Computer-aided diagnosis in real-time endoscopy for all stages of gastric carcinogenesis: Development and validation study.

OBJECTIVE: Using endoscopic images, we have previously developed computer-aided diagnosis models to predict the histopathology of gastric neoplasms. However, no model that categorizes every stage of gastric carcinogenesis has been published. In this study, a deep-learning-based diagnosis model was developed and validated to automatically classify all stages of gastric carcinogenesis, including atrophy and intestinal metaplasia, in endoscopy images. DESIGN: A total of 18,701 endoscopic images were collected retrospectively and randomly divided into train, validation, and internal-test datasets in an 8:1:1 ratio. The primary outcome was lesion-classification accuracy in six categories: normal/atrophy/intestinal metaplasia/dysplasia/early /advanced gastric cancer. External-validation of performance in the established model used 1427 novel images from other institutions that were not used in training, validation, or internal-tests. RESULTS: The internal-test lesion-classification accuracy was 91.2% (95% confidence interval: 89.9%-92.5%). For performance validation, the established model achieved an accuracy of 82.3% (80.3%-84.3%). The external-test per-class receiver operating characteristic in the diagnosis of atrophy and intestinal metaplasia was 93.4 ± 0% and 91.3 ± 0%, respectively. CONCLUSIONS: The established model demonstrated high performance in the diagnosis of preneoplastic lesions (atrophy and intestinal metaplasia) as well as gastric neoplasms.

Endoscopic Kyoto and Kimura-Takemoto Classifications Are Comparable in Predicting High-Risk Gastric Precancerous Lesions.

INTRODUCTION: Severe and extensive gastric atrophy, extensive or incomplete gastric intestinal metaplasia, and gastric dysplasia are considered high-risk gastric precancerous lesions (HGPLs). Endoscopic findings based on the endoscopic Kyoto classification (EKC) and the Kimura-Takemoto classification (KTC) have been reported to be significantly associated with HGPLs. This study aimed to compare these two classifications in predicting active Helicobacter pylori (H. pylori) infection and HGPLs. METHODS: This is a cross-sectional study conducted on naïve dyspeptic patients who underwent upper gastrointestinal endoscopy at a tertiary hospital. Endoscopic findings were scored according to the EKC and KTC. Mapping biopsies were taken, and H. pylori infection was determined using a locally validated rapid urease test and histology. The performance of EKC was compared with that of KTC using the area under the receiver operating characteristic curve (AUC) in predicting active H. pylori infection and HGPLs. RESULTS: There were 292 patients with a median age of 46 and a male-to-female ratio of 1:1. The rates of active H. pylori infection and HGPLs were 61.3% and 14.0%, respectively. The EKC was better than the KTC in predicting active H. pylori infection (AUC: 0.771 vs. 0.658, respectively; p < 0.001). However, these two classifications had comparable performance in predicting HGPLs (AUC: 0.792 vs. 0.791, respectively; p = 0.956). CONCLUSION: Compared to EKC, KTC is inferior in predicting active H. pylori infection but has comparable performance in predicting HGPLs.

Diagnosing and grading gastric atrophy and intestinal metaplasia using semi-supervised deep learning on pathological images: development and validation study.

OBJECTIVE: Patients with gastric atrophy and intestinal metaplasia (IM) were at risk for gastric cancer, necessitating an accurate risk assessment. We aimed to establish and validate a diagnostic approach for gastric biopsy specimens using deep learning and OLGA/OLGIM for individual gastric cancer risk classification. METHODS: In this study, we prospectively enrolled 545 patients suspected of atrophic gastritis during endoscopy from 13 tertiary hospitals between December 22, 2017, to September 25, 2020, with a total of 2725 whole-slide images (WSIs). Patients were randomly divided into a training set (n = 349), an internal validation set (n = 87), and an external validation set (n = 109). Sixty patients from the external validation set were randomly selected and divided into two groups for an observer study, one with the assistance of algorithm results and the other without. We proposed a semi-supervised deep learning algorithm to diagnose and grade IM and atrophy, and we compared it with the assessments of 10 pathologists. The model's performance was evaluated based on the area under the curve (AUC), sensitivity, specificity, and weighted kappa value. RESULTS: The algorithm, named GasMIL, was established and demonstrated encouraging performance in diagnosing IM (AUC 0.884, 95% CI 0.862-0.902) and atrophy (AUC 0.877, 95% CI 0.855-0.897) in the external test set. In the observer study, GasMIL achieved an 80% sensitivity, 85% specificity, a weighted kappa value of 0.61, and an AUC of 0.953, surpassing the performance of all ten pathologists in diagnosing atrophy. Among the 10 pathologists, GasMIL's AUC ranked second in OLGA (0.729, 95% CI 0.625-0.833) and fifth in OLGIM (0.792, 95% CI 0.688-0.896). With the assistance of GasMIL, pathologists demonstrated improved AUC (p = 0.013), sensitivity (p = 0.014), and weighted kappa (p = 0.016) in diagnosing IM, and improved specificity (p = 0.007) in diagnosing atrophy compared to pathologists working alone. CONCLUSION: GasMIL shows the best overall performance in diagnosing IM and atrophy when compared to pathologists, significantly enhancing their diagnostic capabilities.

Clinical Usefulness of Linked Color Imaging for Detection and Characterization of UndifferentiatedType Early Gastric Cancer.

BACKGROUND: Linked color imaging is based on the bioluminescent imaging technique, which enhances differences in mucosal color allowing for contrast-based detection of lesions. There have been reports which have investigated the usefulness of linked color imaging for assessing color values in endoscopy for early gastric cancer cases. However, these primarily focused on differentiated early gastric cancer. This study aimed to assess the efficacy of linked color imaging in analyzing the color differences between cancerous and noncancerous areas in undifferentiated-type early gastric cancer patients compared with conventional white light imaging. METHODS: Forty-six patients were prospectively enrolled with undifferentiated-type early gastric cancer from 3 academic hospitals. All lesions were observed first by white light imaging followed by linked color imaging. An additional biopsy was taken from the surrounding mucosa to check for intestinal metaplasia, and test for Helicobacter pylori was performed. Color difference was measured in accordance with the International Commission on Illumination details. RESULTS: The color difference value with linked color imaging was significantly higher, being more than twice that of white light imaging (26.82 ± 14.18 and 12.60 ± 6.42, P < .001), and this difference appeared to be similar in cases of accompanying Helicobacter pylori infection or intestinal metaplasia. In the subgroup analysis, color difference of poorly differentiated adenocarcinoma was notable in linked color imaging compared to white light imaging. Conversely, no statistically significant finding was present in signet ring cell carcinoma or mixed-type histology. CONCLUSION: Linked color imaging provides a significantly greater color difference between cancerous lesions and background noncancerous mucosa in undifferentiated-type early gastric cancer. Moreover, linked color imaging may differentiate between pathologic subgroups of undifferentiated-type early gastric cancer possibly due to characteristic cellular growth pattern.

Real-time gastric intestinal metaplasia diagnosis tailored for bias and noisy-labeled data with multiple endoscopic imaging.

This work presents real-time segmentation viz. gastric intestinal metaplasia (GIM). Recently, GIM segmentation of endoscopic images has been carried out to differentiate GIM from a healthy stomach. However, real-time detection is difficult to achieve. Conditions are challenging, and include multiple color modes (white light endoscopy and narrow-band imaging), other abnormal lesions (erosion and ulcer), noisy labels etc. Herein, our model is based on BiSeNet and can overcome the many issues regarding GIM. Application of auxiliary head and additional loss are seen to improve performance as well as enhance multiple color modes accurately. Further, multiple pre-processing techniques are utilized for leveraging detection performance: namely, location-wise negative sampling, jigsaw augmentation, and label smoothing. Finally, the decision threshold can be adjusted separately for each color mode. Work undertaken at King Chulalongkorn Memorial Hospital examined 940 histologically proven GIM images and 1239 non-GIM images, obtained over 173 frames per second (FPS). In terms of accuracy, our model is seen to outperform all baselines. Our results demonstrate sensitivity, specificity, positive predictive, negative predictive, accuracy, and mean intersection over union (IoU), achieving GIM segmentation values of 91%, 96%, 91%, 91%, 96%, and 55%, respectively.

Detection of erosions and fat metaplasia of the sacroiliac joints in patients with suspected sacroiliitis using a chemical shift-encoded sequence (IDEAL-IQ).

PURPOSE: To evaluate the performance of a chemical shift-encoded sequence called IDEAL-IQ for detecting sacroiliac joint (SIJ) erosions and fat metaplasia compared to T1-weighted fast spin echo (T1 FSE) using qualitative and quantitative analysis. METHOD: Thirty-four patients with suspicion of sacroiliitis who underwent both MRI and CT were included. Each SIJ was divided into four quadrants for analysis. For qualitative analysis, the diagnostic performance of IDEAL-IQ and T1 FSE for erosions were compared by the McNemar test, using CT as the gold standard. Cochran's Q and McNemar tests were used to determine differences in structural changes detected by different imaging methods. For quantitative analysis, two-sample t test and receiver operating characteristic (ROC) analysis were used for the analysis of histogram parameters of proton density fat fraction (PDFF). RESULTS: Diagnostic sensitivity and accuracy of IDEAL-IQ were greater than T1 FSE for erosions (all P < 0.05). IDEAL-IQ and CT detected more erosions than T1 FSE (all P < 0.05). IDEAL-IQ did not statistically significantly differ from T1 FSE for the detection of fat metaplasia (P = 0.678). All histogram parameters were different between groups with and without fat metaplasia (all P < 0.05) and could distinguish the two groups (all P < 0.05). PDFF75th was the most effective histogram parameter. CONCLUSION: IDEAL-IQ detects SIJ erosions with better accuracy than T1 FSE and is similar to T1 FSE for detection of fat metaplasia, enabling further quantitative analysis of the latter via histogram analysis.

Proton density fat fraction (PDFF) maps aid fat metaplasia evaluation in the sacroiliac joints in additional to T1WI: Improved diagnostic accuracy in axial spondyloarthritis.

PURPOSE: To evaluate the added value of qualitative and quantitative fat metaplasia analysis using proton-density fat fraction (PDFF) map in additional to T1-weighted imaging (T1WI) of the sacroiliac joints (SIJ) for diagnosis of axial spondyloarthritis (axSpA). METHOD: Patients aged 18-45 years with axSpA were enrolled. Non-SpA patients and healthy volunteers were included as controls. All participants underwent 3.0T MRI of the SIJs including semi-coronal T1WI and semi-coronal chemical-shift encoded MRI sequence for generating PDFF map. Each joint was divided into four quadrants for analysis. Two independent readers scored fat metaplasia on T1WI alone or with additional PDFF map and measured PDFF values in different reading sessions. Using clinical diagnosis as the reference, diagnostic accuracy of visual scores and PDFF measurements was evaluated by area under the receiver operating characteristic curve (AUC). Inter-reader agreement was evaluated by the intra-class correlation coefficient (ICC). RESULTS: Forty-nine patients with axSpA and thirty-six controls were included. Qualitative fat metaplasia scores using additional PDFF map performed better than using T1WI alone (AUC: Reader 1, 0.847 vs 0.795, p = 0.082; Reader 2, 0.785 vs 0.719, p = 0.048). AUCs of quantitative analysis using number of quadrants with PDFF value ≥75 % were higher than qualitative analysis using T1WI alone (Reader 1, 0.863 vs 0.795, p = 0.046; Reader 2, 0.823 vs 0.785, p = 0.011). ICCs were 0.854 to 0.922 for qualitative analysis and 0.935 for quantitative analysis. CONCLUSIONS: Additional PDFF map can increase the diagnostic accuracy for axSpA by qualitative and quantitative fat metaplasia analysis, in comparison to using T1WI alone.

Role of magnifying narrow-band imaging endoscopy for diagnosis of Helicobacter pylori infection and gastric precancerous conditions: Few issues.

Standard endoscopy with biopsy and narrow-band imaging with guided biopsy are techniques for the detection of Helicobacter pylori (H. pylori)-related gastritis and precancerous lesions. In this study, the authors compared standard endoscopy and magnified narrow-band imaging (commonly known as NBI-M) in the diagnosis of H. pylori infections, atrophic gastritis, and intestinal metaplasia. Although the sensitivity of NBI-M is better than standard endoscopy, the diagnostic accuracy did not differ substantially between the diagnostic modalities. Future prospective studies may guide endoscopists in difficult cases regarding which modality is more useful and cost-effective for the diagnosis of H. pylori-related gastritis and precancerous conditions.

Comparison of the effectiveness of i-scan and conventional endoscopy in the detection of the endoscopic signs of atrophic gastritis: A clinical trial.

BACKGROUND AND STUDY AIMS: This study aimed to determine whether the use of i-scan endoscopy provides additional benefits to conventional endoscopy in the diagnosis of gastric precancerous lesions. PATIENTS AND METHODS: A total of 120 patients with histologically-verified intestinal metaplasia (IM) or atrophic gastritis (AG) were prospectively evaluated by esophagogastroduodenoscopy. Endoscopic examinations were performed using i-scan and high-definition white-light endoscopy (HD-WLE). The diagnostic yields of both techniques and the number of targeted biopsies per patient were compared. RESULTS: A total of 318 suspicious lesions were detected in 108 patients with i-scan (n = 186) and 81 patients with HD-WLE (n = 132). The diagnostic yields of i-scan and HD-WLE were 81.6% (98/120) versus 77.5% (93/120), respectively (p > 0.05). When only targeted biopsies were taken into account, the diagnostic yields of i-scan and HD-WLE were 89.8% versus 65.4%, respectively (p < 0.05). The mean number of biopsies per patient for i-scan and HD-WLE were 3.27 (393/120) and 7.3 (882/120), respectively (p < 0.05). The mean endoscopic procedure times were 16 and 17 min for i-scan and HD-WLE, respectively (p > 0.05). CONCLUSIONS: Although targeted biopsies with i-scan were not found to be significantly superior to either targeted or random biopsies with HD-WLE, the number of biopsies required to confirm these lesions was much lower.

Performance of chromoendoscopy and narrow-band imaging in the diagnosis of gastric intestinal metaplasia.

BACKGROUND/AIMS: Chromoendoscopy and narrow-band imaging (NBI) have been reported to aid in the diagnosis of gastric intestinal metaplasia (GIM). This study aimed to assess the diagnostic validity of chromoendoscopy combined with NBI in the diagnosis of GIM in Vietnamese. METHODS: A cross-sectional study was carried out on patients with dyspeptic symptoms who underwent esophagogastroduodenoscopy (EGD) at the University Medical Center at Ho Chi Minh City. We compared the detection rates of GIM in the group of patients examined with white-light endoscopy (WLE) alone and those examined with WLE in combination with chromoendoscopy and NBI. RESULTS: A total of 374 patients have been recruited. The additional GIM detection rate after chromoendoscopy combined with NBI was 8.6% (95% confidence interval [CI]: 4.3 - 12.8), p < .005. The rate of GIM within the group of patients biopsied under the guidance of chromoendoscopy combined with NBI was statistically significantly higher than in the group with WLE alone with a distinct rate of 14.4% (95% CI: 6.3 - 2.6), p = .001. CONCLUSIONS: Chromoendoscopy combined with NBI helped to detect the GIM lesions missed by WLE and was a more reliable endoscopic method for the diagnosis of GIM.

Spindle Cell Metaplastic Breast Carcinoma.

INTRODUCTION: Metaplastic breast carcinoma is an uncommon malignancy that constitutes < 5% of all breast cancers. There are 5 subtypes which are spindle cell, squamous cell, carcinosarcoma, matrix-producing and metaplastic with osteoclastic giant cells. Spindle cell carcinoma represents approximately <0.3% of invasive breast carcinomas. It is typically a triple-negative cancer with distinct pathological characteristics, but relatively a non-conclusive on imaging findings. CASE REPORT: An elderly lady presented with an enlarging painful left breast lump for one year. Palpable left breast lump was found on clinical examination. Mammography demonstrated a high density, oval lesion with a partially indistinct margin. Corresponding ultrasound showed a large irregular heterogeneous lesion with solid-cystic areas. Histopathology showed atypical spindle-shaped cells that stained positive for cytokeratins and negative for hormone and human epidermal growth factor receptors, which favoured spindle cell metaplastic carcinoma. Left mastectomy and axillary dissection were performed, and the final diagnosis was consistent with metaplastic spindle cell carcinoma. CONCLUSION: Spindle cell carcinoma of the breast is a rare and aggressive histological type of carcinoma, which may present with benign features on imaging. Tissue diagnosis is essential for prompt diagnosis with multidisciplinary team discussion to guide management and improve patient's outcomes.

Enhanced Visibility in Evaluating Gastric Cancer and Helicobacter pylori-Associated Gastritis Using Linked Color Imaging with a Light-Emitting Diode Light Source.

BACKGROUND: In Japan, laser light source (Laser) endoscopy is widely available, and the characteristics of light-emitting diode light source (LED) endoscopy have not been clarified. AIMS: We assessed the visibility of early gastric cancers (EGCs) and Helicobacter pylori (H. pylori)-associated gastritis for LED endoscopy compared with laser endoscopy using white-light imaging (WLI) and linked color imaging (LCI). METHODS: We assessed 99 lesions between February 2019 and March 2020. The visibility was scored from four (excellent visibility) to one (poor visibility) by evaluating videos including EGCs and gastric mucosa captured using WLI and LCI with LED endoscopy (LED-WLI and LED-LCI, respectively) and laser endoscopy (Laser-WLI and Laser-LCI, respectively). The primary end point was the non-inferiority of the visibility of EGCs and H. pylori-associated gastritis between LED-/Laser-WLI and LED-/Laser-LCI. RESULTS: The visibility scores of EGCs for LED-/Laser-WLI and LED-/Laser-LCI were 3.14/2.97 and 3.39/3.35, respectively. The visibility scores of H. pylori-associated gastritis [intestinal metaplasia (IM), diffuse redness (DR), regular arrangement of collecting venules (RAC) and map-like redness (MR)] for LED-/Laser-WLI and LED-/Laser-LCI were 3.05/2.85 and 3.60/3.50 (IM), 2.76/2.50 and 2.96/2.86 (DR), 2.69/2.44 and 2.77/2.62 (RAC) and 2.97/2.75 and 3.39/3.27 (MR). Non-inferiority was demonstrated for visualizing EGCs and H. pylori-associated gastritis. CONCLUSIONS: LED-WLI and LED-LCI can be used to visualize EGCs and H. pylori-associated gastritis with non-inferiority to Laser-WLI and Laser-LCI. Furthermore, even with LED, LCI was more effective than WLI for evaluating EGCs and H. pylori-associated gastritis. Therefore, LED endoscopy can be used to detect EGCs and evaluate H. pylori-associated gastritis accurately.

Quantification of bone marrow oedema and fat metaplasia in sacroiliac joints in spondyloarthritis patients using histographic magnetic resonance imaging analysis.

OBJECTIVES: To demonstrate a possible basis for a quantitative magnetic resonance imaging (MRI) approach that uses histographic analysis to determine bone marrow oedema (BME) and fat metaplasia at sacroiliac joints (SIJs) level in patients with axial spondyloarthritis (axSpA). METHODS: In this prospective, cross-sectional study, consecutive axSpA patients with inflammatory low back pain underwent 1.5-T MRI. MRI images were scored on a 4-point (0-3) scoring system both for BME and fat metaplasia by two radiologists. A region-of-interest based histographic quantitative analysis was used to assess MRI images. Using the area under the receiver operating characteristic curve (AUC-ROC) approach was tested the diagnostic accuracy of histographic analysis for detecting BME vs. BME and fat metaplasia on MRI images. RESULTS: 17 of the 43 patients (39.5%) included only had a BME lesion, while the remaining 26 patients (60.5%) had both BME and fat metaplasia at the SIJ level. Inter-rater agreement between readers was good (weighted kappa 0.643). On MRI images, BME and BME+fat metaplasia showed significant difference in histographic analysis (p<0.001), with an AUC-ROC of 0.898, and an optimal cut-off point of 311 at histographic analysis in the distinction of BME vs. fat metaplasia. CONCLUSIONS: Histographic analysis could represent a method for quantifying BME on MRI images of SIJs in patients with axSpA. This type analysis can provide important prognostic information and guide the choice of treatment in patients with sacroiliitis.

Osseous metaplasia of the endometrium: A multicenter retrospective study.

OBJECTIVE: To describe clinical and demographic characteristics, ultrasound appearance, and hysteroscopic outcomes of patients with endometrial osseous metaplasia. STUDY DESIGN: We conducted a multicenter retrospective study. We retrospectively reviewed the medical records of all consecutive patients who were referred for hysteroscopy at fourteen institutions in Venezuela, Spain, Morocco, India, Ukraine, Argentina, the United States, and Italy between 1994 and 2018. We identified and included all patients who had a diagnosis of osseous metaplasia at the pathologic report, and data were retrieved from the medical records. RESULTS: Between January 1st, 1994, and December 31st, 2018, 63 patients out of a total of 419,673 women who underwent hysteroscopy had a diagnosis of osseous metaplasia (0.015%). Most patients were 31-40 years old (53.7%), were Caucasian or Hispanic (95.5%), and had at least one previous pregnancy (86.9%). Forty-one out of 63 patients (65.1%) had at least one miscarriage before the index hysteroscopy. Dysmenorrhea, abnormal uterine bleeding, and infertility were reported by 34.9%, 27.0%, and 23.8% of patients. 14.3% of women were asymptomatic. Preoperative transvaginal ultrasound was available and identified a hyperechoic area of variable size with posterior acoustic shadowing in all cases. Hysteroscopy was successful without complications in all 63 cases. Follow-up data were available in 30.2% of patients: 69.2% of patients were infertile, and 44.4% of them conceived and achieved a live birth; all other symptoms improved after hysteroscopic treatment in all patients. CONCLUSIONS: Osseous metaplasia appears associated with multiple unspecific gynecologic symptoms without the predominant role of infertility, as previously suggested. Although endometrial osseous metaplasia is rare, gynecologists should consider this pathologic condition when the characteristic ultrasound appearance is detected, being hysteroscopic treatment effective.

Narrow band imaging for detection of gastric intestinal metaplasia and dysplasia: A systematic review and meta-analysis.

BACKGROUND AND AIMS: Gastric intestinal metaplasia (GIM), a precursor of gastric adenocarcinoma, is challenging to diagnose with white light endoscopy (WLE) and can be missed by random gastric biopsies. Narrowband imaging (NBI) may potentially improve the detection of GIM. However, pooled estimates from prospective studies are lacking. METHODS: Electronic databases were searched for studies comparing NBI and WLE alone for detection of GIM and synchronous dysplasia. Primary outcome was pooled detection rate of GIM by NBI compared with WLE in prospective studies. The secondary outcome was concurrent dysplasia detection. RESULTS: Ten studies were found eligible from 306 articles screened. Eight prospective studies were found eligible for primary endpoint of GIM detection. Two other retrospective studies were included for dysplasia detection. A total of 1366 subjects (694 males, 54.4 ± 5.08 years) underwent upper endoscopy. GIM was detected in 482 (35.3%) subjects. NBI detected GIM in 32% additional subjects (70% vs 38%, RR 1.79; 95% CI 1.34-2.37; P < 0.01). Subgroup analysis revealed newer NBI scopes (GIF260) detected significantly more GIM than WLE (RR 2.47; 95% CI 1.63-3.76; P < 0.01) but not the older (H180) NBI endoscopes (RR 1.33; 95% CI 0.93-1.88; P = 0.11). There was moderate heterogeneity between the studies (I2  = 63%). In five studies (n = 628) that reported dysplasia, there was no significant difference between NBI and WLE in dysplasia detection (RR 1.09; 95% CI 0.81-1.47; P = 0.58). CONCLUSION: Narrowband imaging can significantly increase the detection of GIM when used in addition to standard white light exam during an upper endoscopy.

Endometrial osseous metaplasia in a patient with secondary infertility referred for hysterosalpingogram with ultrasound.

Endometrial osseous metaplasia is a rare entity resulting in the formation of trabecular bone fragments within the uterine cavity and frequently presents as secondary infertility with a history of previous pregnancy loss or termination. The unusual transvaginal ultrasound appearances are important to recognise, as fertility is often restored after hysteroscopic removal of the bone fragments from the uterine cavity.

Gastric intestinal metaplasia assessment between linked color imaging based on endoscopy and pathology.

OBJECTIVE: Cumulative evidence suggests that linked color imaging (LCI) can be used to identify gastric intestinal metaplasia (GIM). We aimed to develop endoscopic grading for GIM (EGGIM) with LCI. METHODS: Two hundred and seventy-seven patients who underwent high-resolution white-light gastroscopy followed by LCI for EGGIM estimation were included. LCI was performed for the entire mucosa, and images of five areas each were recorded from the lesser and greater curvatures of the antrum and corpus, and for the incisura. For each area, scores of 0 (no GIM), 1 (focal GIM, ≤30% of the area), and 2 (extensive GIM, >30% of the area) were attributed for 10 points. If GIM was suspected based on endoscopy findings, targeted biopsies were performed; if GIM was not evident, random biopsies were performed according to the Sydney system to estimate the operative link on GIM (OLGIM). RESULTS: GIM was staged as OLGIM 0, I, II, III, and IV in 136, 70, 37, 28, and 6 patients, respectively. For OLGIM III/IV diagnosis, the area under the receiver operating curve was 0.949 (95% CI 0.916-0.972). EGGIM of 4, with sensitivity and specificity of 94.12% (95% CI 80.3%-99.3%) and 86.42% (95% CI 81.5%-90.5%), respectively, was determined the best cut-off value for identifying OLGIM III/IV patients. CONCLUSIONS: Our findings demonstrated the ability of EGGIM for diagnosing the extent of intestinal metaplasia and showed that EGGIM is related to OLGIM staging. EGGIM of 4 was the best cut-off value for identifying OLGIM III/IV patients.

Intelligent diagnosis of gastric intestinal metaplasia based on convolutional neural network and limited number of endoscopic images.

BACKGROUND: Gastric intestinal metaplasia (GIM) is a precancerous lesion of gastric cancer. Currently, diagnosis of GIM is based on the experience of a physician, which is liable to interobserver variability. Thus, an intelligent diagnostic (ID) system, based on narrow-band and magnifying narrow-band images, was constructed to provide objective assistance in the diagnosis of GIM. METHOD: We retrospectively collected 1880 endoscopic images (1048 GIM and 832 non-GIM) via biopsy from 336 patients confirmed histologically as GIM or non-GIM, from the Kiang Wu Hospital, Macau. We developed an ID system with these images using a modified convolutional neural network algorithm. A separate test dataset containing 477 pathologically confirmed images (242 GIM and 235 non-GIM) from 80 patients was used to test the performance of the ID system. Experienced endoscopists also examined the same test dataset, for comparison with the ID system. One of the challenges faced in this study was that it was difficult to obtain a large number of training images. Thus, data augmentation and transfer learning were applied together. RESULTS: The area under the receiver operating characteristic curve was 0.928 for the pre-patient analysis of the ID system, while the sensitivities, specificities, and accuracies of the ID system against those of the human experts were (91.9% vs. 86.5%, p-value = 1.000) (86.0% vs. 81.4%, p-value = 0.754), and (88.8% vs. 83.8%, p-value = 0.424), respectively. Even though the three indices of the ID system were slightly higher than those of the human experts, there were no significant differences. CONCLUSIONS: In this pilot study, a novel ID system was developed to diagnose GIM. This system exhibits promising diagnostic performance. It is believed that the proposed system has the potential for clinical application in the future.