Introduction

At present, there are three developmental pathways for colorectal cancers, namely, the adenoma-carcinoma sequence, serrated pathway, and de novo pathway. The adenomatous route of colorectal cancer carcinogenesis has been extensively reviewed in the literature. Conversely, the de novo pathway, which leads depressed type polyps to colorectal cancer, lacking an adenomatous phase, recently has garnered increasing attention. Research has indicated that nonpolypoid lesions demonstrate a greater potential for malignancy, underscoring the need for their early detection. This study aims to explore the endoscopic characteristics of depressed type colorectal neoplasms by using magnifying endoscopy (pit pattern classification) and endocytoscopy (EC classification). Both the structural and cellular atypias can be observed in vivo, thus enabling more comprehensive evaluation of the disease.

Aims & Methods

A total of 40719 colorectal neoplasms resected endoscopically or surgically in our institution from April 2001 to June 2022 were subject of this study. Advanced cancers were excluded. Of these, 1463 lesions were submucosally invasive (T1) carcinomas, 6839 lesions were high-grade dysplasia and 32417 lesions were characterized as low-grade dysplasia. Morphologically, they were divided into 3 types: depressed, flat and protruded type comprising 1.2%, 42.9% and 55.9%, respectively. Specifically, the distinctive features of depressed type neoplasms concerning its pit pattern and EC classification were examined.

Results

Most of the flat (93.3%) and protruded type (94.7%) lesions showed type III _L or IV pit pattern, which are common findings in adenomas.On the other hand, most of the depressed type lesions(92.4%) classified by type III s, V _I or V _N pit patterns, are typically seen in carcinomas. As for endocytoscopy, most of the flat and protruded type lesions were identified as EC2 corresponding to adenomas. In contrast, a relatively high rate of depressed type lesions was categorized as EC3a (40.0%) and EC3b (52.0%), consistent with invasive carcinomas. The proportion of T1 carcinomas in depressed type lesions was 61.3%, whereas in the flat and protruded type lesions it was 2.9% and 2.9%, respectively. Also, 10% of lesions less than 5mm in diameter were depressed, 0.01% were flat and 0% were protruded.

Conclusion

This investigation has revealed that depressed lesions can be classified in magnifying endoscopy as type III s, V I or V N pit patterns, and as types EC3a or EC3b in endocytoscopy. Lesions that have these features or these pit patterns also have a propensity to invade the submucosal layer even when they are small in size. Despite their elevated malignant potential, depressed type lesions are frequently disregarded, underscoring the critical importance of recognizing their distinctive features and accurately diagnosing lesions based on their morphology, pit pattern, and EC classification.

Disclosure

None.

Introduction

For training deep learning algorithms in medical imaging, application-specific data is often scarce. Deep learning systems are therefore generally pretrained on large publicly available labeled data sets of general imagery, unrelated to the envisioned application, to have the algorithm learn basic features from these widely available data followed by refinement training on the generally scarce application-specific images. Pretraining might be more effective if the images for pretraining resemble the envisioned application, i.e., domain-specific pretraining.

Aims & Methods

We investigated if pretraining on general endoscopic imagery results in a better performance of five existing AI systems with an application in gastro-intestinal endoscopy, compared to current state-of-the art pretraining approaches (i.e., supervised pretraining with ImageNet and semi-weakly supervised pretraining with the Billion-scale data set). Our group has created an endoscopy-specific dataset called GastroNet for pretraining deep learning systems in endoscopy. GastroNet consists of 5,084,494 endoscopic images retrospectively collected between 2012 and 2020 in seven Dutch hospitals. We created four pretrained models: one using GastroNet and three using ImageNet and/or the Billion-scale data set. The pretraining method was either supervised, self-supervised, or semi-weakly supervised. The pretrained models were subsequently trained towards five independent, commonly used applications in GI endoscopy, using their original application-specific datasets. Outcome parameters were: 1) classification and/or localization performance of the five trained applications; 2) change in performance when the number of available application-specific training data was reduced, to investigate a possible difference in performance drop for the different pretrained models.

Results

Overall, the domain-specific pretrained model resulted in a statistically superior performance for the five different GI applications. More detailed results are presented in Table 1. The superiority was also reflected in a smaller drop in performance when the number of application-specific training data were reduced artificially.

Table 1. Overview of performance of the five different application-specific data sets using four different pretrained models. Cells highlighted in green represent the highest scoring pretrained model per application-specific data set.

Conclusion

Domain-specific pretraining, using unlabeled general endoscopic images, is superior to current state-of-the-art pretraining approaches for developing deep learning algorithms in GI endoscopy. It also allows more effective use of the generally scarce application-specific endoscopy images. These findings might cause a paradigm shift in the development of AI systems in endoscopy.

Introduction

In recent years, cold snare polypectomy (CSP) for small polyps has become popular, but it is considered inappropriate as a treatment for cancer because of the shallow depth of resection at the muscularis mucosa level and the risk of inadequate pathological evaluation. ¹² Therefore, it is important to diagnose advanced histology (AH) (high grade adenoma or adenocarcinoma) prior to treatment. Based on our retrospective review of colorectal neoplastic lesions smaller than 10 mm, the following six endoscopic findings were extracted as significant predictors of AH findings in the multivariate analysis : (1) surface redness, (2) white spots, using white light image, (3) heterogeneity in mucosal pattern, using non-magnified Narrow band imaging (NBI), (4) irregular vessel or surface pattern(equivalent to JNET Type 2B,3), using magnified NBI, (5) loss of lobulation, and (6) non-polypoid growth, using non-magnified chromoendoscopy using indigo carmine. ³

Aims & Methods

The aim of this study was to evaluate the usefulness of image-enhanced endoscopic findings in real-time prediction of AH for small colorectal polyps. Patients scheduled for colonoscopy were recruited for consent prior to the examination at 9 Japanese institutions, and the presence of the aforementioned AH-predictive endoscopic findings was noted for up to 5 lesions per patient when neoplastic polyps of 5-9 mm were detected. If one of the predictive findings was positive, endoscopic resection using the electric current method or surgical resection was applied. The primary endpoint was the negative predictive value (NPV) of the 6 AH-predicting findings using pathology as the gold standard. Secondary endpoints were NPV of AH prediction by modality combination, odds ratios of each predictors and percentage of non-CSP for T1 cancer. Taking account for practical value, the target threshold of NPV was set for 90%.

Results

A total of 1671 patients were enrolled in the study, of which 517 patients with 849 lesions (80 AH lesions) corresponding to 5-9 mm polyps were included in the analysis. The primary endpoint was 0.945 (95% confidence interval (CI): 0.923-0.961). Secondary endpoints were white light + NBI non-magnified/magnified 0.941 (95%CI: 0.920-0.958), white light + NBI non-magnified 0.934 (95%CI: 0.913-0.952), white light + chromoendoscopy 0.926 (95%CI: 0.904-0.944), and white light 0.922 (95%CI: 0.900-0.940). Odds ratio (95%CI) of each 6 findings were as followings; irregular vessel or surface pattern 5.00 (2.59-9.37), loss of lobulation 4.10(1.61-6.56), white spots 3.32(1.60-6.56), non-polypoid growth 3.24(1.13-8.22), heterogeneity 2.53(1.41-4.42), surface redness 1.41(0.56-3.14). All 6 T1 cancer had multiple predictive findings of AH, and the percentage of non-CSP performed was 100%.

Conclusion

Six predictive findings of AH combined with image-enhanced endoscopy were shown to be useful in determining the indication for CSP (UMIN 000040974).

References

1. Ito A, Suga T, Ota H, et al. Resection depth and layer of cold snare polypectomy versus endoscopic mucosal resection. J Gastroenterol 2018;53:1171-1178.
2. Uraoka T, Takizawa K, Tanaka S, et al. Guidelines for Colorectal Cold Polypectomy (supplement to "Guidelines for Colorectal Endoscopic Submucosal Dissection/Endoscopic Mucosal Resection"). Dig Endosc 2022;34:668-675.
3. Iwai T, Imai K, Hotta K, et al. Endoscopic prediction of advanced histology in diminutive and small colorectal polyps. J Gastroenterol Hepatol 2019;34:397-403.

Introduction

While residual subcutaneous IFX levels are found significantly higher compared with IFX IV levels, the relationships between IFX SC levels and the depth of remission in IBD remain elusive.

Aims & Methods

All consecutive IBD patients treated with sc IFX 120mg/2weeks for at least 8 weeks were included in a multicenter study. All patients were switched from iv to sc IFX at least 22 weeks after the drug induction. Crohn’s disease patients with exclusive perineal lesions, pregnant women and patients with a body mass index higher than 30 were excluded. Clinical disease activity assessed by Crohn’s disease Activity Index, Harvey Bradshaw index and the partial Mayo score was recorded during a routine visit and serum CRP, fecal calprotectin and serum IFX and anti-infliximab antibodies levels were measured, regardless of the time of the previous drug injection (Elisa assay). Clinical remission was defined by a CDAI score < 150 or HBI< 5 for CD and a partial Mayo score < 3 for UC. Strong clinical remission was defined by both clinical remission and CRP < 5mg/L and deep remission by the presence of clinical remission, CRP < 5mg/L and fecal calprotectin < 250 µg/g. The results of biomarker and drug concentrations were blinded to those of clinical disease activity.

Results

Among the 71 patients (47 F, 70 % CD, 27 % with concomitant immunosuppressants) enrolled, 55 (77%) patients were in clinical remission, 44 (62%) patients in strong remission and 31 (43%) patients in deep remission. 40 (56%) patients had fecal calprotectin < 250 µg/g regardless the clinical activity. The mean (SD) IFX levels were significantly higher in patients in clinical remission than in those with active disease (17.3 µg/mL (4.9) vs 12.3 µg/mL (4.5) ; p=0.014). Whatever the selected therapeutic target, serum drug levels differed significantly (p=0.01 for strong remission and p<0.001 for deep remission and normal fecal calprotectin). Univariate analysis revealed that only the age was significantly associated with sc IFX levels above 20µg/mL (p=0.045). Conversely, sex, type or duration of IBD, previous anti-TNF and combotherapy were not associated with sc IFX serum drug concentration > 20µg/mL. reported The best cut off values according to the therapeutic target are reported in Table 1. serum sc IFX levels more than 20µg/mL were strongly associated with a deep remission (sensitivity : 0.91, specificity : 0.80, accuracy : 0.85).

Conclusion

Serum sc IFX levels are significantly higher in IBD patients who achieved clinical and/or biomarker remission. serum IFX levels ≥20µg/mL were associated with a deep remission and also with the normalization of fecal calprotectin with an overall accuracy of 0.85 for both.