Lesions In Capsule Endoscopy Research Articles

A Deep Learning Application of Capsule Endoscopic Gastric Structure Recognition Based on a Transformer Model.

Gastric structure recognition systems have become increasingly necessary for the accurate diagnosis of gastric lesions in capsule endoscopy. Deep learning, especially using transformer models, has shown great potential in the recognition of gastrointestinal (GI) images according to self-attention. This study aims to establish an identification model of capsule endoscopy gastric structures to improve the clinical applicability of deep learning to endoscopic image recognition. A total of 3343 wireless capsule endoscopy videos collected at Nanfang Hospital between 2011 and 2021 were used for unsupervised pretraining, while 2433 were for training and 118 were for validation. Fifteen upper GI structures were selected for quantifying the examination quality. We also conducted a comparison of the classification performance between the artificial intelligence model and endoscopists by the accuracy, sensitivity, specificity, and positive and negative predictive values. The transformer-based AI model reached a relatively high level of diagnostic accuracy in gastric structure recognition. Regarding the performance of identifying 15 upper GI structures, the AI model achieved a macroaverage accuracy of 99.6% (95% CI: 99.5-99.7), a macroaverage sensitivity of 96.4% (95% CI: 95.3-97.5), and a macroaverage specificity of 99.8% (95% CI: 99.7-99.9) and achieved a high level of interobserver agreement with endoscopists. The transformer-based AI model can accurately evaluate the gastric structure information of capsule endoscopy with the same performance as that of endoscopists, which will provide tremendous help for doctors in making a diagnosis from a large number of images and improve the efficiency of examination.

Deep Learning and Minimally Invasive Endoscopy: Panendoscopic Detection of Pleomorphic Lesions

Introduction: Capsule endoscopy (CE) is a minimally invasive exam suitable of panendoscopic evaluation of the gastrointestinal (GI) tract. Nevertheless, CE is time-consuming with suboptimal diagnostic yield in the upper GI tract. Convolutional neural networks (CNN) are human brain architecture-based models suitable for image analysis. However, there is no study about their role in capsule panendoscopy. Methods: Our group developed an artificial intelligence (AI) model for panendoscopic automatic detection of pleomorphic lesions (namely vascular lesions, protuberant lesions, hematic residues, ulcers, and erosions). 355,110 images (6,977 esophageal, 12,918 gastric, 258,443 small bowel, 76,772 colonic) from eight different CE and colon CE (CCE) devices were divided into a training and validation dataset in a patient split design. The model classification was compared to three CE experts’ classification. The model’s performance was evaluated by its sensitivity, specificity, accuracy, positive predictive value, negative predictive value, and area under the precision-recall curve. Results: The binary esophagus CNN had a diagnostic accuracy for pleomorphic lesions of 83.6%. The binary gastric CNN identified pleomorphic lesions with a 96.6% accuracy. The undenary small bowel CNN distinguished pleomorphic lesions with different hemorrhagic potentials with 97.6% accuracy. The trinary colonic CNN (detection and differentiation of normal mucosa, pleomorphic lesions, and hematic residues) had 94.9% global accuracy. Discussion/Conclusion: We developed the first AI model for panendoscopic automatic detection of pleomorphic lesions in both CE and CCE from multiple brands, solving a critical interoperability technological challenge. Deep learning-based tools may change the landscape of minimally invasive capsule panendoscopy.

Deep learning and capsule endoscopy: Automatic multi-brand and multi-device panendoscopic detection of vascular lesions.

Background and study aims Capsule endoscopy (CE) is commonly used as the initial exam for suspected mid-gastrointestinal bleeding after normal upper and lower endoscopy. Although the assessment of the small bowel is the primary focus of CE, detecting upstream or downstream vascular lesions may also be clinically significant. This study aimed to develop and test a convolutional neural network (CNN)-based model for panendoscopic automatic detection of vascular lesions during CE. Patients and methods A multicentric AI model development study was based on 1022 CE exams. Our group used 34655 frames from seven types of CE devices, of which 11091 were considered to have vascular lesions (angiectasia or varices) after triple validation. We divided data into a training and a validation set, and the latter was used to evaluate the model's performance. At the time of division, all frames from a given patient were assigned to the same dataset. Our primary outcome measures were sensitivity, specificity, accuracy, positive predictive value (PPV), negative predictive value (NPV), and an area under the precision-recall curve (AUC-PR). Results Sensitivity and specificity were 86.4% and 98.3%, respectively. PPV was 95.2%, while the NPV was 95.0%. Overall accuracy was 95.0%. The AUC-PR value was 0.96. The CNN processed 115 frames per second. Conclusions This is the first proof-of-concept artificial intelligence deep learning model developed for pan-endoscopic automatic detection of vascular lesions during CE. The diagnostic performance of this CNN in multi-brand devices addresses an essential issue of technological interoperability, allowing it to be replicated in multiple technological settings.

Deep Learning and Minimally Invasive Endoscopy: Automatic Classification of Pleomorphic Gastric Lesions in Capsule Endoscopy.

Capsule endoscopy (CE) is a minimally invasive examination for evaluating the gastrointestinal tract. However, its diagnostic yield for detecting gastric lesions is suboptimal. Convolutional neural networks (CNNs) are artificial intelligence models with great performance for image analysis. Nonetheless, their role in gastric evaluation by wireless CE (WCE) has not been explored. Our group developed a CNN-based algorithm for the automatic classification of pleomorphic gastric lesions, including vascular lesions (angiectasia, varices, and red spots), protruding lesions, ulcers, and erosions. A total of 12,918 gastric images from 3 different CE devices (PillCam Crohn's; PillCam SB3; OMOM HD CE system) were used from the construction of the CNN: 1,407 from protruding lesions; 994 from ulcers and erosions; 822 from vascular lesions; and 2,851 from hematic residues and the remaining images from normal mucosa. The images were divided into a training (split for three-fold cross-validation) and validation data set. The model's output was compared with a consensus classification by 2 WCE-experienced gastroenterologists. The network's performance was evaluated by its sensitivity, specificity, accuracy, positive predictive value and negative predictive value, and area under the precision-recall curve. The trained CNN had a 97.4% sensitivity; 95.9% specificity; and positive predictive value and negative predictive value of 95.0% and 97.8%, respectively, for gastric lesions, with 96.6% overall accuracy. The CNN had an image processing time of 115 images per second. Our group developed, for the first time, a CNN capable of automatically detecting pleomorphic gastric lesions in both small bowel and colon CE devices.

Deep Learning for Automatic Identification and Characterization of the Bleeding Potential of Enteric Protruding Lesions in Capsule Endoscopy.

Capsule endoscopy (CE) revolutionized the study of the small intestine, overcoming the limitations of conventional endoscopy. Nevertheless, reviewing CE images is time-consuming. Convolutional Neural Networks (CNNs) are an artificial intelligence architecture with high performance levels for image analysis. Protruding lesions of the small intestine exhibit enormous morphologic diversity in CE images. We aimed to develop a CNN-based algorithm for automatic detection of varied small-bowel protruding lesions. A CNN was developed using a pool of CE images containing protruding lesions or normal mucosa/other findings. A total of 2565 patients were included. These images were inserted into a CNN model with transfer learning. We evaluated the performance of the network by calculating its sensitivity, specificity, accuracy, positive predictive value, and negative predictive value. A CNN was developed based on a total of 21,320 CE images. Training and validation data sets comprising 80% and 20% of the total pool of images, respectively, were constructed for development and testing of the network. The algorithm automatically detected small-bowel protruding lesions with an accuracy of 97.1%. Our CNN had a sensitivity, specificity, positive, and negative predictive values of 95.9%, 97.1%, 83.0%, and 95.7%, respectively. The CNN operated at a rate of approximately 355 frames per second. We developed an accurate CNN for automatic detection of enteric protruding lesions with a wide range of morphologies. The development of these tools may enhance the diagnostic efficiency of CE.

Nomenclature and semantic descriptions of ulcerative and inflammatory lesions seen in Crohn's disease in small bowel capsule endoscopy: An international Delphi consensus statement.

In the medical literature, the nomenclature and descriptions (ND) of small bowel (SB) ulcerative and inflammatory (U-I) lesions in capsule endoscopy (CE) are scarce and inconsistent. Inter-observer variability in interpreting these findings remains a major limitation in the assessment of the severity of mucosal lesions, which can impact negatively on clinical care, training and research on SB-CE. Focusing on SB-CE in Crohn's disease (CD), our aim is to establish a consensus on the ND of U-I lesions. An international panel of experienced SB-CE readers was formed during the 2016 United European Gastroenterology Week meeting. A core group of five CE and inflammatory bowel disease (IBD) experts established an Internet-based, three-round Delphi consensus but did not participate in the voting process. The core group built illustrated questionnaires, including SB-CE still frames of U-I lesions from patients with documented CD. Twenty-seven other experts were asked to rate and comment on the different proposals for the ND of the most frequent SB U-I lesions. For each round, we used a 6-point rating scale (varying from 'strongly disagree' to 'strongly agree'). The consensus was reached when at least 80 % of the voting members scored the statement within the 'agree' or 'strongly agree' categories. A 100% participation rate was obtained for all the rounds. Consensual ND were reached for the following seven U-I lesions: aphthoid erosion, deep ulceration, superficial ulceration, stenosis, edema, hyperemia and denudation. Considering the most frequent SB U-I lesions seen in CE in CD, a consensual ND was reached by the international group of experts. These descriptions and names are useful not only for daily practice and medical education, but also for medical research.

Nomenclature and semantic description of vascular lesions in small bowel capsule endoscopy: an international Delphi consensus statement.

Background and study aims Nomenclature and descriptions of small bowel (SB) vascular lesions in capsule endoscopy (CE) are scarce in the medical literature. They are mostly based on the reader's opinion and thus differ between experts, with a potential negative impact on clinical care, teaching and research regarding SBCE. Our aim was to better define a nomenclature and to give a description of the most frequent vascular lesions in SBCE. Methods A panel of 18 European expert SBCE readers was formed during the UEGW 2016 meeting. Three experts constructed an Internet-based four-round Delphi consensus, but did not participate in the voting process. They built questionnaires that included various still frames of vascular lesions obtained with a third-generation SBCE system. The 15 remaining participants were asked to rate different proposals and description of the most common SB vascular lesions. A 6-point rating scale (varying from ‘strongly disagree’ to ‘strongly agree’) was used successive rounds. The consensus was reached when at least 80 % voting members scored the statement within the ‘agree’ or ‘strongly agree’. Results Consensual terms and descriptions were reached for angiectasia/angiodysplasia, erythematous patch, red spot/dot, and phlebectasia. A consensual description was reached for more subtle vascular lesions tentatively named “diminutive angiectasia” but no consensus was reached for this term. Conclusion An international group has reached a consensus on the nomenclature and descriptions of the most frequent and relevant SB vascular lesions in CE. These terms and descriptions are useful in daily practice, for teaching and for medical research purposes.

Development and validation of a scoring index to predict the presence of lesions in capsule endoscopy in patients with suspected Crohn's disease of the small bowel: a Spanish multicenter study.

Capsule endoscopy (CE) is the first-line investigation in cases of suspected Crohn's disease (CD) of the small bowel, but the factors associated with a higher diagnostic yield remain unclear. Our aim is to develop and validate a scoring index to assess the risk of the patients in this setting on the basis of biomarkers. Data on fecal calprotectin, C-reactive protein, and other biomarkers from a population of 124 patients with suspected CD of the small bowel studied by CE and included in a PhD study were used to build a scoring index. This was first used on this population (internal validation process) and after that on a different set of patients from a multicenter study (external validation process). An index was designed in which every biomarker is assigned a score. Three risk groups have been established (low, intermediate, and high). In the internal validation analysis (124 individuals), patients had a 10, 46.5, and 81% probability of showing inflammatory lesions in CE in the low-risk, intermediate-risk, and high-risk groups, respectively. In the external validation analysis, including 410 patients from 12 Spanish hospitals, this probability was 15.8, 49.7, and 80.6% for the low-risk, intermediate-risk, and high-risk groups, respectively. Results from the internal validation process show that the scoring index is coherent, and results from the external validation process confirm its reliability. This index can be a useful tool for selecting patients before CE studies in cases of suspected CD of the small bowel.

PTH-006 Computer-Aided Lesion Measurement in Capsule Endoscopy Images

Introduction A novel image segmentation algorithm is applied & assessed for lesion size measurements in capsule endoscopy (CE). It is based solely on colour features of CE lesions. Methods The lesion images are available online from KID database (Dataset 1). 1 Size reference standards were made by manual annotation. The proposed algorithm uses the CIE-Lab colour representation instead of the standard RGB used in CE. 1 The components of this space represent lightness (L), the quantity of red (a > 0) or the quantity of green (-a >0), the quantity of yellow (b > 0) or the quantity of blue (-b >0) of a pixel. Initially, the user specifies a point on the lesion of interest (with a single click). Simple Linear Iterative Clustering (SLIC) is applied to group the pixels of similar colour into contiguous regions, called superpixels. 2 Subsequently the superpixels are clustered using the k-means approach into three clusters using information from component a. Thereafter, each superpixel neighbouring to the superpixel that contains the user-specified point is represented by a colour vector (a, b). The proposed algorithm estimates a) its Euclidean distance d1 from the respective vector of the selected superpixel, b) its Euclidean distance d2 from the mean of the respective vectors obtained from all superpixels that do not belong to the cluster of the selected superpixel. Then, the neigerpixel is considered to belong to the abnormal region of interest if d1 Results Seven types of GI lesions were used for evaluation. The measurement accuracy was assessed by comparing the area of the lesion (as identified by the method) with the reference standard area of each lesion. Average accuracies obtained for the measurement of angioectasias, aphthae, chylous cysts, lymphangiectasias, polypoid lesions, stenoses, and ulcers are 98.6%, 92.8%, 94.3%, 99.1%, 80.0%, 82.9%, and 94.8%, respectively. Comparatively, using the well-known colour space proposed by Ohta (I1I2I3) for image segmentation instead of CIE-Lab 2 the results are lower; 97.1%, 92.1%, 92.6%, 97.8%, 75.2%, 78.8%, and 91%, respectively. Conclusion A novel algorithm was proposed & evaluated for accurate computer-aided size measurement of lesions in CE. The overall accuracy on a public dataset was ~92%. This algorithm can be incorporated as a novel measurement tool in contemporary CE reading software. References 1 Iakovidis DK, Koulaouzidis A. Software for enhanced video capsule endoscopy: challenges for essential progress. Nat Rev Gastroenterol Hepatol . 2015; 12 :172–8. 2 Achanta R, et al . Slic superpixels compared to state-of-the-art superpixel methods. IEEE Transactions on Pattern Analysis and Machine Intelligence . 2012; 34 :2274–8. Disclosure of Interest None Declared

PTU-001 Chromoendoscopy for Angiectasias in Capsule Endoscopy; Blue or Just White?

Introduction Blue mode (BM) is one of the features of proprietary software (RAPID®; Medtronic Ltd) for capsule endoscopy (CE) review. BM is a colour coefficient shift of light in the short wavelength range (490–430 nm) superimposed onto a white light (WL) image.1 Recently, studies have shown that BM improves visualisation of vascular and erythematous non-vascular CE lesions.1,2 We aimed to objectively evaluate the validity of BM in CE in assessing the surface annotations of angiectasias as compared to WL. Methods A set of 100 anonymised images of angiectasias was used, with the lesions captured both in WL and BM in the same pose. The entire dataset is available in our online database, KID (http://is-innovation.eu/kid). Three reviewers (2 experts and one novice in CE review) graphically annotated the lesions using Ratsnake annotation tool (http://is-innovation.eu/ratsnake). The images were reviewed in WL and BM twice to estimate the inter- and intra-observer variability (at least 7 days apart). The Jaccard index (JI) was used to assess the similarity (agreement) of the annotations performed by the reviewers. Results Under WL, the average inter-observer agreement ranged between 65±15% (novice vs. expert reviewer) and 67±13% (between experts), while the intra-observer agreement, ranged between 69±17% and 71±13%. Under BM, the average inter-observer agreement ranged between 56±19% (novice vs. expert reviewer) and 78±18% (between experts). The average intra-observer agreement in BM ranged between 69±20% and 73±8. Conclusion BM CE image review does not improve significantly the surface annotations of angiectasias -as compared to WL- for expert or novice reviewers.3,4 References 1 Abdelaal UM, et al. Blue mode imaging may improve the detection and visualisation of small-bowel lesions: A capsule endoscopy study. Saudi J Gastroentero 2015;21:418–22. 2 Krystallis C, et al. Chromoendoscopy in small bowel capsule endoscopy: Blue mode or Fuji Intelligent Colour Enhancement? Dig Liver Dis 2011;43:953–7. 3 Koulaouzidis A, et al. Blue mode does not offer any benefit over white light when calculating Lewis score in small-bowel capsule endoscopy. World J Gastrointest Endosc 2012;4:33–7. 4 Koulaouzidis A, et al. QuickView in small-bowelcapsule endoscopy is useful in certain clinical settings, but QuickView with BlueMode is of no additional benefit. Eur J Gastroenterol Hepatol 2012;24:1099–104. Disclosure of Interest None Declared

Timing of capsule endoscopy influences the diagnosis and outcome in obscure-overt gastrointestinal bleeding.

Capsule endoscopy (CE) has high positive (95%) and negative predictive values (83- 100%) for the bleeding source in patients with obscure gastrointestinal bleeding (OGIB) but the diagnostic yield is influenced by the timing of the examination and the nature of the bleeding. We investigated the role of early CE in the diagnosis of patients with overt OGIB. Patients were divided into 4 groups according to the interval between bleeding and CE. The diagnosis rate and clinical outcomes including type of intervention and recurrent bleeding were compared. The association between the timing of CE and small bowel lesions detected by CE was also analysed. Ninety patients with overt OGIB underwent CE. CE revealed the bleeding source in 46 patients (51%). The diagnosis rate was 73%, 48%, 50% and 35% in 1st, 2nd, 3rd and 4th quartiles (p=0.08). The duration between bleeding and CE was shorter for patients with angioectasia than for those with other abnormalities. The proportion of surgeries and endoscopic interventions was significantly higher in 1st and 2nd quartiles than 3rd and 4th quartiles (p=0.048). Earlier timing of CE achieved a higher diagnostic yield for patients with overt OGIB and consequently resulted in a higher intervention rate.

A feasibility trial of computer-aided diagnosis for enteric lesions in capsule endoscopy

To investigate and evaluate the feasibility of the computer-aided screening diagnosis for enteric lesions in the capsule endoscopy (CE). After developing a series of algorithms for the screening diagnosis of the enteric lesions in CE based on their characteristic colors and contours, the normal and abnormal images obtained from 289 patients were respectively scanned and diagnosed by the CE readers and by the computer-aided screening for the enteric lesions with the image-processed software (IPS). The enteric lesions shown by the images included esoenteritis, mucosal ulcer and erosion, bleeding, space-occupying lesions, angioectasia, diverticula, parasites, etc. The images for the lesions or the suspected lesions confirmed by the CE readers and the computers were collected, and the effectiveness rate of the screening and the number of the scanned images were evaluated, respectively. Compared with the diagnostic results obtained by the CE readers, the total effectiveness rate (sensitivity) in the screening of the commonly-encountered enteric lesions by IPS varied from 42.9% to 91.2%, with a median of 74.2%, though the specificity and the accuracy rates were still low, and the images for the rarely-encountered lesions were difficult to differentiate from the normal images. However, the number of the images screened by IPS was 5000 on average, and only 10%-15% of the original images were left behind. As a result, a large number of normal images were excluded, and the reading time decreased from 5 h to 1 h on average. Though the total accuracy and specificity rates by the computer-aided screening for the enteric lesions with IPS are much lower than those by the CE readers, the computer-aided screening diagnosis can exclude a large number of the normal images and confine the enteric lesions to 5000 images on average, which can reduce the workload of the readers in the scanning of the images. This computer-aided screening technique can make a correct diagnosis as efficiently as possible in most of the patients.

Efficacy of Transit Time Method for Localization of Small Bowel Lesions in Capsule Endoscopy

Backgrounds and Aims: Recently as minimal invasive surgery is performed widely, there is increasing necessity of preoperative localization by capsule endoscopy (CE). However, the localization of small bowel lesions by CE is known to be difficult. And few systematic data of localization methods by CE is reported. Thus we evaluated efficacy of our own transit time method for localization of small bowel lesions by CE in surgically confirmed cases. Patients and Methods: Between October 2002 and January 2006, 18 patients were examined by capsule endoscopy for small bowel pathology, and confirmed by laparoscopic or laparoscopy assisted surgery.

Influence of Experience, Scanning Speed, and the Use of a Jog-Shuttle Device On the Detection Rate in Capsule Endoscopy

Background: The clinical value of capsule endoscopy is clear but many aspects of its practical use remain to be defined. Aim: To determine examiner-dependent factors on the detection rate of lesions in capsule endoscopy. Methods: Two training videos (A and B) of capsule endoscopy lasting 330 seconds each were reviewed by 72 experienced gastroenterologists [advanced capsule endocopists (ADV): 14, intermediate (INT): 7; beginners (BEG): 50]. One video was read using the conventional hardware while the second video was read using an additionally connected jog-shuttle device (JSD) that allowed for immediate adjustment of scanning speed.