|Year : 2016 | Volume
| Issue : 1 | Page : 20-26
Digital contrast enhancement of 18 Fluorine-fluorodeoxyglucose positron emission tomography images in hepatocellular carcinoma
Anil Kumar Pandey, Sanjay Kumar Sharma, Krishan Kant Agarwal, Punit Sharma, Chandrasekhar Bal, Rakesh Kumar
Department of Nuclear Medicine, All India Institute of Medical Sciences, New Delhi, India
|Date of Web Publication||21-Dec-2015|
Department of Nuclear Medicine, All India Institute of Medical Sciences, E-81, Ansari Nagar (East), AIIMS Campus, New Delhi - 110 029
Source of Support: None, Conflict of Interest: None
| Abstract|| |
Purpose: The role of 18 fluorodeoxyglucose positron emission tomography (PET) is limited for detection of primary hepatocellular carcinoma (HCC) due to low contrast to the tumor, and normal hepatocytes (background). The aim of the present study was to improve the contrast between the tumor and background by standardizing the input parameters of a digital contrast enhancement technique. Materials and Methods: A transverse slice of PET image was adjusted for the best possible contrast, and saved in JPEG 2000 format. We processed this image with a contrast enhancement technique using 847 possible combinations of input parameters (threshold "m" and slope "e"). The input parameters which resulted in an image having a high value of 2 nd order entropy, and edge content, and low value of absolute mean brightness error, and saturation evaluation metrics, were considered as standardized input parameters. The same process was repeated for total nine PET-computed tomography studies, thus analyzing 7623 images. Results: The selected digital contrast enhancement technique increased the contrast between the HCC tumor and background. In seven out of nine images, the standardized input parameters "m" had values between 150 and 160, and for other two images values were 138 and 175, respectively. The value of slope "e" was 4 in 4 images, 3 in 3 images and 1 in 2 images. It was found that it is important to optimize the input parameters for the best possible contrast for each image; a particular value was not sufficient for all the HCC images. Conclusion: The use of above digital contrast enhancement technique improves the tumor to background ratio in PET images of HCC and appears to be useful. Further clinical validation of this finding is warranted.
Keywords: 18-fluorin-fluorodeoxyglucose, contrast enhancement, hepatocellular carcinoma, image, positron emission tomography
|How to cite this article:|
Pandey AK, Sharma SK, Agarwal KK, Sharma P, Bal C, Kumar R. Digital contrast enhancement of 18 Fluorine-fluorodeoxyglucose positron emission tomography images in hepatocellular carcinoma. Indian J Nucl Med 2016;31:20-6
|How to cite this URL:|
Pandey AK, Sharma SK, Agarwal KK, Sharma P, Bal C, Kumar R. Digital contrast enhancement of 18 Fluorine-fluorodeoxyglucose positron emission tomography images in hepatocellular carcinoma. Indian J Nucl Med [serial online] 2016 [cited 2019 Dec 12];31:20-6. Available from: http://www.ijnm.in/text.asp?2016/31/1/20/172346
| Introduction|| |
Hepatocellular carcinoma (HCC) is a primary liver cancer derived from hepatocytes.  In more than 80% cases, it develops in association with the liver cirrhosis.  HCC is curable, if detected at an early stage by any surveillance programs. ,,, It can be treated with liver transplantation, resection, or radio frequency ablation. The early detection of HCC has been attempted using 18 fluorodeoxyglucose positron emission tomography (FDG PET). 18 FDG PET exploits the metabolic differences between benign and malignant cells. The accumulation of 18 fluorin-FDG ( 18 F-FDG) in HCC is variable due to the varying degrees of activity of the enzyme "glucose-6-phosphatase" in these tumors. ,,,, Well-differentiated HCC, and low grades tumor have more or less similar degree of activity as compared to the normal liver cells, and thus results in small difference in 18 F-FDG accumulation, and therefore, 18 F-FDG PET picks up only 50-70% of the lesion in these patients. 
It is possible to increase the contrast between the HCC cells, and hepatocytes on 18 F-FDG PET using digital contrast enhancement techniques. Almost all contrast enhancement technique increases the contrast of the targeted lesion at the cost of decreasing contrast in other areas of the image. It is a challenging task to find the optimum balance between the acceptable enhancements of the targeted lesion while maintaining the existing contrast in other areas of the image. Very few digital contrast enhancement techniques have been tried to improve the low contrast in HCC. We conducted a prototype study to select one technique from a set of contrast enhancement techniques available in the literature.  During the prototype study, we realized that many techniques can serve our purpose. However, there is need to standardize the input parameters of any selected technique for the best result. In this study, we selected a contrast enhancement technique that increases the contrast in both the darker, and brighter region of the image simultaneously,  and optimized the value of its input parameters for the best possible contrast for 18 F-FDG PET images in HCC.
| Materials and methods|| |
This was a retrospective study. No ethical approval was required for the study involved in the digital processing of PET images acquired for a clinical indication.
Positron emission tomography-computed tomography acquisition protocol
Patients fasted for at least 4 h before undergoing 18 F-FDG PET-computed tomography (PET-CT). Their blood glucose level was ≤7.7 mmol/l prior to the intravenous injection of 370 MBq of 18 F-FDG. The data acquisition were performed at 45-60 min later with an integrated PET-CT system (Biograph mCT, Siemens Medical Solutions, Erlangen, Germany). No intravenous contrast was used. CT was acquired with covering the patient from skull base to mid-thigh which was matched to PET bed size. Immediately after CT scanning, PET acquisition was performed on covering the same field of view. PET acquisition was done for 2 min/bed, and on a matrix size of 128 × 128. Images were reconstructed using the ordered subset expectation maximization algorithm with 2 iterations, and 8 subsets, and full width half maximum of 5 mm. CT-based attenuation correction of the PET images was applied. Finally, the attenuation corrected PET, CT and fused PET-CT images were available for review.
Digital image enhancement
An experienced nuclear medicine physician adjusted the contrast of a transverse slice of 18 F-FDG PET image containing HCC tumor with the help of a most widely used brightness-contrast tool. Once the best possible contrast in the image in his opinion was achieved, then the slice was saved in the JPEG 2000 format. A total of nine 18 F - FDG PET studies referred the case of suspected HCC were selected, and hence nine JPEG images were saved [Figure 1], and their 2 nd order entropy values are given in [Table 1].
Digital contrast enhancement of these images was carried by standardizing the value of the parameter ("m" and "e") of the contrast enhancement function T (r) represented by equation (1).
In the equation (1), "r0" is the numerical value stored at a pixel location in the image, "m" is the threshold, and "e" determines the slope of the linear portion of the curve represented by the equation (1). The transform function T (r) compresses the input levels below, and above "m" into a narrow range of darker and brighter levels, respectively, and thereby improves contrast.
JPEG image in RGB format were converted into the gray level, and then the image data were converted into double data type before applying equation (1). We have used im2double, an inbuilt function available in the MATLAB, to convert the image to double data type. We made 847 possible combinations of input parameters by varying "m" from 60 to 180 in the steps of 1 and for each value of "m," the value of "e" was varied from 1 to 7 again in the steps of 1. For all 847 combinations, the image was processed. The same process was repeated for all the nine images and in this way 9 × 847 = 7623 images were visually, and quantitatively analyzed for the global contrast enhancement with respect to the original input image. The global contrast enhancement of these images was assessed by the following contrast evaluation metrics:
2 nd order entropy
We calculated 2 nd order entropy based on co-occurrence matrix  using the distance of one pixel in the horizontal direction. 2 nd order entropy gives the information about the image richness in detail taking the spatial relation of the pixel value into account.
Edge content based contrast metrics
We calculated the edge at each pixel position using the standard MATLAB edge function, and Sobel operator, and edge content (EC) by dividing the sum of the value of edge by the matrix size of the image.  The performance of Canny operator have been reported as the best, however, computationally most expensive among other such edge detection operators such as Sobel, Laplacian, and Laplacian of Gaussian. Otherwise, in case of smooth images, Sobel gives much faster results similar to Canny operator. The input image provided by the clinical team was the best images from their end; it was the smooth image as can be seen from the [Figure 1]. Therefore, we used Sobel operator to find the EC in the image.
Absolute mean brightness error
We calculated the value of absolute mean brightness error (AMBE), which is the deviation of the mean intensity of the enhanced image from the mean intensity of the original image by finding the absolute difference between the mean intensity of the output, and the input image.  The mean intensity was calculated as the sum of all pixel values divided by a number of pixels in the image.
Saturation evaluation metric
Saturation evaluation metric (SEM) measures the saturation by computing the number of saturated pixels (black or white pixels which were not saturated before) after applying contrast enhancement,  we calculated the value of saturation metric by dividing the number of saturated pixels (black or white pixels which were not saturated before) by the size of the matrix that is number of pixels in the image.
The calculated values of 2 nd order entropy, ABME, EC, and SEM for each of the 7623 images were stored. We defined criteria for the selection of the image having best possible contrast among the 847 images. We defined the best possible contrast image as the image whose 2 nd order entropy, and EC, have high values, while AMBE and SEM, have low values, respectively. The input parameter ("m" and "e") that resulted in best possible contrast image was considered as standardized input parameters of the contrast enhancement function. The plot of 2 nd order entropy, EC, AMBE, and SEM versus output image parameters ("m" and "e") was carefully analyzed to locate the approximate position on the plot for best possible contrast image. In next step, we drew a separate plot by taking ± 10 data points around this point and determined the standardized parameter.
| Results|| |
For one image, we had 847 data point to be plotted, and there were nine images. For each data points, there were four values namely: 2 nd order entropy, EC, AMBE, and SEM. The plotted data of a representative image appears as an envelope depicting the trend, and the range of variations as shown in [Figure 2]. The plotted data of the individual nine images were zoomed, and the following conclusions were drawn after review:
"With the increase in the value of "m" the upper, and lower limit of 2 nd order entropy and EC also increased, attained a maximum value at some point, after that it becomes constant. The upper and lower limit of the AMBE decreased with increase in the value of "m", and after some point it also attained a maximum value, and remained constant thereafter. The value of SEM initially increased very slowly with an increase in the value of 'm', and after a certain value of "m", it increased steeply. However, at different values of threshold "m", the trendline of contrast enhancement metrics are not the same; in fact in some cases, they are opposite. The plot of eight out of nine input images showed similar global appearance. The result of one input image showed a different trend in terms of AMBE, and SEM."
|Figure 2: The data plot of a representative image. (a) 2nd order entropy on the y-axis, (b) Absolute mean brightness error on the y-axis, (c) Edge content on the y-axis, and (d) Saturation evaluation metric on the y-axis. The x-axis has threshold value ("m") from 60 to 180 and at each threshold there were seven data points ("e") making total 847 data points|
Click here to view
We wrote "…attained a maximum value at some point…" in the previous paragraph because for all nine images these points (maximum value for 2 nd order entropy, EC, AMBE, and SEM) were different, and moreover these values were varying at each threshold that is shown in the zoomed view of [Figure 2], by taking only first 49 data point [Figure 3].
|Figure 3: Data plot of a representative image considering only 49 data values from the beginning of the data plot shown in Figure 1. Here the variation at a particular threshold is obvious. The x-axis has threshold value ("m") from 60 to 66 and at each threshold there were seven data points ("e") making total 49 data points|
Click here to view
[Table 2] shows the 2 nd order entropy, EC, AMBE, and SEM values, of images having the best possible contrast. Eight out of nine images had AMBE in the range of 1.02-8.69. One had AMBE value as 16.49 (S. no. 2), that is, its mean intensity was greater than the input image by 16.49. The values of SEM lied between 0 and 0.10. This indicates that very few pixels in the image either became white or black, as compared to the unprocessed image. The value of AMBE, and SEM together indicates that the overall or global brightness of the images have increased, and also very few pixels became white or black as compared to unprocessed image, and hence the contrast in the image improved. [Table 2] also demonstrates that a particular combination of "m" and "e" value might not be suitable for all 18 F-FDG PET images in HCC because for each image we found different values of "m" and "e." The best possible contrast in seven images was obtained at a threshold value of "m" between 150, and 160, and in another two images at 138 and 175.
|Table 2: The input parameters ("m" and "e") that resulted in best image and their corresponding values of 2nd order entropy, EC, ABME, and SEM|
Click here to view
The difference in entropy, and EC of the output images from their corresponding input images is shown in [Table 3]. The difference in EC is almost zero for all the nine images when the values are rounded to two decimal points, but there was the difference in the values, when the data were considered to six decimal points. This difference might not be appreciable visually, but this also indicates that the contrast enhancement produced by this technique has not deteriorated the edges. The difference in entropy was also very minimal (rounded data up to two decimal points), indicating that after processing, the information content of the images has reduced, however, it is very difficult to appreciate this reduction in information content visually. The image produced as a result of the application of the optimized value of input parameter "m" and "e" has been shown along with the original input image in [Figure 4] and [Figure 5]. From these figures, it is clear that the contrast in the output image is more than that of the input image, and simultaneously, it has not distorted the information in the other areas.
|Figure 4: 18Fluorin-fluorodeoxyglucose positron emission tomography images of a patient with hepatocellular carcinoma. (a) Original image, (b) image having highest value of entropy (c) image having highest value of edge content (slightly more than the original image), (d) image having lowest value of absolute mean brightness error, (e) lowest value of saturation evaluation metric, and (f) image with best contrast based on high value of entropy and edge content and low value of absolute mean brightness error and saturation evaluation metric. Visually it is clear that the selected transformation function in this study has produced better contrast between the tumor and normal liver cell than the original image|
Click here to view
|Figure 5: Original image along with processed image with optimum selected value of "m" and "e" for four different patients (a-d). The images clearly show that the contrast of the image has improved significantly and has not distorted any information available in original image|
Click here to view
|Table 3: Deviation of entropy and EC of the output images from that of the input images|
Click here to view
| Discussion|| |
In this study, we attempted to evaluate the optimum parameters for digital contrast enhancement of 18 F-FDG PET images in HCC. We found the different trend lines for 2 nd order entropy, EC, AMBE, and SEM at "m" = 60, and "m" = 152 of same image data, this may be because at "m" = 60, the darker pixels (pixel value <60) will be compressed into narrow range of darker level. And pixels having values above 60 (set of few black, gray, and white pixels) will be compressed into a narrow range of brighter pixels. Thus, the edge of the tumor that lies in the darker region will improve (because at threshold value 60, the difference between the black and white pixel intensities will increase, thus the contrast will improve), and entropy, and EC will be more at m = 60, and e = 1. As the slope increases, there will be the more smooth transition from black to white, compressed, narrow range of black, and white pixel will relax in both the direction, and therefore, the entropy and EC will decrease. At m = 152, pixel values below 152 (the set of black, gray, and white pixel will be compressed into the black regions) and above 152 (the set of white pixels will be compressed into narrow range of whiter level). Since our tumor is in black region, compression of pixel below 152 into narrow range of black will decrease the sharp transition as in the case of m = 60, and therefore entropy, and EC will be less in comparison to that of m = 60. However, as the slope will increase entropy and EC will improve, because of the relaxation in the compressed, narrow range of black, and white pixels.
The best possible contrast in seven out of nine images was obtained at threshold value "m" between 150 and 160, including both 150, and 160. In another two images, best possible contrast was found at 138, and 175. Eight out of nine images had AMBE in the range of 1.02-8.69. This means, that there is a minimal shift in brightness level after processing the image, and the output images are a little bit brighter than their corresponding input images. One output image had AMBE value of 16.49 (S. no. 2) that is., its mean intensity was greater than the input image by 16.49. The value of SEM lies between 0 and 0.10. This indicates that, after the process very few pixels have either become white, or black as compared to unprocessed images, that is., very less saturation occurred in the output image. The standardized input parameters resulted in the image having the best possible contrast. This study also demonstrates that a particular combination of "m" and "e," might not result in an image having the best possible contrast for all PET images; because for each image we have found the different value of "m" and "e." This is because each image is unique, and requires the unique value of "m" and "e" to produce best possible contrast. Our result suggests that a default parameter may not be the best for all kind of images, a finding that agrees with other authors. ,
Histogram of the input image provides a clue regarding the selection of the threshold. We found that the histograms of all the nine input images were concentrated in the brighter side, that is., from 150 to 255. We also found the threshold in the range 150-160. Therefore, we can say that the selection of threshold from the region of concentrated histogram might provide a best possible contrast in the image. However, to be more precise, we need to increase the number of input images. Our criteria for selection of parameters "m" and "e" that produce best possible contrast in an image was high value of entropy, EC, and low value of AMBE, SEM, as previously suggested by Saleem et al.  However, we did not find a single processed image (from a set of 847 images) that had a highest value of entropy, highest value of EC, lowest value of AMBE, and lowest value of SEM. This may be because with the increase in AMBE, the value of SEM also increases, the entropy increases with more variability in the data in spatial domain, and with the increase in entropy, it is not necessary that EC will also increase. Therefore, we made a compromise in the selection of the standardized parameter. We restricted the value of "m" in the range of 60-180, because in a prototype study we found that the images processed at "m" <60 and "m" >180 had maximum number of saturated pixels producing an image having no details.
To the best of our knowledge, till date the reported sensitivity, and specificity of PET based detection of HCC is based on the most widely used contrast enhancement tool, and very few attempts have been made to improve the contrast between the HCC tumor, and hepatocytes. One study was recently attempted to improve the contrast by using the principle of stochastic resonance  but it cannot be compared with the present study because the method of contrast enhancement is completely different. At this point, our study is different from others that the contrast enhancement technique with standardized input parameters has improved the contrast of those images which were adjusted for best possible contrast by reporting physician with existing routinely used contrast enhancement tool.
We must address the limitations of the present study. This is a small sample size study consisting of nine input images. The nine images included in this study were having histogram concentrated on the brighter side. There may be images whose histogram might be concentrated on the darker side or in the middle part of the histogram. This study lacks the inclusion of input image having the various types of histogram and images having a different spatial distribution of pixel values. Although the sample size was small, the strength of this study is that with one original image, 847 output images were generated using various combinations of "m" and "e" (any one of them can result in possible contrast) the best ways to standardize the input parameters, leaving least possibility of missing best possible contrast achievable in the output image. Also, the selection of best image will vary from reader's perspective. Therefore, the clinical endorsement for the improvement in the contrast of these images is required. Our future plan is to clinically evaluate this digital contrast enhancement technique by selecting more than fifty images of having different spatial patterns of pixels values.
| Conclusion|| |
The use of above mentioned digital contrast enhancement technique increases the overall contrast of the 18 F-FDG PET images in HCC and also the contrast between the HCC tumor and hepatocytes. Further clinical validation of this encouraging technique is required.
Financial support and sponsorship
Conflicts of interest
There are no conflicts of interest.
| References|| |
Parkin DM, Bray F, Ferlay J, Pisani P. Global cancer statistics, 2002. CA Cancer J Clin 2005;55:74-108.
Llovet JM, Burroughs A, Bruix J. Hepatocellular carcinoma. Lancet 2003;362:1907-17.
Bruix J, Sherman M, Llovet JM, Beaugrand M, Lencioni R, Burroughs AK, et al.
Clinical management of hepatocellular carcinoma. Conclusions of the Barcelona-2000 EASL conference. European Association for the Study of the Liver. J Hepatol 2001;35:421-30.
Bruix J, Sherman M, Practice Guidelines Committee, American Association for the Study of Liver Diseases. Management of hepatocellular carcinoma. Hepatology 2005;42:1208-36.
Lim JH, Kim EY, Lee WJ, Lim HK, Do YS, Choo IW, et al.
Regenerative nodules in liver cirrhosis: Findings at CT during arterial portography and CT hepatic arteriography with histopathologic correlation. Radiology 1999;210:451-8.
Matsui O, Kadoya M, Kameyama T, Yoshikawa J, Takashima T, Nakanuma Y, et al.
Benign and malignant nodules in cirrhotic livers: Distinction based on blood supply. Radiology 1991;178:493-7.
Weber G, Cantero A. Glucose-6-phosphatase activity in normal, pre-cancerous, and neoplastic tissues. Cancer Res 1955;15:105-8.
Weber G, Morris HP. Comparative biochemistry of hepatomas. III. Carbohydrate enzymes in liver tumors of different growth rates. Cancer Res 1963;23:987-94.
Messa C, Choi Y, Hoh CK, Jacobs EL, Glaspy JA, Rege S, et al.
Quantification of glucose utilization in liver metastases: Parametric imaging of FDG uptake with PET. J Comput Assist Tomogr 1992;16:684-9.
Okazumi S, Isono K, Enomoto K, Kikuchi T, Ozaki M, Yamamoto H, et al.
Evaluation of liver tumors using fluorine-18-fluorodeoxyglucose PET: Characterization of tumor and assessment of effect of treatment. J Nucl Med 1992;33:333-9.
Torizuka T, Tamaki N, Inokuma T, Magata Y, Sasayama S, Yonekura Y, et al. In vivo
assessment of glucose metabolism in hepatocellular carcinoma with FDG-PET. J Nucl Med 1995;36:1811-7.
Khan MA, Combs CS, Brunt EM, Lowe VJ, Wolverson MK, Solomon H, et al.
Positron emission tomography scanning in the evaluation of hepatocellular carcinoma. J Hepatol 2000;32:792-7.
Gonzalez RC, Woods RE, Eddins SL. Digital Image Processing Using MATLAB. 2 nd
ed. New Delhi: Tata McGraw Hill Education (P) Ltd.; 2011.
Gonzalez RF, Woods RE. Digital image processing. In: Gonzalez RF, Woods RE, editor. Image Enhancement. New York: Addison-Wesley Publishing Company; 1992. p. 163.
Pal NR, Pal SK. Entropic thresholding. Signal Processing 1989;16:97-108.
Saleem A, Beghdadi A, Boashash B. Image quality metrics based multifocus image fusion. 3 rd
European Workshop on Visual Information Processing (EUVIP); 2011.
Chen SD, Ramli AR. Minimum mean brightness error bi-histogram equalization in contrast enhancement. IEEE Trans Consum Electron 2003;49:1310-9.
Hautiere N, Tarel JP, Aubert D, Dumont E. Blind contrast enhancement assessment by gradient ratioing at visible edges. Image Anal Stereol 2008;27:87-95.
Morrow WM, Paranjape RB, Rangayyan RM, Desautels JL. Region-based contrast enhancement of mammograms. IEEE Trans Med Imaging 1992;11:392-406.
Saleem A, Beghdadi A, Boashash B. Image fusion-based contrast enhancement. EURASIP J Image Video Process 2012;2012:10.
Yeh JR, Hsu WT, Chang YC, Lo MT, Lin YH. Application of stochastic resonance for imaging enhancement of computed tomography in hepatocellular carcinoma. IEEE International Conference on Bioinformatics and Biomedicine Workshops (BIBMW); 2011. p. 945-7.
[Figure 1], [Figure 2], [Figure 3], [Figure 4], [Figure 5]
[Table 1], [Table 2], [Table 3]