DOI：10.3760/cma.j.cn311367-20210601-00309

The first paper regarding clinical application of endoscopic ultrasound (EUS) was published in 1980^［1］. Twelve years later in 1992 the first report of endoscopic ultrasound-guided fine-needle aspiration (EUS-FNA) was published^［2］. Since then the pace of development and innovation has accelerated. EUS is now firmly established as an indispensable tool in clinical practice. In the context of diagnostic EUS imaging, the resolution of EUS images has improved in tandem with technological progress, and image enhancement techniques such as contrast harmonic EUS and EUS elastography are also available now. EUS has an important role in cancer staging, in assessment of gastrointestinal wall lesions, and in evaluation of the pancreaticobiliary system. There is growing interest in replacing EUS-FNA with endoscopic ultrasound-guided fine-needle biopsy (EUS-FNB), in order to more consistently obtain tissue cores for histology and molecular analyses^［3］. There is ongoing research into the application of artificial intelligence (AI) in EUS. AI may be especially important to guide clinical decision making if biopsy results are non-diagnostic^［4-5］. The technique of EUS-FNA established the foundation for EUS-guided therapeutic interventions^［6］. EUS-guided celiac plexus neurolysis and EUS-guided drainage of pseudocysts were the initial application of EUS-guided therapeutics. EUS-guided drainage is now performed for all accessible walled-off intra-abdominal or pelvic collections. Endoscopic ultrasound-guided biliary drainage (EUS-BD) after unsuccessful endoscopic retrograde cholangiopancreatography (ERCP) and endoscopic ultrasound-guided gallbladder drainage (EUS-GBD) for management of acute cholecystitis in non-surgical candidates are now viable management options. EUS-guided drainage provides access for adjunctive endoscopic necrosectomy in the management of walled-off pancreatic necrosis (WON). Other EUS-created gastrointestinal anastomoses such as endoscopic ultrasound-guided gastroenterostomy (EUS-GE) in the context of malignant gastric outlet obstruction (GOO), and endoscopic ultrasound-directed transgastric endoscopic retrograde cholangiopancreatography (EDGE) for Roux-en-Y gastric bypass (RYGB) are technically feasible^［7-8］. Other potential applications for interventional EUS include EUS-guided intravascular interventions for portal hypertension^［9-10］, and EUS-guided ablative therapies^［11］, but robust outcome data remain limited for these indications. This review article focuses on advances in EUS-guided tissue acquisition and EUS-guided drainage, two commonly performed procedures, and discusses how these innovations can be incorporated into current clinical practice.

Established role of diagnostic EUS

EUS has an established role for tumor staging, for evaluation of subepithelial lesions, for detection of small tumors and clarification of the nature of pancreaticobiliary disorders. Even more crucially, EUS-guided tissue acquisition allows tissue diagnosis to be made, which is important to guide management strategies^［12］. Lesions located at previously difficult to access sites such as the mediastinum and pancreas can be easily visualized and biopsied under EUS-guidance. The results of EUS-FNA of solid lesions are generally excellent. A meta-analysis reported that the sensitivity and specificity of EUS-FNA in diagnosing the correct aetiology for solid pancreatic masses were 86.8% (95% confidence interval (CI) 85.5% to 87.9%) and 95.8% (95% CI 94.6% to 96.7%), respectively^［13］. Another meta-analysis reported that the sensitivity and specificity of EUS-FNA were 94% (95% CI 91% to 96%) and 98% (95% CI 96% to 99%) respectively for the diagnosis of enlarged intra-abdominal lymph nodes^［14］. However, the diagnostic yield of EUS-FNA was much lower for pancreatic cystic lesions, with sensitivity of only 54% (95% CI 49% to 59%), although specificity was high at 93% (95% CI 90% to 95%)^［15］.

The trend towards using biopsy needles for EUS-guided tissue acquisition

There is now a shift towards tissue acquisition for histological, rather than simply cytological assessment. This is driven by the unmet needs from the use of cytology, and by the availability of newer generation biopsy needles such as reverse and forward bevel Procore^TM needle (Cook Medical, Bloomington, USA), Acquire^TM needle (Boston Scientific, USA), Sharkcore^TM needle (Medtronic Inc, USA) and Trident^TM needle (Micro-Tech Endoscopy, USA). The tips of these biopsy needles are designed such that tissue cores can be obtained during the puncture process. In addition, in the context of pancreatic cystic lesions, a through-the-needle (TTN) microforceps, Moray^® Micro Forceps (Steris Healthcare, USA) can now be inserted through a 19G fine-needle aspiration (FNA) needle to obtain histology from the cyst wall^［16］.

Although FNA needles have served us well, there are intrinsic limitations. Tissue histology is preferred to cytology for pathological diagnosis. For conditions such as subepithelial lesion, lymphoma, autoimmune pancreatitis (AIP) and malignant tumor of unknown primary that require histology and ancillary studies such as immunohistochemistry and molecular analysis for diagnosis, it is essential to obtain tissue cores for pathological evaluation (Fig. 1 and Fig. 2)^［3］. There is a need to overcome procedural inefficiencies associated with EUS-FNA, such as the need for multiple needle passes^［17］, need for rapid on-site evaluation (ROSE)^［18］ or macroscopic on-site evaluation (MOSE)^［19］. The increasing importance of personalized medicine adds to the demand of tissue availability for molecular diagnosis in addition to pathological diagnosis. A randomized controlled  trial (RCT) demonstrated that the rate of tissue adequacy for genomic profiling was significantly higher with fine-needle biopsy (FNB) needles, compared to FNA needles^［20］.There is a need to further improve the diagnostic yield of EUS-guided tissue acquisition for solid lesions and to address the inadequate diagnostic yield for cystic lesions. The rate of procurement of tissue cores using FNA needles is suboptimal. A 19G FNA needle can be used to obtain tissue cores, but the yield is lower than biopsy needle^［21-22］.

A recent meta-analysis compared EUS-FNA and EUS-FNB of solid lesions. EUS-FNB achieved significantly higher diagnostic accuracy (87% vs. 80%, P=0.020) and tissue core rate (80% vs. 62%, P=0.002). FNB achieved diagnosis with significantly fewer passes than FNA (P=0.030). A high diagnostic yield could be achieved by EUS-FNB without the need for ROSE^［23］. A RCT reported that EUS-FNB had a superior diagnostic yield, and was cost-effective, compared to EUS-FNA^［24］.  The TTN microforceps had a technical success rate of 97.12% (95% CI 93.73% to 98.71%), diagnostic yield of 79.60% (95% CI 72.62% to 85.16%) and accuracy of 82.76% (95% CI 77.80% to 86.80%) for pancreatic cystic lesions. There was a significant improvement in diagnostic yield (odds ratio (OR)=4.79, 95% CI1.52 to 15.06, P=0.007) and accuracy (OR=8.69, 95% CI 1.12 to 67.12, P=0.038) when compared to EUS-FNA^［16］.

There is still a role for FNA needles despite the current focus on biopsy needles. This may be in the context of ①potentially higher risk of bleeding from biopsy due to coagulopathy, ②excessive echoendoscope angulation which necessitates a smaller 25G needle, ③cytology alone being sufficient for diagnosis, ④cyst fluid aspiration, or ⑤when the needle is being used to introduce other devices, such as the TTN microforceps, or for EUS-guided therapeutics. Even with the use of biopsy needles, false negative results can occur. When the clinical context is inconsistent with a negative biopsy result, alternative means of tissue acquisition, including surgical excision biopsy, must be considered.

The potential role of AI to complement diagnostic EUS

There is tremendous interest about the use of AI to enhance the performance of endoscopists in terms of lesion detection and diagnosis. The most robust data are in the context of polyp detection during colonoscopy, with the publication of RCT, and such AI systems are now commercially available^［4］. The research into the use of AI for EUS diagnosis is still in the early phase. AI may have a complementary role to EUS-guided tissue acquisition in the sense that it may provide additional reassurance about true negative results for benign diagnoses or prompt a more aggressive approach to acquire tissue for confirmation if the likelihood of malignancy is high. A EUS-based AI model was developed to predict malignancy in intraductal papillary mucinous neoplasm of the pancreas (IPMN). The model was based on patients who had undergone EUS before pancreatectomy and had pathologically confirmed IPMN. AI prediction of malignancy had sensitivity of 95.7% and specificity of 92.6%. The accuracy of AI (94.0%) exceeded endoscopist diagnosis (56.0%)^［25］.  Another EUS-based AI model was developed to diagnose subepithelial lesions using EUS images of patients who had histopathological diagnoses. For subepithelial lesions less than 20 mm, there was no significant difference between the AI model and EUS experts in terms of accuracy and sensitivity. However, for subepithelial lesions larger than 20 mm, the accuracy (90.0% vs. 53.3%) and sensitivity (91.7% vs. 50.0%) of the AI model were significantly higher than EUS experts^［26］.  Another EUS AI model was developed to differentiate AIP from ductal adenocarcinoma of the pancreas (PAC), chronic pancreatitis (CP) and normal pancreas (NP). The AI model differentiated AIP from NP with sensitivity of 99% and specificity of 98%, AIP from CP with sensitivity of 94% and specificity of 71%, and AIP from PAC with sensitivity of 90% and specificity of 93%. The performance of the AI model for differentiating AIP from other conditions was compared against endoscopist performance using case videos. The AI model had significantly higher sensitivity (88.2% (95% CI 63.6% to 98.5%) vs. 53.8% (95% CI 44.4% to 63.0%)) but similar specificity (82.5% (95% CI 70.1% to 91.3%) vs. 86.7% (95% CI 83.0% to 90.0%))^［27］.  The results of these three early phase studies are very promising and demonstrate the potential of AI as an important supplementary diagnostic tool for EUS.

EUS-guided drainage of pancreatic fluid collections

EUS-guided drainage of pseudocyst was first described in 1992^［28］. Since then there has been significant improvement in endoscope design, stent design and accumulation of robust scientific data concerning its efficacy and safety. EUS-guided drainage is now accepted as the first line treatment option for patients with symptomatic walled-off pancreatic fluid collections^［29］. It has been shown to be superior to non-EUS guided endoscopic drainage in terms of efficacy in RCT^［30-32］, and superior to percutaneous drainage in terms of efficacy, less need for reinterventions and shorter length of hospitalization in two retrospective comparative studies^［33-34］. EUS-guided drainage had similar efficacy as surgical drainage but lower costs and shorter duration of hospitalization in a small RCT^［35］. The endoscopically created fistula also facilitates the insertion of an endoscope into the cavity of WON such that direct endoscopic necrosectomy can be performed. In the context of pseudocysts, traditionally double pigtail plastic stents are utilized for drainage. However, the recent development of lumen apposing metallic stents (LAMS) greatly simplified the procedure^［35］ and this process was further simplified when LAMS with an electrocautery tip in the delivery catheter became available, such that a one-step EUS approach for drainage can be completed within minutes without the need for fluoroscopy^［36］. However, the cost of LAMS is higher, and LAMS would need to be removed within one to two months, to avoid the inner flanges causing trauma within a collapsed cavity^［37-39］. Although there is debate about the incremental value of LAMS over plastic stents for pseudocyst drainage, LAMS with inner diameters of 15 to 20 mm is preferred over plastic stents in the context of WON as it facilitates access for endoscopic necrosectomy and can improve treatment efficacy (Fig. 3)^［40-41］. When plastic stents are used, the fistula opening tends to narrow within days and repeat balloon dilation may be needed to allow entry of the endoscope into the WON cavity for endoscopic necrosectomy. Long term clinical success after EUS-guided drainage has been clearly established^［42］.

EUS-guided biliary drainage 

EUS-BD was first reported in 2001^［43］. A plastic stent was inserted during endoscopic ultrasound-guided choledochoduodenostomy (EUS-CDS). Since then the technique has further evolved. Transmural approaches are most straightforward and include EUS-CDS and endoscopic ultrasound-guided hepaticogastrostomy (EUS-HGS). Less commonly EUS-guided biliary rendezvous and EUS-guided antegrade procedures are performed^［44］. Plastic stents are no longer advised for transmural drainage, due to the risk of bile leak and separation of the bile duct from the duodenal wall, and fully covered self-expandable metallic stents (SEMS) preferred. LAMS have been used for EUS-CDS, as it has the theoretical advantage of more secure apposition of the bile duct to the duodenal wall (Fig. 4). A meta-analysis of nine studies (three RCT studies and six retrospective studies) with 483 patients concluded that while there was no difference in technical success between EUS-BD and percutaneous transhepatic biliary drainage (PTBD), EUS-BD was associated with better clinical success (OR=0.45, 95% CI 0.23 to 0.89), fewer post-procedure adverse events (OR=0.23, 95% CI 0.12 to 0.47) and lower rate of reintervention (OR=0.13, 95% CI 0.07 to 0.24)^［45］. This study supported the role of EUS-BD to replace PTBD as second line procedure after unsuccessful ERCP for palliative drainage of malignant biliary obstruction, if the expertise is available. A meta-analysis of four studies (three RCT studies and one retrospective study) with 302 patients reported no difference in technical success, clinical success or total adverse events between primary ERCP and primary EUS-BD for palliative drainage of malignant biliary obstruction. However, given the excellent track record of ERCP and the availability of expertise, ERCP should remain the first line option for palliative drainage of malignant biliary obstruction^［46］.

EUS-guided gallbladder drainage 

The standard of care for treatment of acute cholecystitis is laparoscopic cholecystectomy. Patients who are not surgical candidates and who fail to respond to antibiotic therapy are treated with percutaneous cholecystostomy^［47］. Endoscopic  transpapillary gallbladder drainage (ET-GBD) by ERCP through cystic duct stenting is another option^［48］. The first reports of EUS-GBD were published in 2007^［48-51］. In these early case reports, a plastic stent was used for drainage, in a manner similar to pseudocyst drainage. Unlike a pseudocyst which is adherent to the gut wall, this is not the case for the gallbladder, and there is a risk of bile leak and perforation. Initial data about the use of a novel LAMS for gallbladder and pseudocyst drainage was published in 2012^［35］. LAMS is now preferred for EUS-GBD because it can securely join the gallbladder wall to the duodenal wall. A meta-analysis compared EUS-GBD with ET-GBD in high-risk surgical patients with acute cholecystitis. Five retrospective cohort studies with a total of 857 patients were included in the analysis. EUS-GBD was associated with higher technical (OR=5.22, 95% CI 2.03 to 13.44, P=0.000 6) and clinical success (OR=4.16, 95% CI 2.00 to 8.66, P=0.000 1) compared to ET-GBD. There was no statistically significant difference in the rate of overall adverse events. EUS-GBD was associated with a lower rate of recurrent cholecystitis (OR=0.33, 95% CI 0.14 to 0.79, P=0.01)^［52］. Another meta-analysis^［53］, which included a single underpowered RCT^［54］, and four retrospective cohort studies, compared EUS-GBD (n= 206) with percutaneous gallbladder drainage (PT-GBD) (n=289). There were no statistically significant differences in technical success (OR=0.43, 95% CI 0.12 to 1.58, P=0.21) or clinical success (OR=1.07, 95% CI 0.36 to 3.16, P=0.90) between both procedures. However, EUS-GBD was associated with fewer adverse events than PT-GBD (OR=0.43, 95% CI 0.18 to 1.00, P=0.05), shorter hospital stay (mean difference of －2.53 d, 95% CI －4.28 d to －0.78 d, P=0.005), and fewer reinterventions (OR=0.16, 95% CI 0.04 to 0.42, P<0.01). A recent adequately powered multicentre RCT (n=80) compared EUS-GBD to PT-GBD in patients deemed to be at high surgical risk. While there were no differences in technical success (97.4% vs. 100.0%, P=0.494) or clinical success (92.3% vs. 92.5%, P=1.000), EUS-GBD significantly reduced both one-year adverse events (25.6% vs. 77.5%, P<0.01) and 30-day adverse events (12.8% vs. 47.5%, P=0.001). Significantly less patients in the EUS-GBD group had recurrent acute cholecystitis in one year (2.6% vs. 20.0%, P=0.029)^［55］. A retrospective study with propensity score matching compared EUS-GBD (n=30) with laparoscopic cholecystectomy (n=30) for acute cholecystitis. Clinical outcomes were similar in terms of technical success rate, clinical success rate, hospital stay, 30-day adverse events and mortality^［56］.

EUS-guided gastroenterostomy

Patients with malignant GOO not amenable to curative surgery are usually palliated with endoscopic stenting. For patients with better prognosis and a good functional status, a surgical gastroenterostomy would be preferred as there is a risk of tumour invasion and occlusion following insertion of SEMS with the passage of time, especially if survival exceeds three months^［57］. The feasibility of EUS-GE using LAMS was demonstrated in a porcine study in 2012^［58］. Since then human feasibility studies have been performed and it is being used in clinical practice as a treatment option. Unlike the case of pseudocyst or gallbladder drainage, which are straight forward procedures, and involves simply visualizing and puncturing the target site of drainage to deploy the LAMS, it is somewhat technically more challenging to create an anastomosis between the stomach and proximal jejunum under EUS-guidance. The jejunum is not distended nor adherent to the stomach. To perform EUS-GE, either a balloon or catheter would need to be inserted past the GOO, the jejunal loop of interest distended with water and brought to close proximity with the gastric wall so that it can be visualized by EUS, before the LAMS can be inserted and deployed (Fig. 5)^［59］. The clinical outcomes were addressed in a recent meta-analysis. Twelve studies with 285 patients were included. These studies were mainly retrospective case reports and case series, with only a single prospective case series. The clinical success rate was 90% (95% CI 85% to 94%) with an adverse event rate of 12% (95% CI 8% to 16%)^［60］. EUS-GE was compared with open surgery and laparoscopic gastroenterostomy in two retrospective studies^［61-62］. The key findings were similar success rates, but shorter time to resumption of oral intake and lower costs with EUS-GE.

EUS-directed transgastric ERCP

It can be particularly challenging to perform ERCP in patients with surgically altered anatomy, such as RYGB. Dependent on the length of the roux limb, it may not even be possible to insert an enteroscope to the required depth to visualize the papilla. The clinical success rates of single and double balloon enteroscope assisted ERCP were reported to be only 61.70% and 63.55%, respectively^［63］. Laparoscopy-assisted endoscopic retrograde cholangiopancreatography (LA-ERCP) is the other minimally invasive option for definitive treatment. In patients with unresectable malignant biliary obstruction, PTBD with internalization of stents can be performed for palliative drainage. However, in the context of benign pathologies such as bile duct stones, either laparoscopic or open surgical exploration of the bile duct would be needed for definitive treatment. EDGE uses EUS to access the bypassed stomach in RYGB, create a fistulous tract secured with LAMS, and then perform ERCP through the LAMS (Fig. 6). This procedure was first reported in 2015 in five patients. It was advocated as a cost-effective, minimally invasive option for patients with RYGB^［64］. Since that index publication, further retrospective cases series were published. The results were summarized in a recent systematic review which included nine studies with a total of 169 patients. The technical success rate for being able to perform ERCP was 98%. Minor complications developed in 18.3%, moderate complications in 5.3% and severe complications in 0.6%^［65］. A retrospective cohort study compared EDGE with LA-ERCP and deep enteroscopy-assisted endoscopic retrograde cholangiopancreatography (E-ERCP) for patients with RYGB. The technical success rate for patients in the EDGE group was 100% (n=26), compared with 94% (n=17) and 75% (n=9) in the LA-ERCP and E-ERCP groups (P=0.02), respectively^［66］. Further refinements are necessary for EDGE to be more widely adopted, and the procedure is more relevant in populations with high prevalence of RYGB. It must be remembered that LA-ERCP is technically very easy, with high clinical success rates and low morbidity. A team approach involving both endoscopist and surgeon can easily be adopted, if the surgeon does not perform ERCP. A recent cohort study (n=131) reported success rate of 100%  in biliary cannulation and 99.2% clinical success rate. A single patient had severe chronic marginal ulceration on endoscopic evaluation, necessitating a completion gastrectomy shortly after LA-ERCP. Surgical site infection developed in 9.2% and post-ERCP pancreatitis in 3.8%^［67］. 

Conclusions

A holistic multidisciplinary management approach should be adopted. Treatment must be individualized based on specific needs of the patient and the available expertise. EUS-guided drainage is currently the first line treatment option for symptomatic walled-off pancreatic fluid collections. EUS-BD is a viable alternative to PTBD after failed ERCP for palliation of non-surgical candidates with malignant biliary obstruction. EUS-GBD may be preferred to percutaneous cholecystostomy in non-surgical candidates with acute cholecystitis. EUS-GE is a very promising treatment for malignant GOO, and is particularly useful for non-surgical candidates who have recurrent obstruction due to tumour ingrowth after SEMS insertion. EDGE is not yet ready for prime time, given the limited data and the ease and efficacy of LA-ERCP.