Sage Journals: Discover world-class research

Abstract

The utilization of artificial intelligence (AI) in gastrointestinal (GI) endoscopy has witnessed significant progress and promising results in recent years worldwide. From 2019 to 2023, the European Society of Gastrointestinal Endoscopy has released multiple guidelines/consensus with recommendations on integrating AI for detecting and classifying lesions in practical endoscopy. In Vietnam, since 2019, several preliminary studies have been conducted to develop AI algorithms for GI endoscopy, focusing on lesion detection. These studies have yielded high accuracy results ranging from 86% to 92%. For upper GI endoscopy, ongoing research directions comprise image quality assessment, detection of anatomical landmarks, simulating image-enhanced endoscopy, and semi-automated tools supporting the delineation of GI lesions on endoscopic images. For lower GI endoscopy, most studies focus on developing AI algorithms for colorectal polyps’ detection and classification based on the risk of malignancy. In conclusion, the application of AI in this field represents a promising research direction, presenting challenges and opportunities for real-world implementation within the Vietnamese healthcare context.

Keywords

artificial intelligence colorectal polyp gastrointestinal endoscopy upper GI lesion

Introduction

Gastrointestinal (GI) diseases are associated with high morbidity, mortality, and economic burden, with the total expenditure for GI-related conditions reaching up to $119.6 billion.¹ According to GLOBOCAN 2020, GI cancers, including those of the esophagus, stomach, and colon, rank among the top 10 malignant diseases in terms of both prevalence and mortality rates.² However, if detected at an early stage, the 5-year survival rate of patients can be as high as 80%–90%.^3,4

Endoscopy is essential for diagnosing, screening, and treating various GI diseases, especially GI cancers.^5
–7 However, precancerous or high-risk lesions can be missed during endoscopy. For example, the missed rate of gastric cancer is reported to be 9.4%.⁸ A meta-analysis of 45 studies involving 15,000 colonoscopies reported miss rates of 26% for adenomas, 9% for advanced adenomas, and 27% for serrated polyps.⁹ Therefore, technological solutions are needed to narrow the experience gap between endoscopists and limit medical errors.

Artificial intelligence (AI) has been increasingly applied to various fields of medicine, including endoscopy.¹⁰ Studies on AI applications in endoscopy have shown that computer-aided endoscopy has good accuracy in diagnosing and classifying GI lesions such as colorectal polyps, Barrett’s esophagus, gastric cancer, and esophageal cancers.^11,12 The accuracy is markedly high on still images and videos. In 2022, The European Gastrointestinal Endoscopy Society (ESGE) published scopes for different applications of AI in endoscopy, including detecting and treating neoplastic lesions, diagnosing atrophy, dysplasia, or precancerous conditions. ESGE also emphasized that an AI system can be used in real practice if it achieves a threshold equivalent to experienced endoscopists.¹³

In Vietnam, GI cancers also pose a significant health burden, with colorectal and gastric cancers ranking fourth and fifth, respectively, with over 16,000 new cases in 2022.^2,14 Additionally, many factors may lead to an increased miss rate of GI lesions in Vietnam: overcrowding—the number of daily procedures in major endoscopy centers can be up to 300 cases—and variations in the quality of endoscopy systems across different healthcare facilities.¹⁵ A survey conducted among healthcare professionals specializing in internal medicine, gastroenterology, and GI endoscopy showed that 98% of respondents expressed their need for AI support applications during endoscopy, with a preference for automated lesion detection and delineation.¹⁵ These findings underscore the necessity of developing big data and implementing AI applications in the field of endoscopy in Vietnam.

Overview of AI in gastrointestinal endoscopy worldwide

AI refers to computer systems that perform tasks normally associated with human intelligence. One branch of AI, Machine Learning (ML), focuses on developing algorithms capable of making predictions from large datasets (“Big Data”). Deep learning (DL), a further specialization of ML, utilizes deep neural network structures to replicate the functionality of the human brain. This technology enables the interpretation of large and complex datasets, especially image data. In the endoscopy field, DL is often used in analyzing endoscopic images to create algorithms for detecting and classifying lesions.¹⁶

Building databases

The development of effective AI algorithms for detecting anatomical landmarks and lesions in GI endoscopy depends on the quality of the training dataset. The requirements for the dataset include a large number of images with diversity in aspects including anatomical landmarks; lesions shapes, types, and sizes¹⁷; light modes (white light and images-enhanced modes); image quality (brightness and resolution); and endoscopy systems.^18
–20 Experience in the implementation of AI shows that ensuring a standard resolution (ideally more than 512 × 512 pixels for optimal performance) will facilitate the training of high-quality algorithms for clinical deployment.^18,19,21 Furthermore, the images in the dataset need to be annotated, labeled, and validated by endoscopy experts. These steps are vital to ensure the accuracy and reliability of the dataset, which in turn directly impacts the performance of the AI algorithms developed from it.^17,22,23

Currently, numerous GI endoscopic databases have been published. Most of these databases are open-access, rendering them valuable resources for algorithm training and development. A recent study has summarized the existence of 40 published still image databases, with 34 being open access. Of these, 16 databases focus on polyps, while six focus on small intestinal lesions.²⁴

For the upper GI tract, published endoscopic image datasets encompass various lesion types and anatomical landmarks.^25
–27 These databases mainly consist of still images with varying characteristics for cleanliness, light modes, image quality, and resolutions. However, certain databases still exhibit limitations, such as partially obscured images due to endoscopic instruments, low image resolution, and a lack of lesion annotation and labeling.^26,28,29

As for the lower GI tract, researchers focus on building image and video datasets of colorectal polyps classified and labeled by experts.^30
–35 However, these datasets are limited in diversity and polyp size and lack images/videos captured in enhanced light mode. Some datasets have relatively low image resolutions ranging from 288 × 384 pixels to 480 × 409 pixels.^32,35,36

AI algorithms developed in gastrointestinal endoscopy

According to the ESGE 2022 guidelines, AI algorithms in GI endoscopy include automated detection of anatomical landmarks, quality assessment of endoscopy, and detection and classification of GI lesions.¹³

For the upper GI tract, the research direction of AI focuses on the detection of anatomical landmarks, enhancement of image quality, and detection of precancerous or early-stage lesions.³⁷ First, algorithms for detecting anatomical landmarks of upper GI endoscopy play a vital role in lesion identification, particularly in regions such as the pylorus, Z-line, and cecum.³⁸ Some algorithms categorized the anatomical landmarks of upper GI endoscopy into seven to eleven main zones according to guidelines issued by the World Endoscopy Organization.^39
–41 Moreover, researchers have further divided the stomach into 26 zones to reduce blind spots during endoscopy.⁴² Notably, the accuracy of these AI algorithms consistently exceeds 90%.^38
–42 Several clinical trials have shown that AI-integrated endoscopy systems have significantly lower rates of blind spots (ranging from 2.8 to 5.5%) than conventional endoscopy (ranging from 3.92 to 22.46%).^42
–44 Second, enhancing image quality is crucial in effective AI training. Techniques such as flipping, cropping, resizing, and blurring existing images have been used to diversify and enrich input datasets for AI model optimization.²⁰ Third, for lesion detection, researchers have shown that AI has high accuracy in detecting gastric neoplastic lesions, Barrett’s esophagus, esophageal squamous cell carcinoma, chronic atrophic gastritis, H. pylori infection, and gastric cancer. While most studies were conducted with still images,^37,45
–47 clinical trials have been conducted to assess the application of AI in real-world scenarios. In two clinical trials evaluating AI application in the upper GI, tract, the rates of missed neoplasm gastric lesions (6.1%) and cancer lesions (1.7%) in the AI group were significantly lower than the conventional endoscopy group (27.3% and 6.7%, respectively).^48,49

For the lower GI tract, the principal research directions of AI are centered on cleanliness assessment, detection, and characterization of colonic adenomas/polyps. Studies on using AI to support the assessment of bowel preparation and mucosal cleanliness have reported accuracy ranging from 85% to 95%, with no significant difference between the assessment of endoscopy experts and AI.^50
–52 In clinical trials, AI-supported colonoscopy achieved a lower adenoma miss rate and a higher adenomas/polyps detection rate than the conventional group, especially for small, flat lesions.^53
–58 Regarding polyp classification, a marked increasing number of clinical trials provide evidence of AI’s advantages in detecting small lesions (<5 mm).^59,60 Several AI systems (such as CAD-EYE, EndoBrain, and GI Genius) have been commercialized in the market.⁶¹ (Table 1)

Table 1.

Parameters for evaluating some AI algorithms in GI endoscopy.

AI feature	Sensitivity (%)	Specificity (%)	Accuracy/AUC (%)	Light mode
Upper GI endoscopy
Detecting Barrett’s esophagus³⁷	84–97	80–100	88–95	WLI, NBI
Detecting and predicting esophageal squamous cell carcinoma invasive degree³⁷	71–100	50–96	63–99	WLI, NBI, BLI, ME-NBI/BLI
Detecting esophageal squamous neoplasm⁴⁵	—	—	88 (82–96)	WLI, NBI
Detecting gastric neoplasm⁴⁵	—	—	96 (94–99)	WLI, NBI
Detecting H. pylori status³⁷	67–92	60–99	66–96	WLI, NBI
Detecting and predicting esophageal cancer invasive degree³⁷	28–98	71–100	75–99	WLI, ME-NBI/BLI
Chronic atrophic gastric inflammation⁴⁶	94 (88–97)	96 (88–98)	98 (96–99)	WLI, LCI, BLI, NBI
Detecting gastric cancer⁴⁷	91–98	91–100	90–98	WLI, NBI, BLI
Lower GI endoscopy
Detecting polyp⁶²	70–99	63–99	77–98	WLI
Characterizing polyps⁶²	90–98	70–94	90–96	WLI
Characterizing neoplastic and non-neoplastic polyp⁶³	92 (88–95)	82 (71–89)	—	WLI
Detecting rectum’s ulcer⁶⁴	74 (65–81)	97 (95–99)	91 (83–95)	WLI

AUC, Area Under the Receiver Operating Curve; BLI, Blue Light Imaging; GI, gastrointestinal; LCI, Linked Color Imaging; ME, Magnification endoscopy; NBI, Narrow Band Imaging; WLI, White Light Imaging.

AI research in endoscopy in Vietnam

Since 2019, studies regarding AI applications in the field of GI endoscopy in Vietnam have been conducted using various methodologies. The primary focuses have been developing comprehensive databases of morphology and light modes and constructing AI algorithms for lesion detection using these databases.

Building databases for AI in Vietnam

In Vietnam, the development of image databases for AI training in endoscopy remains limited. To our knowledge, three databases have been established. Two of these databases are data collected from Vietnamese subjects, while the remaining one is a database that employs AI algorithms to annotate lesions within pre-existing datasets.

In 2019, Hoang Trung Hieu and colleagues utilized AI algorithms to annotate features such as polyps, dyed polyps, dyed-resection polyps, and instruments for 4470 unclassified images from the KVASIR and BioMedia 2019 Development datasets. This database has been made publicly accessible and available at https://endoscopy.selab.hcmus.edu.vn (Figure 1).⁶⁵ However, this study did not provide details on the role and contribution of expert knowledge in the review and approval of the AI-generated labels, which may potentially affect the accuracy of the labeled images.

Figure 1.

Endoscopy images of colorectal polyps with AI-generated labels.⁶⁵

To establish an endoscopic image database specifically for Vietnam, serving as educational materials for endoscopy and a foundation for AI training, Dao Viet Hang and colleagues have built two databases of endoscopy images for upper and lower GI tracts since 2020.

For lower GI endoscopy, the database comprises 8190 colonoscopy images with polyps and 4000 images without polyps in white light and flexible spectral imaging color enhancement (FICE) modes. The polyps in these images are delineated and labeled by endoscopists with >5 years of experience.⁶⁶ A portion of the dataset has been published with 1200 polyp images classified and labeled for neoplastic and non-neoplastic lesions, namely BKAI-IGH NeoPolyp.⁶⁷ (Figure 2)

Figure 2.

The BKAI-IGH NeoPolyp dataset and classification labels.⁶⁷

For upper GI endoscopy, in 2022, Dao Viet Hang’s research team developed a database of 8207 endoscopy images in four different light modes, including white light imaging and three image-enhanced endoscopy (IEE) modes: FICE, Blue Light Imaging (BLI), and Linked-Colored Imaging. Among these, 5546 images are upper GI images without lesions of 10 anatomical landmarks (from the pharynx to the duodenum). The remaining images have at least one of five important upper GI lesions including erosive esophagitis, esophageal cancer, gastric cancer, gastritis, and duodenal ulcers (Figure 3).⁶⁸ All images with lesions have been classified, delineated, and labeled by endoscopy experts. The distinguishing features of these upper and lower GI datasets,^66,68 when compared to existing datasets worldwide, are the addition of advanced IEE modes (BLI and LCI) and the expert delineation and labeling of lesions.

Figure 3.

The upper GI endoscopy image database storage system has been labeled.

Initial research of developing AI algorithms in gastrointestinal endoscopy

Researchers in Vietnam have so far implemented studies to develop AI algorithms to improve image quality, identify anatomical locations, detect and delineate lesions, and classify lesions on endoscopy.

Upper GI endoscopy

Bubbles, mucus, or residual food debris in the GI tract are common issues during endoscopy, as these factors can obscure lesions and hinder observation and lesion identification. An algorithm has been developed to assess gastric cleanliness by delineating bubbles during endoscopy. The accuracy of this algorithm was 90% on still images.⁶⁹

To enhance the algorithm training quality and save resources, AI algorithms for simulating IEE modes have been tested and shown promising results. To evaluate the similarity between the images in authentic IEE modes and simulated-IEE modes, we mixed these two types of images and asked the endoscopists to detect which IEE image was simulated. The detection rates of new (<5 years of experience) and experienced endoscopists (⩾5 years of experience) were relatively low (54.7% and 52.7%, respectively).⁷⁰

Identifying anatomical landmarks is a critical task when implementing ML systems to support endoscopy procedures. For anatomical landmark identification, the accuracy of the algorithms developed in Vietnam for detecting 10 anatomical locations from the pharynx to the duodenum is very high, ranging from 97.0% to 98.6% when run on different datasets and network systems.^68,71,72 This algorithm has met the minimum required for anatomical landmarks detection during GI endoscopy procedures, similar to the ESGE guidelines.⁷³ Algorithms detecting anatomical landmarks in upper GI endoscopy have been widely developed and clinically tested on patients with promising results (accuracy > 90%). Some algorithms have also been enhanced with a feature that detects blind spots.^38
–42 Compared to the algorithms developed in Vietnam, the algorithm by Qi He et al. has similar accuracy in anatomical landmarks detection. The WISENSE algorithm classifies and detects blind spots in upper GI endoscopy with a more detailed dataset, up to 26 separate regions of the stomach. However, the dataset lacks data in other positions such as the esophagus and the duodenum.⁴² Other algorithms detect fewer anatomical landmarks. The algorithm by Cogan et al. was designed to detect three basic locations (pylorus, Z-line, and cecum). The algorithm of Takiyama et al. detected seven locations (larynx, esophagus, stomach, duodenum, etc.) while the study by Seong Ji Choi et al. detected eight locations (proximal esophagus; Z-line; stomach cardia and fundus on retroflexed view; stomach body; stomach angle; stomach antrum; duodenal bulb; the second part of duodenum).^39,40 In the future, researchers in Vietnam may refine algorithms for detecting anatomical landmarks by integrating additional features such as blind spot detection.

In 2022, the EndoUnet algorithm, equipped with features for lesion classification, delineation, and labeling, was developed using the upper GI dataset by Hang et al. The overall accuracy of the algorithm was 98.6%. For each of the five important upper GI lesions, the accuracy was 95.1% for erosive esophagitis, 96.9% for esophageal cancer, 88.1% for gastric cancer, 96.9% for gastritis, and 81.9%) for duodenal ulcer.⁷⁴ For the lesion delineating and labeling task, test results on three convolutional neural networks including ResNet-50, DenseNet-121, and VGG-19 showed high accuracy. The two lesions with the most accurate delineation were gastric cancer and esophageal cancer (Dice scores of 87% and 83%, respectively).⁶⁸ Specifically for erosive esophagitis, the author team of Dao Viet Hang and Tran Thanh Hai developed lesion delineation algorithms based on differences in the color and structure of the lesions and the surrounding normal mucosa. The delineation performance of the algorithm needs improvement for erosive esophagitis grades A and M due to the similar characteristics between these lesions and surrounding mucosa.^75,76 Factors related to missed lesions include small lesion sizes and poor mucosal cleanliness. Images with poor cleanliness and those diagnosed with gastric metaplasia or Barrett’s esophagus increase the false-positive rate.⁷⁷ Study results showed that algorithms developed in Vietnam had similar accuracy to previous studies in detecting upper GI lesions (Table 2).^37,45
–47 However, the datasets and algorithms in Vietnam still need to be improved in the detection of other lesions such as Barrett’s esophagus, esophageal neoplasia, and gastric neoplasia, especially lesions of small sizes and at early stages. Besides lesion detection, developing algorithms to classify lesions according to international standards is a potential direction to comprehensively support endoscopists.

Table 2.

Some results of AI applications in GI endoscopy in Vietnam.

Authors	Algorithm/Deep learning network	Features	Results (%)	Dataset	Resolution
Upper GI endoscopy
Tran Thanh Hai (2019)⁷²	MobileNet-V2 and ResNet-18	Anatomical landmarks classification	Sp: 95.9–97.4Se: 99.6–99.7Accuracy: 95.9–97.4	Still images, WLI	1280 × 1024
Ha Van Kien (2020)⁶⁹	Unet	Identify and delineate bubbles	Precision: 91 ± 3Recall: 89 ± 3	Still images, WLI
Phuong Thao Nguyen (2021)⁷⁸	StyleGAN2-ADA	Improving Gastroesophageal Reflux Diseases Classification	Overall accuracy: 91.7%	Still images, WLI	1280 × 1024
Hoang Manh Hung (2022)⁷⁰	CycleGAN	Enhanced-image generation	Accuracy:Junior endoscopists: 54.7%Senior endoscopists: 52.7%	Still imagesWLI, FICE, BLI, LCI	720 × 1280
Hai Vu (2022)⁷¹	CycleGAN	Anatomical landmarks classification	Se: 96.3Sp: 99.6Accuracy: 99.3	- Training: still images- Testing: videoWLI, FICE, BLI, LCI	—
Dao Viet Hang (2022)⁶⁸	EndoUNet	- Anatomical landmark classification- Lesions classification- Lesions segmentation	- Anatomical landmarks classification: 97.0–98.3- Lesions segmentation (DICE score):Esophagus cancer (EC) (78–87), gastric cancer (GC) (83–85), Erosive esophagitis (EE) (42–48), Duodenal ulcer (DU) (58–72),- Lesions classification: 99.5–99.8	Still imagesWLI, FICE	720 × 1280
Phuong Thao Nguyen (2022)⁷⁴	ResNet-50 + DA + FL	Lesions classification	- Precision: DU (81.9); GC (88.1); EC (96.9); Gastritis (92.5); EE (95.1); Normal (99.8)- Recall: DU (92.8); GC (84.4); EC (90.1); Gastritis (94.0); EE (94.8); Normal (99.9)- Accuracy: 98.6	- Training: still images- Testing: videoWLI	1280 × 1024
Dao Viet Hang (2023)⁷⁷	Yolo V8	Erosive esophagitis identification	Se (70.8), Sp (92.6), PPV (90.5), NPV (76), Accuracy (81.7)	Still imagesWLI, FICE, BLI, LCI	720 × 1280
Tran Thanh Hai (2023)^75,76	DCS-UNet	Erosive Esophagitis Segmentation	DICE score- Grade M: 50–52- Grade A: 42–46- Grade B: 55–56	Still imagesWLI, FICE, BLI, LCI	1280 × 960
Bao Anh Trinh (2023)⁷⁹	FocalFPNet Model	Improving Lesion Segmentation	- Esophageal cancer: DICE 0.865; IoU 0.833; Recall 0.826; Precision: 0.908- Peptic Ulcer: DICE: 0.700; IoU: 0.724; Recall: 0.697; Precision: 0.704	Still imagesWLI, FICE, BLI, LCI	1280 × 995 and 1280 × 1024
Lower GI endoscopy
Dao Viet Hang (2020)^66,80	EfficientNet	Polyp detection	- Validation in pixels after training: Se (96.4), Sp (99.8), PPV (94.6), F1 (>95)- Testing dataset: Se (97.6), Sp (94.3), PPV (94.4), NPV (97.5), Accuracy (95.9)	Still imagesWLI, FICE	720 × 1280
Phan Ngoc Lan (2021)⁸¹	NeoUNet	- Polyp segmentation- Identifying polyps at high risk of malignancy	- DICE (segmentation): 0.911- DICE (non-neoplastic polyp): 0.72- DICE (neoplastic polyp): 0.889	Still imagesWLI	—
Nguyen Sy An (2022)⁸²	BlazeNeo	- Polyp segmentation- Identifying polyps at high risk of malignancy	- DICE (segmentation): 0.904- DICE (non-neoplastic polyp): 0.717- DICE (neoplastic polyp): 0.889	Still imagesWLI	—
Trong Hieu Nguyen Mau (2022)⁸³	ConvNeXt	Polyp identification and segmentation	DICE: 88IoU: 82	Still imagesWLI	From 384 × 288 to 1225 × 966
Trong Hieu Nguyen Mau (2022)⁸⁴	ConvTransNet	Polyp identification and segmentation	DICE: 93IoU: 88	Still imagesWLI	From 384 × 288 to 1225 × 966
Kieu Dang Nam (2022)⁸⁵	Resnet101+ DeeplabV2	Improving algorithm’s efficiency on BLI images	DICE:83.8IoU: 91.2	Still imagesWLI, BLI	1280 × 995
Dinh Viet Sang (2022)^86,87	AG-CUResNeSt	Polyp segmentation	DICE: 70.1–94.6IoU: 61.3–90.2	Still imagesWLI	From 384 × 288 to 1225 × 966
Dinh Viet Sang (2022)^88,89	ImageNet + AG-ResUNet	Labeling for polyp segmentation	DICE: 70.2–86.5IoU: 77.2–91.6	Still imagesWLI	From 384 × 288 to 1225 × 966
Nguyen Thanh Duc (2022)⁹⁰	ColonFormer	Polyp segmentation and feature description	DICE: 80.1–92.7IoU: 72.2–93.2	Still imagesWLI	From 384 × 288 to 1225 × 966
Quoc-Huy Trinh (2023)⁹¹	M2UNet and MU	polyp segmentation	DICE: 67.0–90.7IoU: 59.5–85.5	Still imagesWLI	—

BLI, Blue Light Imaging; DICE, DICE Score/DICE Similarity Coefficient; FICE, Flexible spectral Imaging Color Enhancement; IoU, Intersection over Union; LCI, Linked Color Imaging; px, pixels; Se, Sensitivity; Sp, Specificity; WLI, White Light Imaging.

Lower GI endoscopy

Based on a database of colorectal polyps delineated and labeled by experts, Dao Viet Hang and colleagues developed the EfficientNet algorithm for detecting colorectal polyps. When tested on still images, initial results showed the algorithm’s sensitivity and positive predictive value to be 96.4% and 94.6%, respectively.⁶⁶ This algorithm continued to be evaluated for accuracy on a different dataset of still images, including 2000 images with polyps and 2000 without polyps. Trong Hieu et al. used the Kvasir-SEG dataset and developed an algorithm for detecting and delineating colorectal polyps that achieved high accuracy when identifying polyps, with Dice and IoU scores of 88.2% and 81.6%, respectively.⁶⁶ For polyp delineation and classification, the accuracy of the AI algorithm using the BlazeNeo and NeoUnet networks on non-neoplastic polyps were 71.7% and 72.0%, respectively. As for neoplastic polyps, the accuracy of BlazeNeo and NeoUnet were 88.5% and 88.9%. When compared to algorithms using other deep learning networks worldwide, the accuracy of these two algorithms was superior, with a true delineation rate of 90.4% and 91.1%. The processing speed of the algorithm using the BlazeNeo network was slower than that of the NeoUnet network and other deep learning networks.^81,82

To enhance the accuracy of polyp delineation, researchers have adopted various approaches. The methods include using data from WLI images to train polyp segmentation models for BLI images or consistently updating the AI algorithm while collecting images for the dataset.^85,88,89 Some algorithms and deep learning networks have been developed and improved, such as AG-CUResNeSt, MetaFormer (M2UNet), ConvTransNet, and ColonFormer, to improve the accuracy of polyp delineation and describing polyp characteristics.^{83,86,87,91,90} These algorithms and deep learning networks have all demonstrated high accuracy when tested on different polyp datasets.

Besides initial studies with algorithms based on Vietnamese patient databases, Dao Viet Hang et al. conducted a clinical trial on an AI-integrated support system for detecting colorectal polyps developed by the University of Hong Kong. The study participants underwent tandem endoscopy to evaluate the adenoma/polyp detection and missed rates. The research results showed that endoscopy with AI support increased the adenoma detection rate but did not reduce the miss rate compared to the traditional group.⁹²

Almost all studies on developing AI algorithms for endoscopy in Vietnam are currently at the stages of development, calibration, and evaluation on still-image datasets. To be commercialized, an AI algorithm must undergo a comprehensive evaluation of performance and accuracy in real-time settings using endoscopy video datasets.²³ After integration into a device, the algorithm must be examined in clinical trials to compare its effectiveness against novice and expert endoscopists.

Challenges in applying AI in endoscopy

Datasets

The most crucial aspect of developing an AI model with good performance is building a large-scale, high-quality dataset of endoscopy images, annotated and labeled by expert endoscopists experts.²⁰ However, to guarantee the diversity of images, endoscopic images in Vietnam are collected from multiple centers across the country, resulting in several challenges when constructing a comprehensive dataset from multiple healthcare facilities in Vietnam, such as including inconsistency in image quality, difficulties in data extraction, and significant resources required for data cleaning and labeling.¹⁵

First, the healthcare facilities in Vietnam are not uniform in their data storage processes. Most healthcare workers face difficulties in retrieving endoscopy results. The data storage system in Vietnam is still limited, which does not allow for bulk extraction of images and endoscopy results.¹⁵ Additionally, implementing data collection at multiple healthcare facilities requires strict adherence to patient confidentiality policies and approvals from ethical committees and data management authorities.^93
–95 If the data from healthcare facilities is accessible to multiple individuals and organizations, it can increase the risk of privacy and security breaches. Therefore, researchers must establish robust and transparent management systems and procedures to ensure privacy and data security.^96,97 Developing a large dataset for scientific research and AI development research, it is necessary to establish and clarify clear regulations and policies on data security and sharing among collaborating units, and to create an open database among healthcare facilities. Technical mechanisms for sharing, such as cloud servers and access granting, also need to be established.

Secondly, focusing on the data cleaning and labeling process is crucial to developing a high-quality input dataset for training AI algorithms. Image quality can be influenced by various factors, including doctor expertise, available endoscopic systems (generation, type of endoscopes, light modes, resolutions, etc.), level of inflation, cleanliness, artifacts, capture angles, and experiences of endoscopists, etc.^18,98 Therefore, the image dataset must be thoroughly cleaned to maintain patient confidentiality and meet quality standards before proceeding to the labeling process. Another common issue is the potential for bias due to the heterogeneity in physicians’ perspectives, which can lead to incorrect classification and annotation during the labeling stage.^20,21 Furthermore, manual labeling is also time-consuming for doctors. Therefore, developing and sharing standardized annotation tools and common reference standards with clinicians before labeling is important to ensure data quality and save time.^20,99,100 However, establishing common standards may also lead to “overfitting” on the training when the dataset is used for training, resulting in poor performance on validation datasets or actual patients.⁹⁹ To our knowledge, the criteria to collect endoscopic datasets for AI training in Vietnam to ensure diversity include images with both high and low resolution, diverse in lesions (types and morphology), and light modes (WLI and enhance-light mode) (Table 2). To ensure the quality of the input dataset, some studies also collect images that met cleanliness criteria (according to commonly used scales in endoscopy such as BBPS¹⁰¹ and MVS¹⁰²) and annotations reviewed and verified by experts before training^77,81,80 (Supplemental 1).

Resources

Developing AI systems in endoscopy requires highly skilled AI engineers with the knowledge and practical experience in the latest technology advancements. The application of AI in healthcare, particularly in endoscopy, necessitates close collaboration between IT and healthcare professionals to optimize performance.¹⁰³ Vietnam currently has 140 IT training programs, including 57 courses related to AI applications. However, the number of IT professionals and experts in medical technology and big data development remains limited due to inadequate incentives.^103,104

To date, several products have been developed to support real-time lesion detection and characterization in endoscopy, such as Fujifilm’s CADEYE, Olympus’s EndoBrain, and Medtronic’s GI Genius. However, these products require the most advanced endoscopy systems, high resolution input, and high cost.^105,106 While this is a major barrier to the application of AI in healthcare in Vietnam, especially in low-resource centers, it also presents opportunities for researchers in Vietnam to develop AI systems that can be integrated into endoscopic systems with various specifications and image qualities, so that AI technologies can be available to most healthcare facilities.

The body of evidence on AI in healthcare in Vietnam is growing. Most studies were funded by commercial organizations, academic institutions, and international grants, and only a few were funded by national grants.¹⁰⁴ In recent years, the Vietnamese government has begun to prioritize AI development and its applications across various sectors through key national projects. The expansion of international partnerships has offered Vietnamese researchers valuable opportunities to align their works with global AI advancements.¹⁰⁷

Legal and ethical policies

Despite a relatively young field of science, AI application and digitalization in healthcare have been rapidly growing, prompting the establishment of policies and regulations. In other countries, guidelines for digitalization and AI application in clinical practice have been developed in the general context and specific areas of healthcare.^108,109 In Vietnam, the Prime Minister has issued the “National Strategy for Research, Development, and Application of Artificial Intelligence until 2030” to promote AI research, development, and application. In 2020, the Ministry of Health approved the “Digital Transformation Program in Healthcare until 2025” with a vision to establish a smart healthcare system.^107,110 Despite these efforts, Vietnam currently lacks official documentation providing guidelines for the development, evaluation, and implementation of AI algorithms in healthcare. The World Endoscopy Organization has identified challenges in deploying AI-integrated medical devices in clinical practice. These include insufficient data on clinical benefits and cost-effectiveness, lack of reliable guidelines, unclear indications, AI usage costs, and training requirements.¹¹¹ Additionally, the integration of AI in clinical settings raises ethical concerns that demand careful consideration, such as potential harm, legal responsibilities, bias in clinical decisions, impacts on doctor-patient relationships, and workforce implications. According to a recent survey in Vietnam, only 59,3% of endoscopists showed high acceptance of AI-integrated endoscopy systems.¹¹² Endoscopy experts in Asia have also emphasized the need for guidelines on conducting research and applying AI in the field of endoscopy.^113,114

Conclusion and recommendation

Although still in the early stages of applying AI in endoscopy, Vietnamese researchers have applied already developed algorithms and have also started developing Vietnamese databases and algorithms. Some notable accomplishments include building endoscopic image databases that meet the qualified standards and diversity (lesions, light mode, resolution, etc.), and labeled and delineated by endoscopy experts. Algorithms for identifying, detecting, and classifying characteristic lesions of the GI tract have also been developed, showing high sensitivity, specificity, and accuracy with potential for commercialization in endoscopy. This progress presents an opportunity to establish a strong network and collaboration among multiple medical sites, as well as between medical and IT sectors. However, the algorithm development process faces many challenges, such as the lack of rich input database algorithms developed and tested mainly on still image sets with a limited variety of training images. Clear regulations on data security and sharing between collaborating institutions are crucial for the development of large datasets for AI development. Additionally, the government needs to establish a legal framework with detailed guidelines on regulations for developing and applying AI in clinical practice, ensuring adherence to medical ethics principles and protecting patient rights. Addressing these challenges could significantly influence the policymaking decisions regarding the integration of AI in the medical field in Vietnam. For researchers, the next approach is to design clinical trials appropriate to the national healthcare system to evaluate the effectiveness of the newly developed AI algorithms; evaluate the cost-effectiveness, and acceptance of patients, physicians, and endoscopists; and explore the potential of the application of AI in endoscopy training.

Supplemental Material

sj-docx-1-cmg-10.1177_26317745241306562 – Supplemental material for Application of artificial intelligence in gastrointestinal endoscopy in Vietnam: a narrative review

Supplemental material, sj-docx-1-cmg-10.1177_26317745241306562 for Application of artificial intelligence in gastrointestinal endoscopy in Vietnam: a narrative review by Hang Viet Dao, Binh Phuc Nguyen, Tung Thanh Nguyen, Hoa Ngoc Lam, Trang Thi Huyen Nguyen, Thao Thi Dang, Long Bao Hoang, Hung Quang Le and Long Van Dao in Therapeutic Advances in Gastrointestinal Endoscopy

Footnotes

Acknowledgements

None.

Declarations

ORCID iDs

Tung Thanh Nguyen

Hoa Ngoc Lam

Trang Thi Huyen Nguyen

Long Bao Hoang

Long Van Dao

Supplemental material

Supplemental material for this article is available online.

References

Peery

Crockett

Murphy

, et al. Burden and cost of gastrointestinal, liver, and pancreatic diseases in the United States: update 2021. Gastroenterology 2022; 162(2): 621–644.

Sung

Ferlay

Siegel

, et al. Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin 2021; 71(3): 209–249.

Menon

Trudgill

How commonly is upper gastrointestinal cancer missed at endoscopy? a meta-analysis. Endoscopy Int Open 2014; 2(2): E46–E50.

De Rosa

Pace

Rega

, et al. Genetics, diagnosis and management of colorectal cancer (Review). Oncol Rep 2015; 34(3): 1087–1096.

Spiceland

Lodhia

Endoscopy in inflammatory bowel disease: role in diagnosis, management, and treatment. World J Gastroenterol 2018; 24(35): 4014–4020.

Saftoiu

Hassan

Areia

, et al. Role of gastrointestinal endoscopy in the screening of digestive tract cancers in Europe: European Society of Gastrointestinal Endoscopy (ESGE) Position Statement. Endoscopy 2020; 52(4): 293–304.

Kuribayashi

Hosaka

Nakamura

, et al. The role of endoscopy in the management of gastroesophageal reflux disease. DEN Open 2022; 2(1): e86.

Pimenta-Melo

Monteiro-Soares

Libanio

, et al. Missing rate for gastric cancer during upper gastrointestinal endoscopy: a systematic review and meta-analysis. Eur J Gastroenterol Hepatol 2016; 28(9): 1041–1049.

Zhao

Wang

Pan

, et al. Magnitude, risk factors, and factors associated with adenoma miss rate of tandem colonoscopy: a systematic review and meta-analysis. Gastroenterology 2019; 156(6): 1661-74 e11.

10.

Okagawa

Abe

Yamada

, et al. Artificial intelligence in endoscopy. Dig Dis Sci 2022; 67(5): 1553–1572.

11.

Sharma

Hassan

Artificial intelligence and deep learning for upper gastrointestinal neoplasia. Gastroenterology 2022; 162(4): 1056–1066.

12.

Larsen

SLV

Mori

. Artificial intelligence in colonoscopy: a review on the current status. DEN Open 2022; 2(1): e109.

13.

Messmann

Bisschops

Antonelli

, et al. Expected value of artificial intelligence in gastrointestinal endoscopy: European Society of Gastrointestinal Endoscopy (ESGE) Position Statement. Endoscopy 2022; 54(12): 1211–1231.

14.

World Health Organization. Vietnam - Globocan 2022, https://gco.iarc.who.int/media/globocan/factsheets/populations/704-viet-nam-fact-sheet.pdf (2020, accessed 13 May 2024).

15.

Hang

Ngoc Hoa

Hai

VT.

Current status and the demand for building database of digestive endoscopy images and records in health facilities of Vietnam. J Pract Med 2020; 2(1126): 25–28.

16.

Min

Kwak

Cha

JM.

Overview of deep learning in gastrointestinal endoscopy. Gut Liver 2019; 13(4): 388–393.

17.

Willemink

Koszek

Hardell

, et al. Preparing medical imaging data for machine learning. Radiology 2020; 295(1): 4–15.

18.

Thambawita

Strumke

Hicks

, et al. Impact of image resolution on deep learning performance in endoscopy image classification: an experimental study using a large dataset of endoscopic images. Diagnostics (Basel) 2021; 11(12): 2183.

19.

Halevy

Norvig

Pereira

The unreasonable effectiveness of data. IEEE Intell Syst 2009; 24: 8–12.

20.

Yang

Kim

Lim

YJ.

Preparation of image databases for artificial intelligence algorithm development in gastrointestinal endoscopy. Clin Endosc 2022; 55(5): 594–604.

21.

Willemink

Koszek

Hardell

, et al. Preparing medical imaging data for machine learning. Radiology 2020; 295: 4–15.

22.

Diaz

Kushibar

Osuala

, et al. Data preparation for artificial intelligence in medical imaging: a comprehensive guide to open-access platforms and tools. Phys Med 2021; 83: 25–37.

23.

Park

Han

Methodologic guide for evaluating clinical performance and effect of artificial intelligence technology for medical diagnosis and prediction. Radiology 2018; 286(3): 800–809.

24.

Zhu

Gao

Liu

, et al. Public imaging datasets of gastrointestinal endoscopy for artificial intelligence: a review. J Imaging Inform Med 2023; 36(6): 2578–2601.

25.

Jha

Sharma

Dasu

, et al. GastroVision: a multi-class endoscopy image dataset for computer aided gastrointestinal disease detection. In: Workshop on machine learning for multimodal healthcare data, 2023. Springer.

26.

Pogorelov

Schmidt

Riegler

, et al. KVASIR: a multi-class image dataset for computer aided gastrointestinal disease detection. Taipei, Taiwan: ACM Press, 2017.

27.

Smedsrud

Thambawita

Hicks

, et al. Kvasir-Capsule, a video capsule endoscopy dataset. Sci Data 2021; 8(1): 142.

28.

Koulaouzidis

Iakovidis

Yung

, et al. KID project: an internet-based digital video atlas of capsule endoscopy for research purposes. Endosc Int Open 2017; 5(6): E477–E483.

29.

Leenhardt

Le Mouel

J-P

, et al. CAD-CAP: a 25,000-image database serving the development of artificial intelligence for capsule endoscopy. Endoscopy Int Open 2020; 8(03): E415–E420.

30.

Ali

Jha

Ghatwary

, et al. A multi-centre polyp detection and segmentation dataset for generalisability assessment. Sci Data 2023; 10(1): 75.

31.

Bernal

Tajkbaksh

Sanchez

, et al. Comparative validation of polyp detection methods in video colonoscopy: results from the MICCAI 2015 endoscopic vision challenge. IEEE Trans Med Imaging 2017; 36(6): 1231–1249.

32.

Chen

Cheng

, et al. LDPolypVideo benchmark: a large-scale colonoscopy video dataset of diverse polyps. Cham: Springer, 2021. pp. 387–396.

33.

Mesejo

Pizarro

Abergel

, et al. Computer-aided classification of gastrointestinal lesions in regular colonoscopy. IEEE Trans Med Imaging 2016; 35(9): 2051–2063.

34.

Tajbakhsh

Gurudu

Liang

Automated polyp detection in colonoscopy videos using shape and context information. IEEE Trans Med Imaging 2016; 35(2): 630–644.

35.

G-P

Xiao

Chou

Y-C

, et al. Video polyp segmentation: a deep learning perspective. Mach Intell Res 2022; 19(6): 531–549.

36.

Endoscopic Vision Challenge. ETIS-Larib Polyp DB 2014, https://polyp.grand-challenge.org/ETISLarib/

37.

Tokat

van Tilburg

Koch

, et al. Artificial intelligence in upper gastrointestinal endoscopy. Dig Dis 2022; 40(4): 395–408.

38.

Cogan

Tamil

MAPGI: accurate identification of anatomical landmarks and diseased tissue in gastrointestinal tract using deep learning. Comput Biol Med 2019; 111: 103351.

39.

Choi

Khan

Choi

, et al. Development of artificial intelligence system for quality control of photo documentation in esophagogastroduodenoscopy. Surg Endosc 2022; 36(1): 57–65.

40.

Takiyama

Ozawa

Ishihara

, et al. Automatic anatomical classification of esophagogastroduodenoscopy images using deep convolutional neural networks. Sci Rep 2018; 8(1):7497.

41.

Bano

Ahmad

, et al. Deep learning-based anatomical site classification for upper gastrointestinal endoscopy. Int J Comput Assist Radiol Surg 2020; 15(7): 1085–1094.

42.

Zhang

Zhou

, et al. Randomised controlled trial of WISENSE, a real-time quality improving system for monitoring blind spots during esophagogastroduodenoscopy. Gut 2019; 68(12): 2161–2169.

43.

Chen

, et al. Comparing blind spots of unsedated ultrafine, sedated, and unsedated conventional gastroscopy with and without artificial intelligence: a prospective, single-blind, 3-parallel-group, randomized, single-center trial. Gastrointest Endosc 2020; 91(2): 332-9 e3.

44.

Liu

, et al. Evaluation of the effects of an artificial intelligence system on endoscopy quality and preliminary testing of its performance in detecting early gastric cancer: a randomized controlled trial. Endoscopy 2021; 53(12): 1199–1207.

45.

Lui

TKL

Tsui

VWM

Leung

WK.

Accuracy of artificial intelligence-assisted detection of upper GI lesions: a systematic review and meta-analysis. Gastrointest Endosc 2020; 92(4): 821–30 e9.

46.

Shi

Wei

Wang

, et al. Diagnostic value of artificial intelligence-assisted endoscopy for chronic atrophic gastritis: a systematic review and meta-analysis. Front Med (Lausanne) 2023; 10: 1134980.

47.

Visaggi

de Bortoli

Barberio

, et al. Artificial Intelligence in the diagnosis of upper gastrointestinal diseases. J Clin Gastroenterol 2022; 56(1): 23–35.

48.

Shang

Sharma

, et al. Effect of a deep learning-based system on the miss rate of gastric neoplasms during upper gastrointestinal endoscopy: a single-centre, tandem, randomised controlled trial. Lancet Gastroenterol Hepatol 2021; 6(9): 700–708.

49.

Yuan

Liu

Lin

, et al. Effect of an artificial intelligence-assisted system on endoscopic diagnosis of superficial oesophageal squamous cell carcinoma and precancerous lesions: a multicentre, tandem, double-blind, randomised controlled trial. Lancet Gastroenterol Hepatol 2024; 9(1): 34–44.

50.

Lee

, et al. Clean mucosal area detection of gastroenterologists versus artificial intelligence in small bowel capsule endoscopy. Medicine (Baltimore) 2023; 102(6): e32883.

51.

Lee

Calderwood

Karnes

, et al. Artificial intelligence for the assessment of bowel preparation. Gastrointest Endosc 2022; 95(3): 512-8 e1.

52.

Mascarenhas Saraiva

Afonso

Ribeiro

, et al. AI-driven colon cleansing evaluation in capsule endoscopy: a deep learning approach. Diagnostics (Basel) 2023; 13(23): 3394.

53.

Barua

Vinsard

Jodal

, et al. Artificial intelligence for polyp detection during colonoscopy: a systematic review and meta-analysis. Endoscopy 2021; 53(3): 277–284.

54.

Deliwala

Hamid

Barbarawi

, et al. Artificial intelligence (AI) real-time detection vs. routine colonoscopy for colorectal neoplasia: a meta-analysis and trial sequential analysis. Int J Colorectal Dis 2021; 36(11): 2291–2303.

55.

Hassan

Spadaccini

Iannone

, et al. Performance of artificial intelligence in colonoscopy for adenoma and polyp detection: a systematic review and meta-analysis. Gastrointest Endosc 2021; 93(1): 77–85 e6.

56.

Glissen Brown

Mansour

Wang

, et al. Deep learning computer-aided polyp detection reduces adenoma miss rate: a united states multi-center randomized tandem colonoscopy study (CADeT-CS Trial). Clin Gastroenterol Hepatol 2022; 20(7): 1499–507 e4.

57.

Wallace

Sharma

Bhandari

, et al. Impact of artificial intelligence on miss rate of colorectal neoplasia. Gastroenterology 2022; 163(1): 295–304 e5.

58.

Yao

Zhang

Liu

, et al. Effect of an artificial intelligence-based quality improvement system on efficacy of a computer-aided detection system in colonoscopy: a four-group parallel study. Endoscopy 2022; 54(8): 757–768.

59.

Rondonotti

Hassan

Tamanini

, et al. Artificial intelligence-assisted optical diagnosis for the resect-and-discard strategy in clinical practice: the Artificial intelligence BLI Characterization (ABC) study. Endoscopy 2023; 55(1): 14–22.

60.

Houwen

Hazewinkel

Giotis

, et al. Computer-aided diagnosis for optical diagnosis of diminutive colorectal polyps including sessile serrated lesions: a real-time comparison with screening endoscopists. Endoscopy 2023; 55(8): 756–765.

61.

Hann

Troya

Fitting

Current status and limitations of artificial intelligence in colonoscopy. United European Gastroenterol J 2021; 9(5): 527–533.

62.

Milluzzo

Cesaro

Grazioli

, et al. Artificial intelligence in lower gastrointestinal endoscopy: the current status and future perspective. Clin Endosc 2021; 54(3): 329–339.

63.

Wang

Zhong

, et al. Artificial intelligence-assisted detection and classification of colorectal polyps under colonoscopy: a systematic review and meta-analysis. Ann Transl Med 2021; 9(22): 1662.

64.

Maeda

Kudo

Mori

, et al. Fully automated diagnostic system with artificial intelligence using endocytoscopy to identify the presence of histologic inflammation associated with ulcerative colitis (with video). Gastrointest Endosc 2019; 89(2): 408–415.

65.

Hoang

Nguyen

, et al. Enhancing endoscopic image classification with symptom localization and data augmentation. In: MM ‘19: Proceedings of the 27th ACM international conference on multimedia, 2019, pp. 2578–2582.

66.

Viet Hang

Quang Hung

Phuc Binh

, et al. Preliminary of artificial intelligence application in colonoscopic polyps detection in Vietnam. J Med Res 2020; 130(6): 101–110.

67.

Ngoc Lan

Sy An

Viet Hang

, et al. BKAI-IGH NeoPolyp-Small. 2020. https://github.com/GivralNguyen/BKAI-IGH-Neopolyp-Segmentation (2020, accessed 2 June 2024).

68.

Manh

Hang

Long

, et al. EndoUNet: a unified model for anatomical site classification, lesion categorization and segmentation for upper gastrointestinal endoscopy. In: 2022 14th International conference on knowledge and systems engineering (KSE), 2022.

69.

Manh

Dao

, et al. Cleaning assessment in endoscopic esophageal images using U-Net and a classification model. In: 2020 international conference on multimedia analysis and pattern recognition (MAPR), 2020, 8-9 Oct 2020.

70.

Manh Hung

Phuc Binh

Hai

, et al. Preliminary results of applying enhanced-image generating software in upper gastrointestinal endoscopy. Vietnam Med J 2022; 519(1): 39–43.

71.

Hai

Cong

, et al. A robust and high-performance neural network for classifying landmarks in upper gastrointestinal endoscopy images. In: Proceedings of the 11th International Symposium on Information and Communication Technology (SoICT), Hanoi, Vietnam, 2022. pp. 254–261.

72.

Tran

T-H

Nguyen

P-T

Tran

D-H

, et al. Classification of anatomical landmarks from upper gastrointestinal endoscopic images. In: 2021 8th NAFOSTED conference on information and computer science (NICS), 2021.

73.

Marques

Bispo

Pimentel-Nunes

, et al. Image documentation in gastrointestinal endoscopy: review of recommendations. GE Port J Gastroenterol 2017; 24(6): 269–274.

74.

Nguyen

PTT

Dao

Q-T

, et al. Automatic classification of upper gastrointestinal tract diseases from endoscopic images. In: 2022 11th International conference on control, automation and information sciences (ICCAIS), Hanoi, Vietnam. 2022, pp. 442–447.

75.

Tran

T-H

Tran

, et al. DCS-UNet: dual-path framework for segmentation of reflux esophagitis lesions from endoscopic images with U-Net-based segmentation and color/texture analysis. Vietnam J Comput Sci 2023; 10(02): 217–242.

76.

D-H

Nguyen

L-T

Nguyen

V-T

, et al. Boundary delineation of reflux esophagitis lesions from endoscopic images using color and texture, 2021, pp. 1–6.

77.

Tri Thuc

Ngoc Hoa

Thi Ly

, et al. Application of artificial intelligence in detecting reflux esophagitis on endoscopy images. J Med Res 2023; 170(9): 144–151.

78.

Nguyen

Tran

Dao

, et al. Improving gastroesophageal reflux diseases classification diagnosis from endoscopic images using StyleGAN2-ADA. In: International conference on artificial intelligence and big data in digital era, 2021, Springer.

79.

Trinh

TTA

, et al. Improving loss function for a deep neural network for lesion segmentation. In: Proceedings of the 12th international symposium on information and communication technology, Ho Chi Minh, Vietnam: Association for Computing Machinery, 2023, pp. 592–598.

80.

Viet Hang

Thanh Tung

Ngoc Hoa

, et al. Evaluating the accuracy of EfficientNet in colon polyp detection and delineation. Vietnam Med J 2022; 509(1): 319–322.

81.

Lan

Hang

, et al. NeoUNet: towards accurate colon polyp segmentation and neoplasm detection, arXiv:2107.05023, 2021.

82.

Lan

Hang

, et al. BlazeNeo: blazing fast polyp segmentation and neoplasm detection. IEEE Access 2022; 10: 43669–43684.

83.

Nguyen Mau

Trinh

Bui

, et al. Multi kernel positional embedding ConvNeXt for Polyp segmentation. In: 2022 RIVF International conference on computing and communication technologies (RIVF), 2022.

84.

Nguyen-Mau

T-H

Hoang

T-V

Nguyen

H-D

, et al. ConvTransNet: merging convolution with transformer to enhance polyp segmentation. In: Proceedings of the 12th international symposium on information and communication technology, 2023.

85.

Nam

Nguyen

Thuy

, et al. A coarse-to-fine unsupervised domain adaptation method for cross-mode polyp segmentation. In: 2022 14th international conference on knowledge and systems engineering (KSE), 2022, 19–21 Oct 2022.

86.

Hung

Duc

Chien

, et al. AG-ResUNet++: an improved encoder-decoder based method for polyp segmentation in colonoscopy images. In: 2021 RIVF international conference on computing and communication technologies (RIVF), 19–21 Aug 2021.

87.

Sang

Chung

Lan

, et al. AG-CUResNeSt: a novel method for colon polyp segmentation, arXiv:2105.00402, 2021.

88.

Van

Doan

Nguyen

, et al. Online pseudo labeling for polyp segmentation with momentum networks. In: 2022 14th international conference on knowledge and systems engineering (KSE), 2022, 19–21 Oct 2022.

89.

Pham

Dinh

Doan

, et al. Improve polyp semi-supervised segmentation with prioritizing the reliability of unlabeled images. In: Proceedings of the 11th international conference on software and information engineering, Beijing, China: Association for Computing Machinery, 2023, pp. 35–40.

90.

Thanh Duc

Oanh

Nguyen

, et al. ColonFormer: an efficient transformer based method for colon polyp segmentation, arXiv:2205.08473, 2022.

91.

Trinh

Q-H

Bui

N-T

Nguyen-Mau

T-H

, et al. M^2UNet: MetaFormer multi-scale upsampling network for polyp segmentation, arXiv:2306.08600, 2023.

92.

Lui

TKL

Hang

Tsao

SKK

, et al. Computer-assisted detection versus conventional colonoscopy for proximal colonic lesions: a multicenter, randomized, tandem-colonoscopy study. Gastrointest Endosc 2023; 97(2): 325–34 e1.

93.

Government. No: 13/2023/NĐ-CP – Decree on Personal Data Protection.

94.

Ministry of Health. Circular No. 04/TT-BYT Establishment, Functions, Duties, and Powers of the Ethics Council in Biomedical Research, 2020.

95.

Ministry of Health. Circular No. 46/2018/TT-BYT Stipulates Regulations on Electronic Medical Records, 2018.

96.

Bahri

Rägo

CIOMS guide to vaccine safety communication–executive summary. Vaccine 2019; 37(3): 401–408.

97.

Ashcroft

. The declaration of Helsinki. In: The Oxford textbook of clinical research ethics. New York: The Oxford Textbook of Clinical Research Ethics, 2008, pp. 141–148.

98.

Geiger

Cope

, et al. “Garbage In, Garbage Out” Revisited: what do machine learning application papers report about human-labeled training data?, arXiv:2107.02278, 2021.

99.

van der Sommen

de Groof

Struyvenberg

, et al. Machine learning in GI endoscopy: practical guidance in how to interpret a novel field. Gut 2020; 69(11): 2035–2045.

100.

de Lange

Halvorsen

Riegler

. Methodology to develop machine learning algorithms to improve performance in gastrointestinal endoscopy. World J Gastroenterol 2018; 24(45): 5057–5062.

101.

Lai

Calderwood

Doros

, et al. The Boston bowel preparation scale: a valid and reliable instrument for colonoscopy-oriented research. Gastrointest Endosc 2009; 69(3 Pt 2): 620–625.

102.

Basford

Brown

Gadeke

, et al. A randomized controlled trial of pre-procedure simethicone and N-acetylcysteine to improve mucosal visibility during gastroscopy - NICEVIS. Endosc Int Open 2016; 4(11): E1197–E202.

103.

Dao Van

Tran Quoc

. Application of big data and artificial intelligence in hospital management Vietnam. J Commun Med 2024; 65(CD5): 282–287.

104.

Vuong

, et al. Artificial intelligence vs. natural stupidity: evaluating AI readiness for the Vietnamese medical information system. J Clin Med 2019; 8(2): 168.

105.

FujiFilm Vietnam. CAD EYE endoscopy support program, https://www.fujifilm.com/vn/en/healthcare/endoscopy/endoscopy-processors/cadeye (2024,accessed 11 October 2024).

106.

Medtronic. GI genius intelligent endoscopy module, https://www.medtronic.com/covidien/en-us/products/gastrointestinal-artificial-intelligence/gi-genius-intelligent-endoscopy.html (accessed 11 October 2024).

107.

Prime Minister. National strategy for research, development, and application of artificial intelligence until 2030. https://thuvienphapluat.vn/van-ban/Cong-nghe-thong-tin/Quyet-dinh-127-QD-TTg-2021-Chien-luoc-nghien-cuu-phat-trien-va-ung-dung-Tri-tue-nhan-tao-den-2030-463680 (accessed 11 October 2024).

108.

Kolbinger

Veldhuizen

Zhu

, et al. Reporting guidelines in medical artificial intelligence: a systematic review and meta-analysis. Commun Med (Lond) 2024; 4(1): 71.

109.

de Hond

AAH

Leeuwenberg

Hooft

, et al. Guidelines and quality criteria for artificial intelligence-based prediction models in healthcare: a scoping review. NPJ Digit Med 2022; 5(1): 2.

110.

Ministry of Health. Decision 5316 QD-BYT digital transformation program in healthcare until 2025, https://thuvienphapluat.vn/van-ban/The-thao-Y-te/Quyet-dinh-5316-QD-BYT-2020-phe-duyet-chuong-trinh-chuyen-doi-so-y-te-den-2025-460152.aspx (2020, accessed 1 October 2024).

111.

Mori

East

Hassan

, et al. Benefits and challenges in implementation of artificial intelligence in colonoscopy: world endoscopy organization position statement. Dig Endosc 2023; 35(4): 422–429.

112.

Hang

Trang

NTH

Uyen

VTT

, et al. Experiences and acceptance of endocsopists regarding the implementation of an AI algorithim intergrated system in gastrointestinal endoscopy in Vietnam. Vietnam Med J 2024; 542(3): 272–276.

113.

Ahmad

Stoyanov

Lovat

LB.

Barriers and pitfalls for artificial intelligence in gastroenterology: ethical and regulatory issues. Tech Innovations Gastrointestinal Endoscopy 2020; 22(2): 80–84.

114.

Lee

QUEK

SXZ

KOO

, et al. Survey on the perceptions of Asian endoscopists to artificial intelligence. Preprint. Research Square.2024.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.02 MB