- Email: [email protected]
Localization of prostatic tumor’s infection based on normalized mutual information MRI image segmentation
G Model
ARTICLE IN PRESS
JIPH-1172; No. of Pages 5
Journal of Infection and Public Health xxx (2019) xxx–xxx
Contents lists available at ScienceDirect
Journal of Infection and Public Health journal homepage: http://www.elsevier.com/locate/jiph
Localization of prostatic tumor’s infection based on normalized mutual information MRI image segmentation Guoping Lu a,∗,1 , Lixin Zhou b,1 a b
Department of Urology, Guangxi Minzu Hospital, Nanning, Guangxi 530001, China Department of Radiology, Guangxi Minzu Hospital, Nanning, Guangxi 530001, China
a r t i c l e
a b s t r a c t
i n f o
Article history: Received 8 May 2019 Received in revised form 11 August 2019 Accepted 15 August 2019 Keywords: Normalized mutual information Magnetic resonance imaging Image segmentation Prostate cancer Localization
To investigate the effect of normalized mutual information (MRI) image segmentation in accurate localization of prostate cancer with infection and the role in disease treatment, the normalized mutual information method is used to measure the similarity of images, so as to select the maps. Then, the popular global weighted voting method and normalized mutual information method are applied to calculate the weights and carry out the label image fusion. The map selection method based on mutual information substantially completes the segmentation of the MRI image prostate. The prostate position is basically found on the 10 test images, and the positioning of the prostate organs is deviated in the worst case. In the case of poor multi-map segmentation, it usually happens when those are not well represented in the map. Because of the structural similarity of medical images, multi-atlas segmentation based on normalized mutual information method can be done. Using the prior information of atlas, the atlas label image can be selected. After fusion, the final segmentation of the test image can be completed, which has a high accuracy for the location of prostate cancer. This method can accurately delineate the target area in radiotherapy of prostate cancer and reduce the damage of rectum, bladder and other organs caused by radiotherapy. However, there are still some problems in this study, such as inadequate segmentation accuracy, long data processing time and so on. There is still a certain distance from practicality, and further research is needed. © 2019 The Authors. Published by Elsevier Limited on behalf of King Saud Bin Abdulaziz University for Health Sciences. This is an open access article under the CC BY-NC-ND license (http:// creativecommons.org/licenses/by-nc-nd/4.0/).
Introduction With the continuous improvement of people’s living standards, the average life expectancy is also extended, and the incidence of prostate cancer in the world shows a significant upward trend [1]. According to statistics, the number of new cases of prostate cancer worldwide reached 680,000 in 2002, accounting for 11.7% of all new cases of different types of cancer. In China, the incidence of prostate cancer is rising rapidly, especially obvious in some big cities, and it has become a s¨ erious disaster area¨. Some patients with prostate cancer will be accompanied by a certain degree of infection, mostly chronic, with perineal and suprapubic pain, may have frequent urination, urgency and rectal irritation symptoms, or even acute urinary retention. Therefore, the study of prostate cancer is urgent.
∗ Corresponding author. E-mail address: luguoping [email protected] (G. Lu). 1 These authors equally contributed to this work.
Among the prostate cancer detection and radiation therapy, the first key issue is the accurate location of the prostate tumor, including its location and size. At the same time, it is important to accurately divide the prostate and surrounding organs by applying a prescribed dose of radiation to the glandular tissue [2,3]. At present, transrectal ultrasound imaging is a commonly used method of prostate imaging examination, which is widely used in the examination and diagnosis of prostate cancer. Prostate cancer in the CT image shows that the contour of the prostate edge is unclear, the tissue density is not uniform, and the anterior wall of the rectum and the bladder wall in front of the prostate may infiltrate [4–6]. Magnetic resonance imaging (MRI) is an imaging technique that uses image reconstruction of a signal generated by resonance of a nucleus in a strong magnetic field. Unlike other imaging techniques, magnetic resonance imaging has the advantages of high sensitivity, rich image information, high soft tissue resolution, no ionizing radiation, no adverse effects on the body, and has great advantages for breast cancer diagnosis. With the development of new magnetic resonance imaging technology, Dynamic Contrast-Enhanced Magnetic Resonance Imaging (DCE-MRI) technology provides more
https://doi.org/10.1016/j.jiph.2019.08.011 1876-0341/© 2019 The Authors. Published by Elsevier Limited on behalf of King Saud Bin Abdulaziz University for Health Sciences. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
Please cite this article in press as: Lu G, Zhou L. Localization of prostatic tumor’s infection based on normalized mutual information MRI image segmentation. J Infect Public Health (2019), https://doi.org/10.1016/j.jiph.2019.08.011
G Model JIPH-1172; No. of Pages 5
ARTICLE IN PRESS G. Lu, L. Zhou / Journal of Infection and Public Health xxx (2019) xxx–xxx
2
relevant information for tumor diagnosis. It is based on continuous scanning based on rapid imaging sequences, accurately reflecting the cellular information of tissue cell components, extracellular space, capillary permeability, microcirculation and perfusion of lesions. On the magnetic resonance imaging (MRI), the blood flow of the tumor tissue is rich, and it is different from the water content of the normal tissue, so that it can be more obvious that the prostate tissue is proliferated and enlarged. Therefore, MRI is superior to CT in showing the intrinsic tissue structure of the prostate and showing the dip of the tumor tissue to the surrounding adjacent adipose tissue, which is more diagnostic than CT. With the continuous cross-infiltration of various disciplines and the development of computer computing capabilities, medical image registration is becoming more and more intelligent. From the perspective of clinical application, the voxel-based registration method is better than the feature-based method. For a period of time in the future, the registration method based on mutual information will gain more attention, and gradually make medical image registration meet the clinical needs of more cases [7]. The registration method based on mutual information does not need to make assumptions about the relationship between the gray levels of the registration images, nor does it need to perform image segmentation or feature extraction before registration, and has good registration accuracy and robustness. And it is a fully automated registration method that has been successfully applied in medical image registration. Methods Normalized mutual information applied to image preprocessing In general, the image data of the patient is pre-processed after it is acquired. Image preprocessing often removes image noise, enhances image contrast, and segments regions of interest in an image. It is generally visually enhanced or segmented with interest. The image preprocessing uses the classical z-score normalization algorithm to adjust the dynamic range of the map image and the test image. This method normalizes the data with the mean and
standard deviation of the original data [8,9]. The processed data has a mean value of 0 and a standard deviation of 1, which is in accordance with the standard normal distribution. Mark the patient image to be segmented as T, and mark each image in the map as Ri . The conversion function is: Ri =
Ri − i i
(1)
Ri is the original image in the map image, i is the mean of the map image Ri , i is the standard deviation of Ri , and Ri is the normalized image of Ri . Medical image registration based on normalized mutual information Since the user does not need a prior knowledge of the registration image, the manual process requires almost no manual intervention, so the method is applicable to a wide range, and currently applies to most different modal image registrations. However, pure mutual information as a similarity measure of medical image registration also has many shortcomings, such as mutual information as a similarity measure is more sensitive to the performance of overlapping areas of images. When the overlap area is reduced, the sample points of the image are correspondingly reduced, and the reduction of the sample points is the reduction of the number of pixels participating in the statistical mutual information [10]. This leads to a significant decrease in the statistical index of the probability distribution and an increased likelihood of misregistration. In order to reduce the influence of image overlap area on mutual information similarity measure, Studholme first proposed the normalized mutual information theory in 1999, and its expression is as follows: NMI (A, B) =
H (A) + H (B) H (A, B)
(2)
H(A) and H(B) represent the information entropy of images A and B, respectively, and H(A, B) represents the joint entropy of images A and B.
Fig. 1. Multi-spectral segmentation process.
Please cite this article in press as: Lu G, Zhou L. Localization of prostatic tumor’s infection based on normalized mutual information MRI image segmentation. J Infect Public Health (2019), https://doi.org/10.1016/j.jiph.2019.08.011
G Model
ARTICLE IN PRESS
JIPH-1172; No. of Pages 5
G. Lu, L. Zhou / Journal of Infection and Public Health xxx (2019) xxx–xxx
3
Fig. 2. Segmentation experimental results.
Multi-spectral image segmentation method
Statistical analysis
The multi-spectral segmentation method is mainly divided into three processes. The first step is the registration process. Each map image is registered with the test image to be segmented, and a certain deformation is made to make it as similar as possible to the test image. The second step is map selection. The registration result obtained by the first step, according to a certain strategy, may be a priori or posteriori, to select the appropriate map image. Third, label fusion process, will soon get the map label. The algorithm is combined to obtain a final fusion label that is propagated to the test image to obtain the final segmentation [11–13]. The specific segmentation process is shown in Fig. 1.
From January 2016 to January 2017, 20 patients with pathologically confirmed prostate cancer in our hospital were enrolled in the study. All patients were incapable or unwilling to undergo surgery in the mid-high-risk period and volunteered to receive intensitymodulated radiation therapy. The patient’s age ranged from 55 to 78 years old. Tumor staging was performed in 13 patients with T2b T˜ 3a and 7 patients with T3b T˜ 4 . Paired t-test was performed on the volume of CTV based on MRI fusion images using SPSS 21.0. The difference was statistically significant at P < 0.05. Paired t-test was performed on the 40, 45, 50, 55, 60, 65, 70 Gy exposure dose volume and the mean dose and maximum dose of the rectum and bladder in the radiotherapy plan based on MRT fusion image. P < 0.05 indicated that the difference was statistically significant.
Radiotherapy targeting area and dose determination Before implementing 3D-CRT or IMRT radiotherapy for prostate cancer patients, the first task is to determine the target area. In the field of tumor radiotherapy, it mainly includes gross tumor area (GTV), clinical target area (CTV), internal target area (ITV), and planned target area (PTV) four target areas. The second task is to determine the dose that normal organ tissue adjacent to the target area can withstand [14]. The study found that CTV needs to include the entire prostate, because the prostate cancer capsules with different stages have a high rate of infection. The prostate is anatomically adjacent to the rectum and the wrist, so the dose of the rectum and the bladder during the radiotherapy of the cancer patient in the organ is strictly controlled to reduce and avoid the side effects caused by the radiotherapy. A large number of studies have shown that when the rectum receives a radiation of 70 Gy or more, the volume percentage must be strictly controlled below 25%; otherwise the probability of toxic side effects of the second or more than second grade of the rectum will increase greatly [15].
Results Classification of medical image registration According to the medical image modality, it is divided into single-modal medical image registration and multi-modal medical image registration. Single-modal medical image registration refers to the registration of two images obtained from the same imaging device, generally used in digital subtraction angiography imaging, and growth and development monitoring. The two images of multimodal image registration are from different imaging devices, which are mainly used for neurosurgery diagnosis, surgical positioning and radiation therapy planning design. Anatomical images such as CT, MRI, and DSA combined with functional images such as SPECT, PET, and EEG can provide positioning information during epileptic surgery. In addition, since MRI is suitable for the description of the
Please cite this article in press as: Lu G, Zhou L. Localization of prostatic tumor’s infection based on normalized mutual information MRI image segmentation. J Infect Public Health (2019), https://doi.org/10.1016/j.jiph.2019.08.011
G Model JIPH-1172; No. of Pages 5
ARTICLE IN PRESS G. Lu, L. Zhou / Journal of Infection and Public Health xxx (2019) xxx–xxx
4
Fig. 3. Dice coefficient distribution result.
contour of the tumor tissue, and the dose can be accurately calculated by the CT image, the two-image registration fusion images are usually required in the field of radiation therapy. At present, multi-modal image registration is a research hotspot in the field of medical image registration. Segmentation of experiment results In the MRI detection images of 20 patients, 10 test images ae selected. According to the research method of this paper, the similarity is calculated by using the normalized mutual information method for each test image, and the top 5 maps are selected according to the order of the values. After calculating the weight, the fusion label is calculated and segmented, and the Dice coefficient and Hausdorff distance are analyzed to evaluate the segmentation method. The partial segmentation results are shown in Fig. 2. Among them, Fig. 2(a) is the best of the segmentation results, and Fig. 2(b) is the worst segmentation result. This method performs better on some images but not on others. As can be seen in the figure, among the worst three results, there is a certain deviation in the positioning of the prostate, and discrete isolated regions appear. In the best case, the algorithm can get better results. Dice coefficient and Hausdorff distance In the 10 segmented images of this study, the largest Dice coefficient is 0.94, the worst is 0.64, and the average is 0.76, as shown in Fig. 3. The Hausdorff distance results of the 10 test image segmentation results are shown in Fig. 4. Where the Dice coefficient is higher, the segmentation is better, and the Hausdorff distance is correspondingly smaller; where the Dice coefficient is lower, the segmentation is poorer and the Hausdorff distance is also larger. CTV volume The mean volume of CTV (prostate + seminal vesicle) delineated by CT images of 20 patients with prostate cancer is 72.35 cm3 , and the mean volume of CTV (prostate + seminal vesicle) delineated by CT-MRI fusion image is 61.58 cm3 . The volume of the fusion image is 14.89% smaller than that of the CT sketch alone, and the difference is statistically significant (t = 3.243, P = 0.018). It is found that increasing the target dose of prostate cancer without increasing the damage of normal tissues and organs is the key to improving the efficacy of radiotherapy for prostate cancer, that is, radiotherapy has a significant dose-effect relationship with prostate cancer.
Fig. 4. Hausdorff distance distribution results.
Table 1 Comparison of CT image and CT-MRI fusion image in rectal exposure dose.
V40 /% V45 /% V50 /% V55 /% V60 /% V65 /% V70 /% Dmean /Gy
CT
CT-MRI
t
P
41.95 ± 7.98 38..17 ± 7.56 32.54 ± 6.78 25.34 ± 5.88 20.23 ± 5.18 16.07 ± 4.65 11.55 ± 3.92 34.79 ± 3.35
37.23 ± 6.97 33.34 ± 6.45 29.18 ± 5.32 21.19 ± 4.67 16.93 ± 3.94 12.36 ± 3.04 8.76 ± 2.18 31.87 ± 2.89
1.868 2.005 2.453 2.932 3.012 3.253 3.316 1.996
0.096 0.075 0.041 0.032 0.024 0.010 0.006 0.029
˜ Gy radiation. Dmean : mean ˜ 70 : percentages of rectal volumes exposed to 40 Gy70 V40 V dose of rectum.
While the rectum and bladder are the main threat organs that limit the dose of the target area, and the volume of the rectum to receive high doses is closely related to its severe late toxic side effects. This study shows that the use of CT-MRI fusion images to delineate the target area can significantly reduce the volume of the rectum exposed to high doses when the CT image is used to ensure the target dose (Table 1). Conclusion Medical image registration is an important branch of medical image processing and has a wide range of applications in medical image research. Due to the structural similarity of the medical images, a method based on multi-map segmentation can be used, and the a priori information in the map is used to select the map label image, and the final segmentation of the test image is completed after the fusion. The image registration method based on mutual information is currently the most popular registration method, which is applied to medical image registration in various modes and has achieved successful application. In view of the fact that the prostate has a small proportion in the MRI image, so the contribution in the image similarity is small, and the map selection is susceptible to the surrounding tissue of the prostate. It is proposed to constrain ROI with the label image during the dimensionality reduction projection process, to make the selected map image more accurate. The volume of PCV (prostate + seminal vesicle) delineated by CT-MRI fusion image is smaller than that of CT image, which can significantly reduce the volume of rectum irradiated by higher dose and reduce the dose of the highest dose point of bladder. Therefore, the use of CT-MRI fusion images in prostate cancer radiotherapy to map the target area can improve the accuracy of target area delineation, and reduce the incidence of toxic and side effects in the
Please cite this article in press as: Lu G, Zhou L. Localization of prostatic tumor’s infection based on normalized mutual information MRI image segmentation. J Infect Public Health (2019), https://doi.org/10.1016/j.jiph.2019.08.011
G Model JIPH-1172; No. of Pages 5
ARTICLE IN PRESS G. Lu, L. Zhou / Journal of Infection and Public Health xxx (2019) xxx–xxx
rectum, bladder and other dangerous organs, thus providing possibility for a better overall survival rate for patients with prostate cancer. Although the method of map selection and map fusion has been improved in the research, it has been improved in the final segmentation, but it is still not enough. There are still many details to be solved, including not high segmentation accuracy, long processing time etc. There is still a certain distance from the practical. The effects of different parameters on the segmentation effect have not been fully studied, such as the influence of region size and different regularization parameters. In addition, the similarity calculation is carried out by using the grayscale features of the image. Even after the image is uniformly preprocessed, the contrast between the brightness and the darkness may be inconsistent in the prostate region, which makes the difference in the similarity calculation, and the phase consistency adjustment can be considered in the future. Conflict of interest None declared. References [1] Dahran N, Szewczykbieda M, Wei C, Vinnicombe S, Nabi G. Normalized periprostatic fat MRI measurements can predict prostate cancer aggressiveness in men undergoing radical prostatectomy for clinically localised disease. Sci Rep 2017;7(1):4630. [2] Freedman JN, Collins DJ, Bainbridge H, Rank CM, Nill S, Kachelrieß M, et al. T2-weighted 4D magnetic resonance imaging for application in magnetic resonance-guided radiotherapy treatment planning. Invest Radiol 2017;52(10):1. [3] Padgett KR, Stoyanova R, Pirozzi S, Johnson P, Piper J, Dogan N, et al. Validation of a deformable MRI to CT registration algorithm employing same day planning MRI for surrogate analysis. J Appl Clin Med Phys 2018;19(2):258–64.
5
[4] Drozdzal M, Chartrand G, Vorontsov E, Shakeri M, Di Jorio L, Tang A, et al. Learning normalized inputs for iterative estimation in medical image segmentation. Med Image Anal 2017;44:1–13. [5] Ali TFT, Elhariri MA, Riad MM. Diffusion-weighted MRI in prostatic lesions: diagnostic performance of normalized ADC using normal peripheral prostatic zone as a reference. Egypt J Radiol Nucl Med 2018;49(1):239–44. [6] Perez-Lopez R, Nava RD, Figueiredo I, Mateo J, Collins DJ, Koh DM, et al. Multiparametric magnetic resonance imaging of prostate cancer bone disease: correlation with bone biopsy histological and molecular features. Invest Radiol 2017;53(2):96. [7] Huang W, Zhang J, Sun H, Ma H, Cai Z. An anomaly detection method based on normalized mutual information feature selection and quantum wavelet neural network. Wirel Pers Commun 2017;96(3-4):1–21. [8] Yang F, Ford JC, Dogan N, Padgett KR, Breto AL, Abramowitz MC, et al. Magnetic resonance imaging (MRI)-based radiomics for prostate cancer radiotherapy. Transl Androl Urol 2018;7(3): 445–58. [9] Lahoti AM, Dhok AP, Rantnaparkhi CR, Rawat JS, Chandak NU, Tawari HS. Role of magnetic resonance imaging, magnetic resonance spectroscopy and transrectal ultrasound in evaluation of prostatic pathologies with focus on prostate Cancer. Pol J Radiol 2018;82:827–36. [10] Ermpongkosol SP, Aramay S, Vattanakul T, Phongkitkarun S. The association between the outcomes of extraperitoneal laparoscopic radical prostatectomy and the anthropometric measurements of the prostate by magnetic resonance imaging. Int Braz J Urol 2018;44(2):238–47. [11] Dahran N, Szewczykbieda M, Wei C, Vinnicombe S, Nabi G. Normalized periprostatic fat MRI measurements can predict prostate cancer aggressiveness in men undergoing radical prostatectomy for clinically localised disease. Sci Rep 2017;7(1):4630. [12] Ogawa T, Kojima I, Ishii R, Sakamoto M, Murata T, Suzuki T, et al. Clinical utility of dynamic-enhanced MRI in salivary gland tumors: retrospective study and literature review. Eur Arch Otorhinolaryngol 2018;275(1060):1–9. [13] Piras P, Teresi L, Puddu PE, Torromeo C, Young AA, Suinesiaputra A, et al. Morphologically normalized left ventricular motion indicators from MRI feature tracking characterize myocardial infarction. Sci Rep 2017;7(1):12259. [14] Gnep K, Fargeas A, Gutiérrezcarvajal RE, Commandeur F, Mathieu R, Ospina JD, et al. Haralick textural features on T2 -weighted MRI are associated with biochemical recurrence following radiotherapy for peripheral zone prostate cancer. J Magn Reson Imaging 2017;45(1):103. [15] Zhao W, Quan Z, Huang X, Ren J, Wen D, Zhang G, et al. Grading of pancreatic neuroendocrine neoplasms using pharmacokinetic parameters derived from dynamic contrast-enhanced MRI. Oncol Lett 2018;15(6):8349–56.
Please cite this article in press as: Lu G, Zhou L. Localization of prostatic tumor’s infection based on normalized mutual information MRI image segmentation. J Infect Public Health (2019), https://doi.org/10.1016/j.jiph.2019.08.011
ARTICLE IN PRESS
JIPH-1172; No. of Pages 5
Journal of Infection and Public Health xxx (2019) xxx–xxx
Contents lists available at ScienceDirect
Journal of Infection and Public Health journal homepage: http://www.elsevier.com/locate/jiph
Localization of prostatic tumor’s infection based on normalized mutual information MRI image segmentation Guoping Lu a,∗,1 , Lixin Zhou b,1 a b
Department of Urology, Guangxi Minzu Hospital, Nanning, Guangxi 530001, China Department of Radiology, Guangxi Minzu Hospital, Nanning, Guangxi 530001, China
a r t i c l e
a b s t r a c t
i n f o
Article history: Received 8 May 2019 Received in revised form 11 August 2019 Accepted 15 August 2019 Keywords: Normalized mutual information Magnetic resonance imaging Image segmentation Prostate cancer Localization
To investigate the effect of normalized mutual information (MRI) image segmentation in accurate localization of prostate cancer with infection and the role in disease treatment, the normalized mutual information method is used to measure the similarity of images, so as to select the maps. Then, the popular global weighted voting method and normalized mutual information method are applied to calculate the weights and carry out the label image fusion. The map selection method based on mutual information substantially completes the segmentation of the MRI image prostate. The prostate position is basically found on the 10 test images, and the positioning of the prostate organs is deviated in the worst case. In the case of poor multi-map segmentation, it usually happens when those are not well represented in the map. Because of the structural similarity of medical images, multi-atlas segmentation based on normalized mutual information method can be done. Using the prior information of atlas, the atlas label image can be selected. After fusion, the final segmentation of the test image can be completed, which has a high accuracy for the location of prostate cancer. This method can accurately delineate the target area in radiotherapy of prostate cancer and reduce the damage of rectum, bladder and other organs caused by radiotherapy. However, there are still some problems in this study, such as inadequate segmentation accuracy, long data processing time and so on. There is still a certain distance from practicality, and further research is needed. © 2019 The Authors. Published by Elsevier Limited on behalf of King Saud Bin Abdulaziz University for Health Sciences. This is an open access article under the CC BY-NC-ND license (http:// creativecommons.org/licenses/by-nc-nd/4.0/).
Introduction With the continuous improvement of people’s living standards, the average life expectancy is also extended, and the incidence of prostate cancer in the world shows a significant upward trend [1]. According to statistics, the number of new cases of prostate cancer worldwide reached 680,000 in 2002, accounting for 11.7% of all new cases of different types of cancer. In China, the incidence of prostate cancer is rising rapidly, especially obvious in some big cities, and it has become a s¨ erious disaster area¨. Some patients with prostate cancer will be accompanied by a certain degree of infection, mostly chronic, with perineal and suprapubic pain, may have frequent urination, urgency and rectal irritation symptoms, or even acute urinary retention. Therefore, the study of prostate cancer is urgent.
∗ Corresponding author. E-mail address: luguoping [email protected] (G. Lu). 1 These authors equally contributed to this work.
Among the prostate cancer detection and radiation therapy, the first key issue is the accurate location of the prostate tumor, including its location and size. At the same time, it is important to accurately divide the prostate and surrounding organs by applying a prescribed dose of radiation to the glandular tissue [2,3]. At present, transrectal ultrasound imaging is a commonly used method of prostate imaging examination, which is widely used in the examination and diagnosis of prostate cancer. Prostate cancer in the CT image shows that the contour of the prostate edge is unclear, the tissue density is not uniform, and the anterior wall of the rectum and the bladder wall in front of the prostate may infiltrate [4–6]. Magnetic resonance imaging (MRI) is an imaging technique that uses image reconstruction of a signal generated by resonance of a nucleus in a strong magnetic field. Unlike other imaging techniques, magnetic resonance imaging has the advantages of high sensitivity, rich image information, high soft tissue resolution, no ionizing radiation, no adverse effects on the body, and has great advantages for breast cancer diagnosis. With the development of new magnetic resonance imaging technology, Dynamic Contrast-Enhanced Magnetic Resonance Imaging (DCE-MRI) technology provides more
https://doi.org/10.1016/j.jiph.2019.08.011 1876-0341/© 2019 The Authors. Published by Elsevier Limited on behalf of King Saud Bin Abdulaziz University for Health Sciences. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
Please cite this article in press as: Lu G, Zhou L. Localization of prostatic tumor’s infection based on normalized mutual information MRI image segmentation. J Infect Public Health (2019), https://doi.org/10.1016/j.jiph.2019.08.011
G Model JIPH-1172; No. of Pages 5
ARTICLE IN PRESS G. Lu, L. Zhou / Journal of Infection and Public Health xxx (2019) xxx–xxx
2
relevant information for tumor diagnosis. It is based on continuous scanning based on rapid imaging sequences, accurately reflecting the cellular information of tissue cell components, extracellular space, capillary permeability, microcirculation and perfusion of lesions. On the magnetic resonance imaging (MRI), the blood flow of the tumor tissue is rich, and it is different from the water content of the normal tissue, so that it can be more obvious that the prostate tissue is proliferated and enlarged. Therefore, MRI is superior to CT in showing the intrinsic tissue structure of the prostate and showing the dip of the tumor tissue to the surrounding adjacent adipose tissue, which is more diagnostic than CT. With the continuous cross-infiltration of various disciplines and the development of computer computing capabilities, medical image registration is becoming more and more intelligent. From the perspective of clinical application, the voxel-based registration method is better than the feature-based method. For a period of time in the future, the registration method based on mutual information will gain more attention, and gradually make medical image registration meet the clinical needs of more cases [7]. The registration method based on mutual information does not need to make assumptions about the relationship between the gray levels of the registration images, nor does it need to perform image segmentation or feature extraction before registration, and has good registration accuracy and robustness. And it is a fully automated registration method that has been successfully applied in medical image registration. Methods Normalized mutual information applied to image preprocessing In general, the image data of the patient is pre-processed after it is acquired. Image preprocessing often removes image noise, enhances image contrast, and segments regions of interest in an image. It is generally visually enhanced or segmented with interest. The image preprocessing uses the classical z-score normalization algorithm to adjust the dynamic range of the map image and the test image. This method normalizes the data with the mean and
standard deviation of the original data [8,9]. The processed data has a mean value of 0 and a standard deviation of 1, which is in accordance with the standard normal distribution. Mark the patient image to be segmented as T, and mark each image in the map as Ri . The conversion function is: Ri =
Ri − i i
(1)
Ri is the original image in the map image, i is the mean of the map image Ri , i is the standard deviation of Ri , and Ri is the normalized image of Ri . Medical image registration based on normalized mutual information Since the user does not need a prior knowledge of the registration image, the manual process requires almost no manual intervention, so the method is applicable to a wide range, and currently applies to most different modal image registrations. However, pure mutual information as a similarity measure of medical image registration also has many shortcomings, such as mutual information as a similarity measure is more sensitive to the performance of overlapping areas of images. When the overlap area is reduced, the sample points of the image are correspondingly reduced, and the reduction of the sample points is the reduction of the number of pixels participating in the statistical mutual information [10]. This leads to a significant decrease in the statistical index of the probability distribution and an increased likelihood of misregistration. In order to reduce the influence of image overlap area on mutual information similarity measure, Studholme first proposed the normalized mutual information theory in 1999, and its expression is as follows: NMI (A, B) =
H (A) + H (B) H (A, B)
(2)
H(A) and H(B) represent the information entropy of images A and B, respectively, and H(A, B) represents the joint entropy of images A and B.
Fig. 1. Multi-spectral segmentation process.
Please cite this article in press as: Lu G, Zhou L. Localization of prostatic tumor’s infection based on normalized mutual information MRI image segmentation. J Infect Public Health (2019), https://doi.org/10.1016/j.jiph.2019.08.011
G Model
ARTICLE IN PRESS
JIPH-1172; No. of Pages 5
G. Lu, L. Zhou / Journal of Infection and Public Health xxx (2019) xxx–xxx
3
Fig. 2. Segmentation experimental results.
Multi-spectral image segmentation method
Statistical analysis
The multi-spectral segmentation method is mainly divided into three processes. The first step is the registration process. Each map image is registered with the test image to be segmented, and a certain deformation is made to make it as similar as possible to the test image. The second step is map selection. The registration result obtained by the first step, according to a certain strategy, may be a priori or posteriori, to select the appropriate map image. Third, label fusion process, will soon get the map label. The algorithm is combined to obtain a final fusion label that is propagated to the test image to obtain the final segmentation [11–13]. The specific segmentation process is shown in Fig. 1.
From January 2016 to January 2017, 20 patients with pathologically confirmed prostate cancer in our hospital were enrolled in the study. All patients were incapable or unwilling to undergo surgery in the mid-high-risk period and volunteered to receive intensitymodulated radiation therapy. The patient’s age ranged from 55 to 78 years old. Tumor staging was performed in 13 patients with T2b T˜ 3a and 7 patients with T3b T˜ 4 . Paired t-test was performed on the volume of CTV based on MRI fusion images using SPSS 21.0. The difference was statistically significant at P < 0.05. Paired t-test was performed on the 40, 45, 50, 55, 60, 65, 70 Gy exposure dose volume and the mean dose and maximum dose of the rectum and bladder in the radiotherapy plan based on MRT fusion image. P < 0.05 indicated that the difference was statistically significant.
Radiotherapy targeting area and dose determination Before implementing 3D-CRT or IMRT radiotherapy for prostate cancer patients, the first task is to determine the target area. In the field of tumor radiotherapy, it mainly includes gross tumor area (GTV), clinical target area (CTV), internal target area (ITV), and planned target area (PTV) four target areas. The second task is to determine the dose that normal organ tissue adjacent to the target area can withstand [14]. The study found that CTV needs to include the entire prostate, because the prostate cancer capsules with different stages have a high rate of infection. The prostate is anatomically adjacent to the rectum and the wrist, so the dose of the rectum and the bladder during the radiotherapy of the cancer patient in the organ is strictly controlled to reduce and avoid the side effects caused by the radiotherapy. A large number of studies have shown that when the rectum receives a radiation of 70 Gy or more, the volume percentage must be strictly controlled below 25%; otherwise the probability of toxic side effects of the second or more than second grade of the rectum will increase greatly [15].
Results Classification of medical image registration According to the medical image modality, it is divided into single-modal medical image registration and multi-modal medical image registration. Single-modal medical image registration refers to the registration of two images obtained from the same imaging device, generally used in digital subtraction angiography imaging, and growth and development monitoring. The two images of multimodal image registration are from different imaging devices, which are mainly used for neurosurgery diagnosis, surgical positioning and radiation therapy planning design. Anatomical images such as CT, MRI, and DSA combined with functional images such as SPECT, PET, and EEG can provide positioning information during epileptic surgery. In addition, since MRI is suitable for the description of the
Please cite this article in press as: Lu G, Zhou L. Localization of prostatic tumor’s infection based on normalized mutual information MRI image segmentation. J Infect Public Health (2019), https://doi.org/10.1016/j.jiph.2019.08.011
G Model JIPH-1172; No. of Pages 5
ARTICLE IN PRESS G. Lu, L. Zhou / Journal of Infection and Public Health xxx (2019) xxx–xxx
4
Fig. 3. Dice coefficient distribution result.
contour of the tumor tissue, and the dose can be accurately calculated by the CT image, the two-image registration fusion images are usually required in the field of radiation therapy. At present, multi-modal image registration is a research hotspot in the field of medical image registration. Segmentation of experiment results In the MRI detection images of 20 patients, 10 test images ae selected. According to the research method of this paper, the similarity is calculated by using the normalized mutual information method for each test image, and the top 5 maps are selected according to the order of the values. After calculating the weight, the fusion label is calculated and segmented, and the Dice coefficient and Hausdorff distance are analyzed to evaluate the segmentation method. The partial segmentation results are shown in Fig. 2. Among them, Fig. 2(a) is the best of the segmentation results, and Fig. 2(b) is the worst segmentation result. This method performs better on some images but not on others. As can be seen in the figure, among the worst three results, there is a certain deviation in the positioning of the prostate, and discrete isolated regions appear. In the best case, the algorithm can get better results. Dice coefficient and Hausdorff distance In the 10 segmented images of this study, the largest Dice coefficient is 0.94, the worst is 0.64, and the average is 0.76, as shown in Fig. 3. The Hausdorff distance results of the 10 test image segmentation results are shown in Fig. 4. Where the Dice coefficient is higher, the segmentation is better, and the Hausdorff distance is correspondingly smaller; where the Dice coefficient is lower, the segmentation is poorer and the Hausdorff distance is also larger. CTV volume The mean volume of CTV (prostate + seminal vesicle) delineated by CT images of 20 patients with prostate cancer is 72.35 cm3 , and the mean volume of CTV (prostate + seminal vesicle) delineated by CT-MRI fusion image is 61.58 cm3 . The volume of the fusion image is 14.89% smaller than that of the CT sketch alone, and the difference is statistically significant (t = 3.243, P = 0.018). It is found that increasing the target dose of prostate cancer without increasing the damage of normal tissues and organs is the key to improving the efficacy of radiotherapy for prostate cancer, that is, radiotherapy has a significant dose-effect relationship with prostate cancer.
Fig. 4. Hausdorff distance distribution results.
Table 1 Comparison of CT image and CT-MRI fusion image in rectal exposure dose.
V40 /% V45 /% V50 /% V55 /% V60 /% V65 /% V70 /% Dmean /Gy
CT
CT-MRI
t
P
41.95 ± 7.98 38..17 ± 7.56 32.54 ± 6.78 25.34 ± 5.88 20.23 ± 5.18 16.07 ± 4.65 11.55 ± 3.92 34.79 ± 3.35
37.23 ± 6.97 33.34 ± 6.45 29.18 ± 5.32 21.19 ± 4.67 16.93 ± 3.94 12.36 ± 3.04 8.76 ± 2.18 31.87 ± 2.89
1.868 2.005 2.453 2.932 3.012 3.253 3.316 1.996
0.096 0.075 0.041 0.032 0.024 0.010 0.006 0.029
˜ Gy radiation. Dmean : mean ˜ 70 : percentages of rectal volumes exposed to 40 Gy70 V40 V dose of rectum.
While the rectum and bladder are the main threat organs that limit the dose of the target area, and the volume of the rectum to receive high doses is closely related to its severe late toxic side effects. This study shows that the use of CT-MRI fusion images to delineate the target area can significantly reduce the volume of the rectum exposed to high doses when the CT image is used to ensure the target dose (Table 1). Conclusion Medical image registration is an important branch of medical image processing and has a wide range of applications in medical image research. Due to the structural similarity of the medical images, a method based on multi-map segmentation can be used, and the a priori information in the map is used to select the map label image, and the final segmentation of the test image is completed after the fusion. The image registration method based on mutual information is currently the most popular registration method, which is applied to medical image registration in various modes and has achieved successful application. In view of the fact that the prostate has a small proportion in the MRI image, so the contribution in the image similarity is small, and the map selection is susceptible to the surrounding tissue of the prostate. It is proposed to constrain ROI with the label image during the dimensionality reduction projection process, to make the selected map image more accurate. The volume of PCV (prostate + seminal vesicle) delineated by CT-MRI fusion image is smaller than that of CT image, which can significantly reduce the volume of rectum irradiated by higher dose and reduce the dose of the highest dose point of bladder. Therefore, the use of CT-MRI fusion images in prostate cancer radiotherapy to map the target area can improve the accuracy of target area delineation, and reduce the incidence of toxic and side effects in the
Please cite this article in press as: Lu G, Zhou L. Localization of prostatic tumor’s infection based on normalized mutual information MRI image segmentation. J Infect Public Health (2019), https://doi.org/10.1016/j.jiph.2019.08.011
G Model JIPH-1172; No. of Pages 5
ARTICLE IN PRESS G. Lu, L. Zhou / Journal of Infection and Public Health xxx (2019) xxx–xxx
rectum, bladder and other dangerous organs, thus providing possibility for a better overall survival rate for patients with prostate cancer. Although the method of map selection and map fusion has been improved in the research, it has been improved in the final segmentation, but it is still not enough. There are still many details to be solved, including not high segmentation accuracy, long processing time etc. There is still a certain distance from the practical. The effects of different parameters on the segmentation effect have not been fully studied, such as the influence of region size and different regularization parameters. In addition, the similarity calculation is carried out by using the grayscale features of the image. Even after the image is uniformly preprocessed, the contrast between the brightness and the darkness may be inconsistent in the prostate region, which makes the difference in the similarity calculation, and the phase consistency adjustment can be considered in the future. Conflict of interest None declared. References [1] Dahran N, Szewczykbieda M, Wei C, Vinnicombe S, Nabi G. Normalized periprostatic fat MRI measurements can predict prostate cancer aggressiveness in men undergoing radical prostatectomy for clinically localised disease. Sci Rep 2017;7(1):4630. [2] Freedman JN, Collins DJ, Bainbridge H, Rank CM, Nill S, Kachelrieß M, et al. T2-weighted 4D magnetic resonance imaging for application in magnetic resonance-guided radiotherapy treatment planning. Invest Radiol 2017;52(10):1. [3] Padgett KR, Stoyanova R, Pirozzi S, Johnson P, Piper J, Dogan N, et al. Validation of a deformable MRI to CT registration algorithm employing same day planning MRI for surrogate analysis. J Appl Clin Med Phys 2018;19(2):258–64.
5
[4] Drozdzal M, Chartrand G, Vorontsov E, Shakeri M, Di Jorio L, Tang A, et al. Learning normalized inputs for iterative estimation in medical image segmentation. Med Image Anal 2017;44:1–13. [5] Ali TFT, Elhariri MA, Riad MM. Diffusion-weighted MRI in prostatic lesions: diagnostic performance of normalized ADC using normal peripheral prostatic zone as a reference. Egypt J Radiol Nucl Med 2018;49(1):239–44. [6] Perez-Lopez R, Nava RD, Figueiredo I, Mateo J, Collins DJ, Koh DM, et al. Multiparametric magnetic resonance imaging of prostate cancer bone disease: correlation with bone biopsy histological and molecular features. Invest Radiol 2017;53(2):96. [7] Huang W, Zhang J, Sun H, Ma H, Cai Z. An anomaly detection method based on normalized mutual information feature selection and quantum wavelet neural network. Wirel Pers Commun 2017;96(3-4):1–21. [8] Yang F, Ford JC, Dogan N, Padgett KR, Breto AL, Abramowitz MC, et al. Magnetic resonance imaging (MRI)-based radiomics for prostate cancer radiotherapy. Transl Androl Urol 2018;7(3): 445–58. [9] Lahoti AM, Dhok AP, Rantnaparkhi CR, Rawat JS, Chandak NU, Tawari HS. Role of magnetic resonance imaging, magnetic resonance spectroscopy and transrectal ultrasound in evaluation of prostatic pathologies with focus on prostate Cancer. Pol J Radiol 2018;82:827–36. [10] Ermpongkosol SP, Aramay S, Vattanakul T, Phongkitkarun S. The association between the outcomes of extraperitoneal laparoscopic radical prostatectomy and the anthropometric measurements of the prostate by magnetic resonance imaging. Int Braz J Urol 2018;44(2):238–47. [11] Dahran N, Szewczykbieda M, Wei C, Vinnicombe S, Nabi G. Normalized periprostatic fat MRI measurements can predict prostate cancer aggressiveness in men undergoing radical prostatectomy for clinically localised disease. Sci Rep 2017;7(1):4630. [12] Ogawa T, Kojima I, Ishii R, Sakamoto M, Murata T, Suzuki T, et al. Clinical utility of dynamic-enhanced MRI in salivary gland tumors: retrospective study and literature review. Eur Arch Otorhinolaryngol 2018;275(1060):1–9. [13] Piras P, Teresi L, Puddu PE, Torromeo C, Young AA, Suinesiaputra A, et al. Morphologically normalized left ventricular motion indicators from MRI feature tracking characterize myocardial infarction. Sci Rep 2017;7(1):12259. [14] Gnep K, Fargeas A, Gutiérrezcarvajal RE, Commandeur F, Mathieu R, Ospina JD, et al. Haralick textural features on T2 -weighted MRI are associated with biochemical recurrence following radiotherapy for peripheral zone prostate cancer. J Magn Reson Imaging 2017;45(1):103. [15] Zhao W, Quan Z, Huang X, Ren J, Wen D, Zhang G, et al. Grading of pancreatic neuroendocrine neoplasms using pharmacokinetic parameters derived from dynamic contrast-enhanced MRI. Oncol Lett 2018;15(6):8349–56.
Please cite this article in press as: Lu G, Zhou L. Localization of prostatic tumor’s infection based on normalized mutual information MRI image segmentation. J Infect Public Health (2019), https://doi.org/10.1016/j.jiph.2019.08.011