potentially offers the most accurate noninvasive means of estimating in vivo volumes. A new digital image processing method of renal segmentation on unenhanced abdominal CT image was presented by us in 2000 [6]. Our method has two steps of image processing: (a) Single slice processing using the character of pixel distribution of the kidney, and (b) removal of residual noise using batch processing with folding method.
J.C. Kim / Journal of Clinical Imaging 28 (2004) 135–137
Before practical and easy application of this digital image processing in daily clinical fields, we performed experimental animal study to estimate the accuracy and usefulness of this method in renal volume measurement on abdominal CT in two 4-month old pigs. The result in our study revealed that the mean percentage error of the renal volume of both kidneys of two pigs measured by digital image processing and actual renal volume was less than 3.7%. The extraction method in our study was based on the difference of density of the other organs. Therefore, it was not easy or simple to extract only the renal parenchyma from surrounding solid organs with densities similar to those of kidneys on CT. Because of the fact that the right kidney abuts the liver and left kidney is near to the spleen on CT scan in the anterior – posterior direction, partial volume effect can also be produced. In fact, there was difficulty in detection of anatomical position for automatic segmentation and noise removal in our study, because the size of the kidney was smaller than the size of the adjacent large organs such as liver or spleen. Especially in binary image of the kidney, it was often difficult to differentiate the kidney and hole during the course of mesh image formation. Considering the relative small size of the kidney, similar density to adjacent liver or spleen, advertent overlapping of mesh and pixel, and possible partial volume effect on CT, the mean percentage error of 2.8% is considered to be tolerable and negligible. Breiman et al. [7] performed contiguous 1-cm-thick CT scanning in dog kidneys in vivo, and reported that the mean percentage error of volume calculations using the sum-of-areas technique was 3.86% for eight dog kidneys. Brenner et al. [4] reported that calculated CT volumes of cadaver kidneys and spleens were within F 10% of directly measured volumes, with accuracy affected by respiratory movement, rapid change in vivo blood volume, low CT number-gradient at the object’s periphery, observer error in cursor tracing of the desired structure, and mathematical errors inherent in Simpson’s rule. Compared with the results of above two articles, 2.8% discrepancy rate in our study was estimated to be more relevant and reliable. Lerman et al. [8] accomplished contrast enhanced cine CT scans in 14 anesthetized dogs and the volume of right kidneys was determined after boundary identification, and
137
the mean renal volumes ( F S.E. of the mean) in vivo (78.2 F 2.4 cc) were compared with those of excised postmortem subjects (66.1 F 2.2 cc) determined by fluid displacement (r = 0.86; P < .001). If we also use cine CT scan, our results would be better than before. Our study, however, has critical limitation that the sample size (i.e., two) is too small to be statistically significant in scientific thesis. Another limitation is based on the fact that the volume of surgically removed pig kidney is not exactly equal to the volume of pig kidney in vivo. The difference may be due to the blood, filtrate, and urine contents of the in vivo kidney. In conclusion, automatic segmentation and volume measurement of the kidney using our digital image processing on unenhanced abdominal CT in two pigs were feasible, reliable and useful. The renal volume determined by this digital image processing was almost as same as the actual renal volume. The method used in this study will be clinically applicable in clinical fields to determine renal volume both in normal persons and patients.
References [1] Lalli AF. Renal enlargement. Radiology 1965;84:688 – 94. [2] Thornbury JR, Weiss SL. The urinary tract. In: Amberg JR, Kelsey CA, Peters ME, Rogers LF, Sackett JF, Stropher CM, Thornbury JR, Unger JM, Weiss SL, Wicks JD, editors. Paul and Juhl’s essentials of radiologic imaging. Philadelphia (PA): JB Lippincott, 1987. pp. 581 – 2. [3] Heymsfield SB, Fulenwider T, Nordlinger B, Sones P, Kutner N. Accurate measurement of liver, kidney, and spleen, volume, and mass by computerized axial tomography. Ann Intern Med 1979;90:185 – 7. [4] Brenner DE, Whtley NO, Houk TL, Aisner J, Wiernik P, Whitley J. Volume determination in computed tomography. JAMA 1982;247: 1299 – 302. [5] Moss AA, Froedman MA, Brito AC. Determination of liver, kidney, and spleen volumes by computed tomography. J Comput Assist Tomogr 1981;5:12 – 4. [6] Kim SH, You SW, Kim SJ, Kim JC, Park JW. Segmentation of kidney without using contrast medium on abdominal CT image. 5th International Conference on Signal Processing Proceedings 2000;147 – 52. [7] Breiman RS, Beck JW, Korobkin M, Glenny R, Akwari OE, Heaston DK, Moore AV, Ram PC. Volume determination using computed tomography. Am J Roentgenol 1982;138:329 – 33. [8] Lerman LO, Bentley MD, Bell MR, Rumberger JA, Romero JC. Quantitation of the in vivo kidney volume with cine computed tomography. Invest Radiol 1990;25:1206 – 11.