The Application Of Image Segmentation And Surgical Navigation Techniques In Hepatobiliary Surgery

Research BackgroundEach year, the incidence of hepatocellular carcinoma (HCC) rises and more than 800,000 patients die from the disease. HCC is the second most common cause of cancer death worldwide and is associated with hepatitis virus infection in 80% of cases. Hepatectomy and liver transplantation have been considered as the main curative treatments. They are important ways for HCC patients to acquire long-term survival currently. For hepatectomy, the accurate preoperative measurement of liver volume, function and vascular anatomy is essential in that postoperative hypohepatia/liver failure to be prevented, remnant liver volume with sufficient vascular inflow and outflow should be reserved, and the biliary draining system with adequate drainage needs to be preserved. Simultaneously, the tumor has to be removed completely with adequate margins (R0 resection) and the perioperative mortality should be reduced as much as possible. Liver transplantation is also an important way of treatment, including the deceased donor liver transplant and the adult living donor liver transplant (LDLT). The evaluation of the total and segmental liver volumes and intrahepatic anatomy is crucial for LDLT because they are major factors in predicting the safe outcome for both donor and recipient. In general, when dealing with HCC patients under liver surgery, both evaluating the liver volume and assessing intrahepatic vascular anatomy are important constituent parts of preoperative surgical planning.Surgeons require simple yet accurate method to evaluate liver volume in order to perform the liver surgery safely. As an accurate and noninvasive detection method, computed tomography (CT) plays an important role in quantitative radiology and precision surgery. There are three methods to do the liver segmentation based on CT images through the computer technology:manual volumetry, semiautomatic interactive volumetry and automated volumetry. Manual volumetry is the current "gold-standard" for liver volume calculation. It can provide accurate results, but it is very time consuming, tedious and subjective. All these defects limit its applications in clinical work. With the development of computer technology, there are many semiautomatic interactive volumetry and automated volumetry methods come forth. These methods are with high computational speed. If the accuracy of these methods in calculation of liver volumes can reach the level of manual volumetry, they will be beneficial additions to clinical work.Although liver surgery is the radical treatment of HCC, the epidemiological investigation shows that there are only 10-30% of affected patients are currently classified as surgical resectable due to most of them are associated with hepatitis cirrhosis. Over 80% patients with HCC have to choose non-radical treatments such as transcatheter hepatic arterial chemoembolization or ablation. The indications of local ablation are widely, it is a safely and effective way to treat HCC patients who can not undergo the hepatectomy and it can be an interim treatment of liver transplantation. There are several common guidance methods such as under direct vision/intraoperative ultrasonic guidance, under laparoscopic guidance, percutaneous puncture under US/CT/MRI guidance and so on. The percutaneous puncture has been widely used with advantages of little trauma, less pain, short hospital stay, repetitive punctures and so on. But the surgery must be guided with imaging equipment such as US, CT or MRI. All these equipments have different limitations respectively.Based on the above issues, we aim to establish the "gold standard" that use to compare and evaluate of the accuracy, consistency, and efficiency of a semi-automatic interactive medical image three-dimensional visualization system, MI-3DVS (software registration number:2008SR18798) which was developed by Southern Medical University and South China Normal University and two automatic liver volumetry (statistical shape model, SSM and probabilistic atlas, PA) which were developed by Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences to appraise the preponderance and deficiency of MI-3DVS, SSM and PA. So we could choose preferred liver segmentation and liver evaluation methods preoperatively for liver surgery. MI-3DVS can offer omnibearing observations from multi-viewpoints of abdominal organs and evaluation of the liver volume,. We could use them to estimate the spatial relationships between vessels and nearby organs. Under the guideline of Couinaud’s liver segment theory, we can also use MI-3DVS to perform the individualized liver segments. We collected and analyzed the clinical data of patients with HCC underwent hepatic resection based on MI-3DVS in our department to estimate if the MI-3DVS could improve the accuracy of liver surgery and enhance the safety of hepatectomy. Fianlly, an optical tracking liver puncture navigation system was developed. The accuracy, stability and practicability of this system through a series of puncture experiments were evaluated. We aimed to overcome the shortcomings of existing guiding methods and provide an accurate, stable and quick way to orientation and puncture.Part Ⅰ. Comparison and evaluation of one semi-automatic interactive and two automatic liver volumctry methods on clinical CT imagesObjectives1. Optimization the CT imaging methods to acquire high quality CT triphase images.2. Comparison and evaluation of the accuracy, consistency, and efficiency of MI-3DVS, SSM, PA with manual volumetry on clinical CT images to appraise the preponderance and deficiency of these three methods.Materials and Methods1. Subjects42 patients examined with 64/256-slice multidetector CT (64/256-MDCT) scanning in the Department of Hepatobiliary Surgery (Ⅰ), Zhujiang Hospital, Southern Medical University between April 2013 and January 2014 were randomly chosen. Among these, were 27 cases with of normal livers (group A,21 men and 6 women) and 15 patients with disease (group B,9 men and 6 women). In group A, the mean age of the men was 50.0±15.0 years (range,5 to 81 years) and that of the women was 57.8 ±10.9 years (range,41 to 70 years). In group B, the mean age of the men was 48.3± 15.3 years (range,24 to 72 years) and that of the women was 60.2±13.6 years (range,38 to 77 years). In group B, the disease models included hepatocellular carcinoma (7 patients) and intra-hepatic calculi (8 patients). Portal venous phase images were used in this study because they can maximize the intensity difference between liver parenchyma and non-liver tissue.2. Equipment and facilities(1) CT scanner:PHILIPS BRILLIANCE 64/256 slice multi-detector CT scanner; CT contrast medium:Iopamiro; double tube high pressure syringe; system-provided by 256-MDCT Mxview workstation for post-processing of the images. (2) One blade server from the digital medical clinical trial center of Southern Medical University. (3) One high performance workstation. (4) One semi-automatic interactive volumetry sofrware:MI-3DVS, two automatic liver volumetry softwares: SSM and PA.3. The collection and storage of CT imagesThe process was presentated in this reference:Yang J, Fang C H, Fan Y F, et al. To assess the benefits of medical image three-dimensional visualization system assisted pancreaticoduodenctomy for patients with hepatic artery variance [J]. Int J Med Robot,2014.4. The processing methods of CT images(1) I used the Microsoft GeoS to perform 300 liver segmentation as exercises. The score results showed that the ability of my segmentation is equal to an experience radiologist. Then I used this software to perform the manual volumetry and the results were rechecked by another board-certified radiologist to ensure the quality. So, the gold-standard of this study was established. (2) All the semi-automatic interactive and automatic liver volumetry were performed by the radiologist.5. The brief introduction of evaluation indexes of liver segmentation based on CT images(1) Liver volume:V;(2) Segmentation time:T;(3) Relative absolute volume difference, VD. VD>0 means over-segmentation; VD<0 means under-segmentation; VD=0 means the segmentation volume is equal to the manual volumetry.6. Statistical analysisWhen the test of homogeneity of variances was not significant (P>0.05), an analysis of variance (ANOVA) was performed. When the test of ANOVA was not significant (P>0.05), the least significant difference (LSD) or Bonferroni test was used to do the pairwise comparison. When the test of homogeneity of variances was significant (P<0.05), the Kruskal Wallis test was performed. Finally, when the test of Kruskal Wallis was significant (P<0.05), the multiple Student t-test with Bonferroni correction was chosen to further compare the two sets of data. Linear association was evaluated with the Pearson correlation coefficient (r). Intra-class correlation coefficient (ICC) was chosen to evaluate the agreement between automated volumetry or semi-automatic interactive volumetry and manual volumetry. The evaluation of the level of agreement between two volumetric methods was performed by the method described by Bland and Altman. SPSS 13.0 software (SPSSInc, Chicago, IL, USA) was used to perform the statistical analysis; P<0.05 was considered statistically significant.Results1. The liver volumes obtained from different methodsIn group A, the mean gold standard manual volume was 1203.3±285.6 cm3, the mean liver volume obtained by MI-3DVS was 1182.4±279.3 cm3, the mean liver volume obtained by SSM was 1170.2±279.9 cm3, and the mean liver volume obtained by PA was 1233±274.5 cm3. Compared with gold standard, the VD error was-1.69% with MI-3DVS,-2.75% with SSM and 3.06% with PA. In group B, the mean gold standard manual volume was 1272.3±313.2cm3, the mean liver volume obtained by MI-3DVS was 1233.1±317.4cm3, the mean liver volume obtained by SSM was 1228.2±300.0cm3, and the mean liver volume obtained by PA was 1314.4±301.6cm3. Compared with gold standard, the VD error was -3.20% with system-provided by 256-MDCT Mxview workstation for post-processing of the images. (2) One blade server from the digital medical clinical trial center of Southern Medical University. (3) One high performance workstation. (4) One semi-automatic interactive volumetry sofrware:MI-3DVS, two automatic liver volumetry softwares: SSM and PA.3. The collection and storage of CT imagesThe process was presentated in this reference:Yang J, Fang C H, Fan Y F, et al. To assess the benefits of medical image three-dimensional visualization system assisted pancreaticoduodenctomy for patients with hepatic artery variance [J]. Int J Med Robot,2014.4. The processing methods of CT images(1)1 used the Microsoft GeoS to perform 300 liver segmentation as exercises. The score results showed that the ability of my segmentation is equal to an experience radiologist. Then I used this software to perform the manual volumetry and the results were rechecked by another board-certified radiologist to ensure the quality. So, the gold-standard of this study was established. (2) All the semi-automatic interactive and automatic liver volumetry were performed by the radiologist.5. The brief introduction of evaluation indexes of liver segmentation based on CT images(1) Liver volume:V;(2) Segmentation time:T;(3) Relative absolute volume difference, VD. VD>0 means over-segmentation; VD<0 means under-segmentation; VD=0 means the segmentation volume is equal to the manual volumetry.6. Statistical analysisWhen the test of homogeneity of variances was not significant (P>0.05), an analysis of variance (ANOVA) was performed. When the test of ANOVA was not significant (P>0.05), the least significant difference (LSD) or Bonferroni test was used to do the pairwise comparison. When the test of homogeneity of variances was significant (P<0.05), the Kruskal Wallis test was performed. Finally, when the test of Kruskal Wallis was significant (P<0.05), the multiple Student t-test with Bonferroni correction was chosen to further compare the two sets of data. Linear association was evaluated with the Pearson correlation coefficient (r). Intra-class correlation coefficient (ICC) was chosen to evaluate the agreement between automated volumetry or semi-automatic interactive volumetry and manual volumetry. The evaluation of the level of agreement between two volumetric methods was performed by the method described by Bland and Altman. SPSS 13.0 software (SPSSInc, Chicago, IL, USA) was used to perform the statistical analysis; P<0.05 was considered statistically significant.Results1. The liver volumes obtained from different methodsIn group A, the mean gold standard manual volume was 1203.3±285.6 cm3, the mean liver volume obtained by MI-3DVS was 1182.4±279.3 cm3, the mean liver volume obtained by SSM was 1170.2±279.9 cm3, and the mean liver volume obtained by PA was 1233±274.5 cm3. Compared with gold standard, the VD error was -1.69% with MI-3DVS,-2.75% with SSM and 3.06% with PA. In group B, the mean gold standard manual volume was 1272.3±313.2cm3, the mean liver volume obtained by MI-3DVS was 1233.1±317.4cm, the mean liver volume obtained by SSM was 1228.2±300.0cm3, and the mean liver volume obtained by PA was 1314.4±301.6cm3. Compared with gold standard, the VD error was -3.20% with MI-3DVS -3.35% with SSM and 4.14% with PA. Multiple comparisons of liver volume differences among MI-3DVS, SSM, and PA indicated that the liver volume obtained by different methods was very close and without statistically significant differences:MI-3DVS vs manual in group A (P=0.783) and in group B (P=0.731), SSM vs manual in group A (P=0.665) and in group B (P=0.696), PA vs manual in group A (P=0.696) and in group B (P=0.710), SSM vs MI-3DVS in group A (P=0.875) and in group B (P=0.963), PA vs MI-3DVS in group A (P=0.505) and in group B (P=0.475), PA vs SSM in group A (P=0.410) and in group B (P=0.447). The multiple comparisons of VD error among MI-3DVS, SSM, and PA showed that the MI-3DVS, SSM, and PA did not have statistically significant differences:SSM vs MI-3DVS in group A (P=0.233) and in group B (P=0.066), PA vs MI-3DVS in group A (P=0.715) and in group B (P=0.262), PA vs SSM in group A (P=0.406) and in group B (P=0.326).2. The processing time of different methodsThe mean processing time for manual volume was 41.78±10.09 min and 47.93±6.32 min for group A and B, respectively. The mean processing time for MI-3DVS was 27.63±4.50 min and 27.53±4.00 min for group A and B, respectively. The mean processing time for SSM was 1.28±0.51 min in group A and 1.23±0.60 min in group B, respectively. The mean processing time for PA was 1.19±0.20 min in group A and 1.16±0.09 min in group B, respectively. (1) There were significant differences between MI-3DVS and manual volumetry between SSM and manual volumetry, between PA and manual volumetry, between SSM and MI-3DVS, between PA and MI-3DVS, both in group A and in group B (all P<0.001). All the three methods are faster than manual volumerty. (2) Both in group A and B, there were statistically significant differences between MI-3DVS and SSM, MI-3DVS and PA. The automatic methods are faster than semi-automatic method. (3) However, there were no statistically significant differences between SSM and PA in group A (P=0.950) or group B (P=0.960).3. The agreement between different methodsIn group A, the ICC between MI-3DVS and manual volumetry is 0.982 and the r is 0.984, the ICC between SSM and manual volumetry is 0.980 and the r is 0.986, the ICC between PA and manual volumetry is 0.978 and the r is 0.984. In group B, the ICC between MI-3DVS and manual volumetry is 0.987 and the r is 0.994, the ICC between SSM and manual volumetry is 0.985 and the r is 0.996, the ICC between PA and manual volumetry is 0.955 and the r is 0.982. The results of Bland-Altman analysis showed that in both groups, MI-3DVS, SSM and PA achieved excellent agreement with manual volumetry.Conclusions1. In both groups, an excellent agreement in volumes were achieved between SSM vs the gold standard, PA vs the gold standard, and MI-3DVS vs the gold standard. The accuracy is quite satisfied.2. In both groups, the computational speed of SSM and PA are better than MI-3DVS. It indicates that SSM and PA have high efficiency.3. MI-3DVS, SSM and PA are all good at calcuating the liver volume and they can be qualified for the clinical work. Meanwhile, MI-3DVS has powerful capabilities of image postprocessing. It can be used to segment the liver and observe the liver from multiple viewpoints. It can display the relationships between lesions and normal organs or vessels clearly. It can be used for virtual surgery. MI-3DVS is an important assistive tool for clinical work currently.Part Ⅱ. The research of MI-3DVS in individualized liver segments and preoperative planning for hepatectomyObjectives1. Evaluating the practical capacity of MI-3DVS in realizing the digital individualivzed liver segments.2. Estimating the clinical value of MI-3DVS in improving the accuracy of hepatectomy in preoperative planning and enhancing the safety of hepatic surgery.Materials and Methods1. SubjectsBetween January 2013 and June 2014,34 consecutive patients with HCC underwent hepatic resection based on MI-3DVS in the Department of Hepatobiliary Surgery (Ⅰ), Zhujiang Hospital, Southern Medical University. All (29 men and 5 women) the cases were collected for the study. The mean age of the men was 49.7±12.5 years (range,23 to 75 years) and that of the women was 48.2±16.9 years (range,32 to 67 years). Among the 34 patients,4 of them underwent anatomical liver resection (group A) and the others underwent local segmental resection (group B).2. Equipment and facilities(1) CT scanner:PHILIPS BRILLIANCE 64/256 slice multi-detector CT scanner; CT contrast medium:Iopamiro; double tube high pressure syringe; 64/256-MDCT Mxview workstation for post-processing of the images. (2) One blade server from the digital medical clinical trial center of Southern Medical University. (3) One high performance computer. (4) Medical image three-dimensional visualization system, MI-3DVS.3. The collection and storage of CT images are the same as described in part Ⅰ4. The three-dimensional reconstruction of abdominal images, realization of individual liver segments and the principle of surgical planning(1) The process of three-dimensional reconstruction of abdominal images:① Download the required CT data from the server and import them to MI-3DVS.② Choose corresponding modules to perform the individual three-dimensional reconstruction.③Finish the reconstruction and amend the models by professional knowledge.(2) The realization of individualized liver segments:based on the Couinaud’s liver segments theory, the image segmentation technology was used to extract the liver, tumor and vessel. The vascular centerline was extracted, the vascular tree was modeled and then the liver was segmented automatically according to the principle of blood-supply by portal vein and blood drainage by hepatic vein. Finally, the liver was three-dimensional visualize.(3) The principle of surgical planning:according to standard oncologic resection, the tumor had to be removed completely with adequate margins (RO resection), which was chosen to be at least 1 cm. The remnant liver volume with sufficient vascular inflow and outflow and the biliary draining system is preserved with adequate drainage when the tumor was removed, the remnant liver can maintain the vital movement of human body.5. The brief introduction of evaluation criteria of related indexes(1) Predicted excisional liver volume, PELV. It was obtained from preoperative surgery planning by MI-3DVS.(2) Actual excisional liver volume, AELV. AELV=|volume from preoperative CT scan-volume from postoperative CT scan|. On the basis of one Nagasue’s finding (Nagasue N, Yukaya H, Ogawa Y, et al. Human liver regeneration after major hepatic resection. A study of normal liver and livers with chronic hepatitis and cirrhosis[J]. Ann Surg,1987,206(1):30-39):it takes about 7 to 10 days for liver to start to regeneration. The liver volume from the postoperative CT scan (within 5 days after hepatic surgery) can be thought to be equal to the volume just after hepatectomy.(3) Absolute error, AE. AE=|PELV-AELV|.(4) Percentage error, PE. PE=AE÷AELV.6. Statistical analysisSimple linear regression analysis was conducted to examine the relationship between AELV and PELV. Linear association was evaluated with the Pearson correlation coefficient (r). Intra-class correlation coefficient was chosen to evaluate the agreement between AELV and PELV. The evaluation of the level of. agreement between AELV and PELV was performed by the method described by Bland and Altman. The statistical analysis was performance with SPSS 13.0 software (SPSSInc, Chicago, IL, USA) and P<0.05 was considered statistically significant.Results1. The difference between actual excisional liver volume and predicted excisional liver volumeOn the whole, the average AELV was 319.9±157.6 cm3 and the average PELV was 298.8±157.6 cm3 with an average absolute error of 21.1±5.9 cm3 (average percentage error,8.0±4.0%). In group A, the average AELV was 350.7±139.0 cm3 and the average PELV was 331.1±143.8 cm3 with an average absolute error of 19.6±4.8 cm3 (average percentage error,6.5±3.2%). In group B, the average AELV was 315.8±161.6 cm3 and the average PELV was 294.5±161.1 cm3 with an average absolute error of 21.4±6.1 cm3 (average percentage error,8.2±4.1%).2. The relationship between actual excisional liver volume and predicted excisional liver volumeThere was a strong positive correlation between the AELV and the PELV in group A and B. The simple regression equation is AELV=1.034×PELV-31.48 in group A (r=0.97;P<0.001) and AELV=0.996xPELV-20.08 in group B (r=0.92; P< 0.001). The ICC in group A is 0.994 and the r is 0.97. The ICC is 0.992 in group B and the r is 0.92. The Bland-Altman analysis showed that PELV achieved excellent agreement with AELV in both groups.3. The results of individualivzed liver segmentsThe individualized liver segments from MI-3VDS are in strict accordance with the Couinaud’s liver segment theory. Compaed with the actual surgery, the digital liver segments remained the same with actual surgery.Conclusions1. Although in both groups, there was a strong positive correlation between the AELV and the PELV, there were still statistically significant differences between them. In consideration of the clinical result:none of the 34 patients had hypohepatia/ hepatic failure after hepatectomy and the mortality was 0% during the perioperative period. The difference could be accepted.2. The individualized liver segments function of MI-3DVS is based on the actual vascular branches and liver shape. The foundation is Couinaud’s liver segments theory. The realization of this process is implemented by the computational technology and the results are real. It provides a new way to study the anatomy of liver.3. MI-3DVS is a valuable surgical assistive tool in surgical planning and it can enhance the safety of hepatic resection. It can be qualified for preoperative surgical planning.Part Ⅲ. A pilot study of a percutaneous puncture navigation systemObjectives1. Exploring the establishment of an optical tracking puncture navigation system.2. A pilot study of the accuracy and stability of the system in practical application.Materials and Methods1. SubjectAn optical tracking puncture navigation system development by our team.2. Equipment and facilities(1) NDI Polaris optical tracking system. (2) Microsoft Kinect depth camera. (3) High performance workstation:Lenovo Thinkpad w540. (4) An abdominal biopsy phantom CIRS Model 057. (5) Abdominal surgical navigation system software, the registeration number is 2014SR000461. (6) Siemens MDCT scanner and its image postprocessing workstation. (7) Several puncture needles.3. The workflow of the navigation system(1) Connecting the devices,seting up the equipments and constructing the navigation system. (2) Acquisition of the phantom CT images through the CT scanner and transmit them to the navigation system. (3) Registration:the purpose of this process is to establish the spatial relationships between the image coordinate system of patients and the coordinate system of tracking equipment and its accuracy impacts the accuracy of the navigation sytem seriously. (4) The CT data analysis, the segmentation of the liver and tumor was performd, an interventional plan was chosen: the puncture path, entry point and entry angle. (5) Puncture under the navigation guide:finding the entry point, adjusting the entry angle and performing the puncture. The interface of the navigation system shows the position of the needle tip, the difference between actual and planning route and the error was shown in real-time. (6) Postoperative CT scan. The distance between the centroid of the trager and the tip of the needle was calculated.4. The evaluation indexes(1) Navigation time:the time-consumed by the navigation procedure. (2) Target Placement Error (TPE):the distance between the centroid of the target and the tip of the needle.5. Statistical analysisThis section does not involve statistical analysis and only present some statistical description.Results1. The time-consumed during navigation system setup is about 15 min.2.72 puncture experiments were performed on 8 "liver tumors" and the TPE is 5.14±2.41mm (range,1.18-11.05mm)Conclusions1. This navigation system is established on the existing operation environment and it is an user-friendly equipment. The operation of this system would not affect the regular work.2. The time-consumed of this system is only about 15 min. The whole procedure is simple and easy to learn.3. The results of total 72 puncture experiments showed that the accuracy of this system is high, the effect is reliable and the stability is satisfied.