Introduction

The optical diagnosis of T1 colorectal cancer (CRC) with deep invasion (≥T1b CRC) is crucial in determining subsequent treatment. Recently, computer-aided diagnosis (CADx) has become available for the diagnosis of deep invasive cancer, but until now, it must be combined with image enhanced endoscopy (IEE) or endocytoscopy. We have developed a CADx system to diagnose deep invasive cancer using non-magnified white light imaging (WLI).

Aims & Methods

This study aimed to compare the diagnostic accuracy of the CADx in T1 CRC diagnosis with that of non-expert endoscopists and expert endoscopists. A total of 365 images of 37 T1 CRCs (16 T1a CRCs and 21 T1b CRCs) from prospectively enrolled consecutive patients were included in the study. Images were independently diagnosed by two non-expert endoscopists and two expert endoscopists; sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and accuracy of T1b CRC diagnosis by the CAD system were calculated and compared with the endoscopists diagnostic accuracies.

Results

The sensitivity, specificity, PPV, NPV, and accuracy of CAD were 70.2%, 60.4%, 84.2%, 49.3%, and 70.7%, respectively, while those of non-experts were 39.9%, 80.6%, 82.7%, 36.6%, and 51.9%, respectively. The accuracy rate was significantly higher in the CAD group (p < 0.01). On the other hand, those of experts were 47.9%, 87.7%, 79.2%, 63.2%, and 67.7%, respectively, which were comparable to the diagnostic accuracy of CAD. The diagnostic time per image was significantly shorter for CAD than for endoscopists (0.49 sec. for CAD, 7.72 sec. for non-expert, and 5.77 sec. for expert, p < 0.01, p < 0.01, comparison with CAD).

Conclusion

Non-magnified WLI-based CADx is useful for non-expert endoscopists in diagnosing T1 CRC with deep invasion.

Disclosure

This work was supported by JSPS KAKENHI Grant Number 21K07885 and JFE (The Japanese Foundation for Research and Promotion of Endoscopy) Grant.

Introduction

Deep submucosal invasion is not considered an independent risk factor for lymph node metastasis in T1 colorectal cancer (CRC). This study aimed to evaluate the feasibility and effectiveness of diagnostic endoscopic resection for T1 CRCs with deep invasion.

Aims & Methods

This retrospective study included consecutive patients with T1 CRC who underwent either endoscopic submucosal dissection (ESD) or endoscopic mucosal resection (EMR). Pedunculated lesions (Paris classification Ip) were excluded. Low-risk T1 CRCs were defined as lesions with submucosal invasion ≥1000 µm that were resected en bloc with R0 margins and lacked lymphovascular invasion (LVI), high-grade budding (>BD1), or poor differentiation. These lesions were considered eligible for expanded curative resection.

Results

A total of 123 T1 CRC lesions in 121 patients were resected endoscopically (26 by EMR, 97 by ESD). The overall en bloc and R0 resection rates were 89.4% (100/123) and 65.9% (81/123), respectively. For T1b CRCs, these rates were 86.7% (52/60) and 56.7% (34/60). No perforations occurred. Expanded curative resection was achieved in 26.7% (16/60) of T1b cases. Among non-curatively resected T1b lesions, 50.0% (30/60) had LVI, 41.7% (25/60) had positive or indeterminate vertical margins (VM1/VMX), and 13.3% (8/60) showed significant tumor budding.

Conclusion

Diagnostic endoscopic resection for T1 CRCs with deep submucosal invasion is feasible and safe, with a notable rate of expanded curative resection. This strategy may support accurate pathological evaluation and informed treatment planning in selected patients.

Introduction

In early-stage colorectal cancer (CRC), distinguishing CRC with deep invasion is crucial in determining subsequent treatments. However, diagnosis remains difficult. Recently, we have developed and reported a computer-aided diagnostic (CAD) system to diagnose CRC with deep invasion, using only non-magnified white light imaging.1) This study aimed to compare the diagnostic accuracy of CAD for early-stage CRC with that obtained in a prospective study using magnified narrow-band imaging endoscopy (M-NBI) and dye-based magnified chromoendoscopy (MCE).2)

Aims & Methods

Among 1173 lesions registered in a multicenter prospective study, 1428 images of 162 early-stage CRC cases (125 Tis CRCs, 16 T1a CRCs and 21 T1b CRCs) registered at our hospital were included. Sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and accuracy rate of the CAD system for diagnosing T1b (depth of invasion ≥1000 μm) CRC were all calculated, and compared with those of M-NBI and MCE diagnoses. Optical diagnoses by M-NBI and MCE were performed by experts in real time.

Results

The sensitivity, specificity, PPV, NPV, and accuracy rate of CAD were 57.1%, 85.8%, 37.5%, 93.1%, and 82.1%, respectively. Whereas those of M-NBI were 71.4%, 93.6%, 62.5%, 95.7%, and 90.2%, respectively, and those of MCE were 85.7%, 91.5%, 60.0%, 97.7%, and 90.7%, respectively. The overall accuracy rate of CAD was significantly lower than those of M-NBI and MCE diagnoses. However, the accuracy rates for depressed lesions (IIa, IIa+IIc) were 87.5% for M-NBI, 81.3% for MCE, and 87.5% for CAD, and for polypoid lesions (Ip, Is) were 83.1% for M-NBI, 84.5% for MCE, and 76.1% for CAD, indicating that the accuracy rates of CAD were similar to those of M-NBI and MCE diagnoses in these cases.

Conclusion

CAD using non-magnified white light imaging is useful for the diagnosis of deep invasive cancer in depressed or polypoidal early-stage CRCs.

References

1) Tokunaga M, Matsumura T, et al. Gastrointest Endosc. 2021;93:647-653.
2) Matsumura T et al. Clin Gastroenterol Hepatol. 2023

Disclosure

This work was supported by JSPS KAKENHI Grant Number 21K07885 and JFE (The Japanese Foundation for Research and Promotion of Endoscopy) Grant.

Introduction

The incidence of early-stage colorectal cancer (CRC) rises with the introduction of colorectal screening programs; there is a shift towards less invasive treatment in screen-detected compared to non-screen-detected CRCs.

Aims & Methods

This retrospective, single-center cohort study aimed to assess the impact of colorectal screening on the incidence of submucosal invasive (pT1) CRC and compare treatment characteristics of screen-detected and non-screen-detected lesions between 2010 and 2023.
Screen-detected pT1 CRC (involving lesions detected with the pilot CRC screening in Csongrád-Csanád county in 2015, and those with the Hungarian population-based screening program [EFOP-1.8.1-VEKOP-15-2016-00001] initiated in 2019) lesion, patient and treatment characteristics were compared to pT1 lesions detected outside the screening program in average-risk patients aged 50-70 years during the screening period (2015-2016 and 2019-2023). The impact of screening on pT1 CRC incidence was assessed with investigating new pT1 CRC cases throughout the screening period (2015-2016 and 2019-2023) compared to non-screening periods (2010-2014 and 2017-2018).

Results

During the 14-year study period, a total of 182 pT1 CRCs were detected in average-risk patients aged 50-70 years. 81 of these lesions (44.5%) were detected during non-screening periods. During the screening period, 44 lesions were screen-detected (24.2%, 16 of these detected during the pilot screening) and 57 lesions were non-screen-detected (31.3%). Mean pT1 incidence was higher in the screening than in the non-screening periods (14.7 vs. 11.6 lesions/year, p<0.01); with the highest incidence during the pilot screening (2015-2016; 20 lesions/year), and the lowest during the pandemic (2020-2021, 9 lesions/year).
Patient and lesion characteristics were similar in screen-detected and non-screen-detected lesions (mean age: 63.1 vs 63.5 years, male ratio: 62.8% vs 64.3%, follow-up period: 3.85 vs 3.70 years; location: 31.8% vs 33.3% rectal, morphology: 34.1% vs 42.1% pedunculated, mean lesion size: 21.0 vs 24.7 mm). No difference in histologic prognostic factors were found between the two groups. Treatment choices were similar for both screen-detected, and non-screen-detected lesions: local excision in 65.9% vs 72.2% (incl. ER in 63.6% vs 63.2%, and minimally invasive local surgical resection in 2.3% vs 7.0%), and surgical oncologic resection in 34.1% vs 29.8%, respectively; p=0.167. Screen-detected colon cancers were more likely to be treated with surgical oncologic resection (43.3% vs 28.9%, p=0.218), while screen-detected rectal cancers were preferred to be treated with local excision (85.7% vs 68.4% [ER: 78.6% vs 47.4%], p=0.416) compared to non-screen-detected lesions.
During the follow-up, adverse outcomes (local recurrence and/or dissemination) occurred in two screen-detected, and in one non-screen-detected patient, p=0.797. Lymph node involvement rate was similar in those undergoing surgical oncologic resection in screen-detected, and non-screen-detected patients (6.7% vs 5.9%, respectively; p=0.927).

Conclusion

The incidence of submucosally invasive CRC rises with the implementation of colorectal screening program. There is no difference in the management and outcomes of screen-detected and non-screen-detected pT1 CRCs, but local excision, especially endoscopic resection is preferred for screen-detected rectal lesions.

Introduction

Accurate prediction of lymph node metastasis (LNM) in submucosal invasive (T1) colorectal cancer (CRC) is essential for determining the need for additional surgery following endoscopic resection. Conventional histopathological assessments, however, are limited by interobserver variability. This study aimed to develop and externally validate a stacking-based artificial intelligence (AI) model integrating histopathological whole slide images (WSIs) and clinical parameters to predict the risk of LNM in T1 CRC.

Aims & Methods

This retrospective study included 1,056 patients with histologically confirmed T1 CRC treated at Showa Medical University Northern Yokohama Hospital. Of these, 593 cases were used for model training and 100 for internal testing. External validation was conducted on 25 cases from two independent institutions.
The proposed stacking ensemble model incorporated the Multiple Instance Self-Training (MIST) score derived from WSIs, patient sex, and tumor location (colon vs. rectum). Base learners included logistic regression, XGBoost, and random forest, with a meta-learner synthesizing their outputs to generate the final LNM prediction. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC) in both internal and external cohorts.

Results

LNM was observed in 69 (11.6%) training cases, 15 (15.0%) test cases, and 2 (8.0%) external validation cases. The model achieved an AUC of 0.6965 in the internal test cohort and 0.7500 in the external validation cohort, demonstrating stable and generalizable predictive performance.
A representative case involved a 64-year-old man with a 16-mm type 0-Is lesion in the sigmoid colon resected via endoscopic mucosal resection (EMR). Histopathology showed well-differentiated tubular adenocarcinoma (tub1 > tub2), submucosal invasion depth of 3,300 µm (T1b), tumor budding grade BD1, Ly0, V1a, and negative resection margins (HM0, VM0). Despite guideline-based recommendations prompting additional surgical resection, no LNM was found (N0, 0/16), and the patient remained recurrence-free five years postoperatively—consistent with the AI model’s prediction.

Conclusion

The stacking-based AI model exhibited robust and generalizable performance in predicting LNM in T1 CRC, both internally and across external institutions. This approach offers a complementary tool to conventional pathology and may enhance individualized decision-making regarding the necessity of additional surgical intervention.

Introduction

Delayed bleeding (DB) is a major complication of colorectal endoscopic submucosal dissection (ESD). There are several definitions of DB, and one of the main ones is "cases requiring hemostasis with hematochezia or hemoglobin (Hb) decrease ≥2 g/dL as compared to the last preoperative level" [1]. However, it is not clear whether or not cases with Hb decrease actually requires endoscopic hemostasis, and this is sometimes difficult to determine in actual clinical practice.

Aims & Methods

We defined DB as a case requiring hemostasis with hematochezia or a Hb decrease ≥2 g/dL after colorectal ESD. The purpose of this study was to clarify the significance of Hb decrease in DB after colorectal ESD. Of the 595 patients (615 tumors) who underwent ESD or hybrid ESD for colorectal tumors at our hospital between June 2009 and April 2022, 145 patients (145 tumors) with suspected DB (suspected DB; with hematochezia or a Hb decrease ≥2 g/dL) were analyzed. We compared the hemostasis group (H group; hemostasis performed) and the no-hemostasis group (no-H group; no endoscopy performed or endoscopy performed but no hemostasis performed after hematochezia or Hb decrease) in a retrospective study.

Results

Thirty-two patients with 32 tumors were categorized into the H group, while 113 patients with 113 tumors were in the no-H group. Hematochezia was significantly higher in the H group than in the no-H group (100% vs 83.2%, p = 0.008, Fisher's exact test). On the other hand, the incidence of Hb decrease on postoperative day (POD) 1 was significantly higher in the no-H group than in the H group (26.5% vs 0%, p <0.001, Fisher's exact test). There were 7 cases in the 145 cases who had both hematochezia and a Hb decrease on POD 1, and every 7 cases were in the no-H group. Among the H group, only 8 cases (25%) had a Hb decrease ≥2 g/dL on the day of hemostasis compared to the preoperative level. On multivariate logistic regression analysis, date from ESD to first hematochezia ≤Day3 (OR 3.57, 95%CI 1.46–8.74; p = 0.005) was identified as the significant factor contributing to the no-H group, followed by bleeding duration ≤1 day (OR 4.58, 95%CI 1.45–14.43; p = 0.009).

Conclusion

The duration and the first date of hematochezia seemed to be important in determining whether to perform hemostasis when DB was suspected. On the other hand, there were no cases in which a Hb decrease ≥2 g/dL on POD 1 required hemostasis. This suggests that a Hb decrease may be a low priority factor in DB after colorectal ESD.
​

References

1. Tajiri H, Kitano S. Complications associated with endoscopic mucosal resection: Definition of bleeding that can be viewed as accidental. Dig Endosc 2004; 16: S134– S136.

Introduction

Optical diagnosis attempts to guide the endoscopist in making a treatment choice, however, selecting lesions with possible covert cancer (CONECCT IIA lesions: JNET IIA, no macronodule, no depressed area), high-risk lesions (CONECCT IIC lesions: JNET IIB or macronodule or depressed area) or overt submucosal invasive carcinoma (SMIC) (CONECCT III lesions: JNET III area) remains challenging.

Aims & Methods

We aim to describe the endoscopic and oncologic outcomes of endoscopically removed T1-colorectal cancers (T1-CRC), according to their pre-resection classification by optical diagnosis. This is a multicenter prospective cohort study in 13 centers. Procedural and oncological outcomes were analyzed using univariate analysis.

Results

7495 colorectal ESDs were included from 09/2019 to 09/2024, with a median lesion size of 50mm [35-65], rectal location in 35.3% and right colon in 37.9%. Lesions were classified as CONECCT IIA in 28.4%, CONECCT IIC in 60.3% and CONECCT III in 5.3%. En-bloc, R0 and curative resection rates were 96.3%, 88.1% and 81.4%, with significant decreasing trends as the CONECCT classification advanced (97.3%; 91.9%; 90.7% in CONECCT IIA, 96.3%; 87.6%; 80.6% in CONECCT IIC and 93.2%; 74.2%; 32.0% in CONECCT III, respectively; p<0.001). T1-CRC occurred in 10.4% and the prevalence significantly differed according to the optical diagnosis: 2.3% in CONECCT IIA, 10.8% in CONECCT IIC and 49.6% in CONECCT III. En-bloc, R0 and curative resection rates were 96.7%, 74.9% and 23.6% respectively, with a significant decreasing trend in curative resection rates as the CONECCT classification increased (25.5% CONECCT IIA, 26.5% CONECCT IIC and 13.9% CONECCT III; p=0.002). Vertical margins were free in 80.3%. Overall median depth of SMIC was 1500µm [750-2560], measuring 1426µm in CONECCT IIA, 1840µm in CONECCT IIC and 2744µm in CONECCT III (p<0.001), corresponding to deep SMIC in 46.8% in CONECCT IIA, 62.2% CONECCT IIC and 77.1% CONECCT III (p<0.001). Lympho-vascular invasion and tumor budding were more frequently observed in CONECCT III at 30.0% and 34.3% respectively (vs 23.9%; 23.9% in CONECCT IIA and 16.4%; 20.6% in CONECCT IIC; p=0.002). Poor differentiation was reported in 5.3% with no significant difference between categories. Finally a trend was noted in the number of prognostic risk factors per lesion: no high-risk factors in 34.0% of CONECCT IIA, 29.5% in CONECCT IIC and 18.6% in CONECCT III, versus 3 high-risk factors in 4.3% in CONECCT IIA, 7.0% in CONECCT IIC and 11.2% in CONECCT III (p=0.049).

Conclusion

Endoscopic resection of T1-CRC demonstrates high en-bloc and R0 resection rates. However, curative outcomes and oncologic features worsen significantly with the presence of risky optical diagnostic criteria, highlighting the importance of adequate optical diagnosis and considering intramuscular resection techniques for advanced lesions, balancing over- and undertreatment as nearly 50% of CONECCT III lesions were T1-CRC.

References

Brule C, Pioche M, Albouys J, et al. The COlorectal NEoplasia Endoscopic Classification to Choose the Treatment classification for identification of large laterally spreading lesions lacking submucosal carcinomas: A prospective study of 663 lesions. United European Gastroenterol J, 2022;10():80-92