Advertisement

Novel “resect and analysis” approach for T2 colorectal cancer with use of artificial intelligence

Published:April 29, 2022DOI:https://doi.org/10.1016/j.gie.2022.04.1305

      Background and Aims

      Because of a lack of reliable preoperative prediction of lymph node involvement in early-stage T2 colorectal cancer (CRC), surgical resection is the current standard treatment. This leads to overtreatment because only 25% of T2 CRC patients turn out to have lymph node metastasis (LNM). We assessed a novel artificial intelligence (AI) system to predict LNM in T2 CRC to ascertain patients who can be safely treated with less-invasive endoscopic resection such as endoscopic full-thickness resection and do not need surgery.

      Methods

      We included 511 consecutive patients who had surgical resection with T2 CRC from 2001 to 2016; 411 patients (2001-2014) were used as a training set for the random forest–based AI prediction tool, and 100 patients (2014-2016) were used to validate the AI tool performance. The AI algorithm included 8 clinicopathologic variables (patient age and sex, tumor size and location, lymphatic invasion, vascular invasion, histologic differentiation, and serum carcinoembryonic antigen level) and predicted the likelihood of LNM by receiver-operating characteristics using area under the curve (AUC) estimates.

      Results

      Rates of LNM in the training and validation datasets were 26% (106/411) and 28% (28/100), respectively. The AUC of the AI algorithm for the validation cohort was .93. With 96% sensitivity (95% confidence interval, 90%-99%), specificity was 88% (95% confidence interval, 80%-94%). In this case, 64% of patients could avoid surgery, whereas 1.6% of patients with LNM would lose a chance to receive surgery.

      Conclusions

      Our proposed AI prediction model has a potential to reduce unnecessary surgery for patients with T2 CRC with very little risk. (Clinical trial registration number: UMIN 000038257.)

      Graphical abstract

      Abbreviations:

      AI (artificial intelligence), AUC (area under the curve), CEA (carcinoembryonic antigen), CI (confidence interval), CRC (colorectal cancer), EFTR (endoscopic full-thickness resection), LNM (lymph node metastasis), RF (random forest), TAMIS (transanal minimally invasive surgery), TEM (transanal endoscopic microsurgery)
      Stage T2 colorectal cancer (CRC) invades the muscularis propria of the bowel wall but has not advanced into subserosa or pericolic tissue. About 25% of patients with T2 CRC have lymph node metastases (LNMs) at histopathologic assessment.
      • Mori Y.
      • Kudo S.E.
      • Endo S.
      • et al.
      Morphology as a risk factor for the malignant potential of T2 colorectal cancer.
      • Fields A.C.
      • Lu P.
      • Hu F.
      • et al.
      Lymph node positivity in T1/T2 rectal cancer: a word of caution in an era of increased incidence and changing biology for rectal cancer.
      • Kajiwara Y.
      • Ueno H.
      • Hashiguchi Y.
      • et al.
      Risk factors of nodal involvement in T2 colorectal cancer.
      • Baxter N.N.
      • Garcia-Aguilar J.
      Organ preservation for rectal cancer.
      • Ushigome H.
      • Ohue M.
      • Kitamura M.
      • et al.
      Evaluation of risk factors for lymph node metastasis in T2 lower rectal cancer to perform chemoradiotherapy after local resection.
      Therefore, current standard therapy for all T2 CRC patients is surgical resection with segmental colectomy and lymph node dissection, an invasive procedure with significant resources and patient burden.
      Recently, endoscopic full-thickness resection (EFTR) has emerged as an option to remove early-stage colorectal tumors.
      • Schmidt A.
      • Beyna T.
      • Schumacher B.
      • et al.
      Colonoscopic full-thickness resection using an over-the-scope device: a prospective multicentre study in various indications.
      • Zwager L.W.
      • Bastiaansen B.A.J.
      • Bronzwaer M.E.S.
      • et al.
      Endoscopic full-thickness resection (eFTR) of colorectal lesions: results from the Dutch colorectal eFTR registry.
      • Guillaumot M.A.
      • Barret M.
      • Jacques J.
      • et al.
      Endoscopic full-thickness resection of early colorectal neoplasms using an endoscopic submucosal dissection knife: a retrospective multicenter study.
      • Kuellmer A.
      • Behn J.
      • Beyna T.
      • et al.
      Endoscopic full-thickness resection and its treatment alternatives in difficult-to-treat lesions of the lower gastrointestinal tract: a cost-effectiveness analysis.
      • Kuellmer A.
      • Mueller J.
      • Caca K.
      • et al.
      Endoscopic full-thickness resection for early colorectal cancer.
      • Meier B.
      • Stritzke B.
      • Kuellmer A.
      • et al.
      Efficacy and safety of endoscopic full-thickness resection in the colorectum: results from the German Colonic FTRD Registry.
      • Toyonaga T.
      • Ohara Y.
      • Baba S.
      • et al.
      Peranal endoscopic myectomy (PAEM) for rectal lesions with severe fibrosis and exhibiting the muscle-retracting sign.
      • Ono S.
      • Kobayashi R.
      • Ito S.
      • et al.
      A case of successful full-thickness resection using endoscopic submucosal dissection and transanal suturing of rectal cancer.
      EFTR enables full-thickness bowel resection and provides a complete specimen for histopathologic diagnosis. EFTR is less invasive and thus associated with fewer resources and patient burden than surgical resection of colorectal tumors. A candidate for EFTR with T2 CRC is presented in Figure 1 in which the tumor size was <20 mm, for which R0 resection was expected to be performed with acceptable adverse event rates.
      • Schmidt A.
      • Beyna T.
      • Schumacher B.
      • et al.
      Colonoscopic full-thickness resection using an over-the-scope device: a prospective multicentre study in various indications.
      If new tools can accurately select patients with T2 CRC who do not have LNM after EFTR, EFTR without additional surgical resection would be an acceptable treatment option for such patients.
      Figure thumbnail gr1
      Figure 1T2 colorectal cancer without lymph node metastasis (T2N0). A, Endoscopic appearance. The tumor is 14 mm in diameter and located in the ascending colon. B, Specimen after surgical resection (H&E, orig. mag. X1). The diagnosis was moderately differentiated adenocarcinoma invading into the muscularis propria, no lymphovascular invasion, and no lymph node metastasis (retrieved 20 lymph nodes) (stage pT2N0).
      We recently reported a multicenter study on the development and validation of a novel pretherapy prediction tool for LNM in T1 CRC using artificial intelligence (AI), known as the artificial neural network.
      • Kudo S.E.
      • Ichimasa K.
      • Villard B.
      • et al.
      Artificial intelligence system to determine risk of T1 colorectal cancer metastasis to lymph node.
      Here we investigate a novel prediction tool for LNM in T2 CRC, which was simpler and easier to develop, known as the random forest (RF).

      Methods

      Study design

      We included data from all patients with pT2 CRC who underwent surgical resection with lymph node dissection from April 8, 2001 to October 13, 2016 at Showa University Northern Yokohama Hospital. Exclusion criteria were synchronous CRC; transanal endoscopic microsurgery (TEM); familial adenomatous polyposis, Lynch syndrome (definite or suspected without genetic testing for variants in the mismatch repair genes according to the Japanese guidelines for the clinical practice of hereditary colorectal cancer
      • Tomita N.
      • Ishida H.
      • Tanakaya K.
      • et al.
      Japanese Society for Cancer of the Colon and Rectum (JSCCR) guidelines 2020 for the clinical practice of hereditary colorectal cancer.
      ), or ulcerative colitis; preoperative chemotherapy or radiotherapy; and missing data.

      Endpoints

      The primary aim of the study was to develop and validate a simple, noninvasive tool for accurately predicting LNM in patients with T2 CRC using data obtained after endoscopic resection to help triage patients to the least invasive and burdensome yet safe and effective treatment.

      AI tool

      For this purpose, we divided patients into 2 datasets according to the date of treatment following the recommendation of the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis statements.
      • Moons K.G.
      • Altman D.G.
      • Reitsma J.B.
      • et al.
      Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD): explanation and elaboration.
      Data from 411 patients (2001-2014) were used for training the AI system, and the remaining 100 patients (2014-2016) were used to validate the AI system.
      • Moons K.G.
      • Altman D.G.
      • Reitsma J.B.
      • et al.
      Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD): explanation and elaboration.
      We used an RF algorithm for machine learning with a learning algorithm with fast training speed and high model robustness.
      • Breiman L.
      Random forests.
      The RF model applied 220 decision trees, and the maximum leaf node depth was 9. Receiver-operating characteristic curves were used to assess the discriminating power of the tool. Subsequently, its area under the curve (AUC) was compared with the prediction model using a nomogram incorporating 4 factors (lymphatic invasion, vascular invasion, serum carcinoembryonic antigen [CEA], and patient sex) that was constructed based on the results of multivariate logistic regression analysis.

      Prediction variables

      The prediction model used 8 variables: patient age, patient sex, tumor size, tumor location, lymphatic invasion, vascular invasion, histologic differentiation, and CEA level. Clinical factors (patient age and sex, tumor location and size, and serum CEA level [in ng/mL]) were extracted from the electronic health record system. Tumor location was dichotomized as 1 of 2 colonic segments (cecum or ascending to the sigmoid colon) or rectum. Tumor size was measured after formalin fixation. CEA was measured before surgical resection. All resected specimens were examined by dedicated GI pathologists who were not aware of the present study in accordance with Japanese guidelines for histopathologic assessment of CRC.
      Japanese Society for Cancer of the Colon and Rectum
      Japanese classification of colorectal, appendiceal, and anal carcinoma: the 3d English edition.
      ,
      • Hashiguchi Y.
      • Muro K.
      • Saito Y.
      • et al.
      Japanese Society for Cancer of the Colon and Rectum (JSCCR) guidelines 2019 for the treatment of colorectal cancer.

      Statistical analysis

      Our main analyses comprised all eligible patients. In prespecified subgroup analyses, we assessed the AI tool for small T2 CRCs (diameter ≤2 cm) and T2 CRCs in the rectum, because these are technically easier to target for minimally invasive treatment options such as EFTR,
      • Schmidt A.
      • Beyna T.
      • Schumacher B.
      • et al.
      Colonoscopic full-thickness resection using an over-the-scope device: a prospective multicentre study in various indications.
      TEM, and transanal minimally invasive surgery (TAMIS).
      Statistical analyses were performed using R, version 3.4.1 (R Foundation for Statistical Computing, Vienna, Austria). When investigating the predicting performance of the AI system, the sensitivity, specificity, and rate of surgery reduction were calculated. Reduction of surgery was defined as cases with LNM negativity diagnosed by the AI system.
      Data are presented as mean ± standard deviation and median. The 95% confidence intervals (CIs) were calculated using the exact Clopper-Pearson method.

      Study registration and ethics

      This study was registered in the University Hospital Medical Information Network clinical trial registry (no. UMIN 000038257). The ethics committee of Showa University Northern Yokohama Hospital approved the study protocol (no. 18H-006). Informed consent was obtained using an “opt-out” method under the approval of the local ethics committee in which information about the current trial was presented to patients and lack of refusal was regarded as consent to the study. The study was conducted in accordance with the Declaration of Helsinki. All authors had access to the study data, and all authors reviewed and approved the final manuscript.

      Results

      Study cohort

      Figure 2 shows the study flowchart. Five hundred forty patients with pT2 CRC were treated during the study period. Of these, 3 patients with multiple CRCs at the time of resection, 2 patients who underwent TEM, 20 patients with suspected Lynch syndrome, and 4 patients with missing data in the patient files and local registries for variables were excluded. Thus, 511 patients were eligible and included in the analyses. Of these, 411 were assigned to the training cohort and 100 to the validation cohort (Table 1). Rates of LNM in the training and validation datasets were 26% (106/411) and 28% (28/100), respectively. The mean number of removed lymph nodes per patient was 20 (standard deviation, 12), and the median number was 18.
      Figure thumbnail gr2
      Figure 2Patient flowchart. CRC, Colorectal cancer; LNM, lymph node metastasis.
      Table 1Baseline characteristics of the training and validation cohorts for artificial intelligence system
      Training (n = 411)Validation (n = 100)Total (n = 511)
      Age, y66 ± 1267 ± 1266 ± 12
      Sex
       Male216 (53)41 (41)257 (50)
       Female195 (47)59 (59)254 (50)
      Tumor location
       Colon273 (66)72 (72)345 (68)
       Rectum138 (34)28 (28)166 (32)
      Tumor size, mm31 ± 1530 ± 1331 ± 15
      Lymphatic invasion
       Positive216 (53)32 (32)248 (49)
       Negative195 (47)68 (68)263 (51)
      Vascular invasion
       Positive258 (63)59 (59)317 (62)
       Negative153 (37)41 (41)194 (38)
      Histologic differentiation
       Poorly differentiated adenocarcinoma/mucinous carcinoma/signet-ring cell carcinoma16 (4)1 (1)17 (3)
       Tubular differentiated adenocarcinoma/papillary adenocarcinoma395 (96)99 (99)494 (97)
       Carcinoembryonic antigen2.1 ± 1.72.2 ± 2.12.1 ± 1.8
      Lymph node metastasis
       Positive106 (26)28 (28)134 (26)
       Negative305 (74)72 (72)377 (74)
      Values are mean ± standard deviation or n (%).

      Performance of the AI system

      The AUC of the AI system in the validation data was .93, compared with the .88 of the nomogram (Fig. 3). Figure 4 presents the importance of each variable used in the AI system. Lymphatic invasion was the most influential factor for LNM among 8 examined factors. Figure 5 indicates the rate of LNM in high- and low-risk groups according to the threshold (scaled from 1 to 2) of the AI system compared with the postoperative mortality in the United States.
      • Jafari M.D.
      • Jafari F.
      • Halabi W.J.
      • et al.
      Colorectal cancer resections in the aging US population: a trend toward decreasing rates and improved outcomes.
      With 1.18 threshold, sensitivity and specificity were 96% (95% CI, 90%-99%) and 88% (95% CI, 80%-94%), respectively. In this case, 64% of patients (95% CI, 54%-73%) with T2 CRC could avoid surgery, whereas 1.6% of patients (95% CI, 0%-8.4%) with T2 CRC with LNMs would lose a chance to receive surgery. When the threshold was set at 1.19, which was lower than the risk of surgical mortality over age 75 to 79 years, the frequency of the most influential factor, namely lymphatic invasion, was higher in the high-risk group than in the low-risk group (91% vs 2%) (Table 2).
      Figure thumbnail gr3
      Figure 3Receiver operating characteristic curves of the validation cohort (n = 100), tumor size ≤2 cm (n = 27), and rectum (n = 28). AUC, Area under the curve.
      Figure thumbnail gr4
      Figure 4Importance of each variable used in the artificial intelligence system. CEA, Carcinoembryonic antigen.
      Figure thumbnail gr5
      Figure 5Risk of lymph node metastasis in high- and low-risk groups according to the threshold of the artificial intelligence system compared with postoperative mortality.
      • Jafari M.D.
      • Jafari F.
      • Halabi W.J.
      • et al.
      Colorectal cancer resections in the aging US population: a trend toward decreasing rates and improved outcomes.
      LNM, Lymph node metastasis.
      Table 2Comparison of characteristics between high- and low-risk groups categorized by the artificial intelligence system
      High risk (n = 34)Low risk (n = 66)
      Age, y65 ± 1267 ± 12
      Sex
       Male14 (41)27 (41)
       Female20 (59)39 (59)
      Tumor location
       Colon21 (62)51 (77)
       Rectum13 (38)15 (23)
      Tumor size, mm26 ± 1032 ± 14
      Lymphatic invasion
       Positive31 (91)1 (2)
       Negative3 (9)65 (98)
      Vascular invasion
       Positive25 (74)34 (52)
       Negative9 (26)32 (48)
      Histologic differentiation
       Poorly differentiated adenocarcinoma/mucinous carcinoma/signet-ring cell carcinoma0 (0)1 (1)
       Tubular differentiated adenocarcinoma/papillary adenocarcinoma0 (100)99 (99)
       Carcinoembryonic antigen1.9 ± 1.32.4 ± 2.4
      Lymph node metastasis
       Positive26 (76)2 (3)
       Negative8 (24)64 (97)
      Values are mean ± standard deviation or n (%).

      Subgroup analyses

      In the validation dataset, 27 tumors were 2 cm or smaller, and 28 tumors were located in the rectum and thus eligible for the predefined subgroup analyses. Supplementary Table 1 (available online at www.giejournal.org) shows clinicopathologic features in subanalysis. For lesions ≤2 cm, LNM was found in 9 tumors (33%), and the AUCs for this cohort was .86 (Fig. 3). For lesions in the rectum, LNM was found in 9 (32%), and the AUC for this cohort was .86 (Fig. 3).

      Discussion

      The developed AI system predicted the presence of LNM in patients with T2 CRC with high sensitivity and specificity. This may enable the novel “resect and analysis” approach for T2 CRC, in which we determine the need for additional surgery after EFTR of T2 CRC according to the AI-assisted risk stratification and vulnerability of patients. This innovative approach may contribute to reduction of surgery by up to 64% with the minimalized risk of overlooking presence of LNM.
      EFTR is an emerging endoscopic resection technique for complex lesions that cannot be easily treated by current EMR or endoscopic submucosal dissection techniques.
      • Schmidt A.
      • Beyna T.
      • Schumacher B.
      • et al.
      Colonoscopic full-thickness resection using an over-the-scope device: a prospective multicentre study in various indications.
      • Zwager L.W.
      • Bastiaansen B.A.J.
      • Bronzwaer M.E.S.
      • et al.
      Endoscopic full-thickness resection (eFTR) of colorectal lesions: results from the Dutch colorectal eFTR registry.
      • Guillaumot M.A.
      • Barret M.
      • Jacques J.
      • et al.
      Endoscopic full-thickness resection of early colorectal neoplasms using an endoscopic submucosal dissection knife: a retrospective multicenter study.
      ,
      • Kuellmer A.
      • Mueller J.
      • Caca K.
      • et al.
      Endoscopic full-thickness resection for early colorectal cancer.
      ,
      • Meier B.
      • Stritzke B.
      • Kuellmer A.
      • et al.
      Efficacy and safety of endoscopic full-thickness resection in the colorectum: results from the German Colonic FTRD Registry.
      ,
      • Brewer Gutierrez O.I.
      • Akshintala V.S.
      • Ichkhanian Y.
      • et al.
      Endoscopic full-thickness resection using a clip non-exposed method for gastrointestinal tract lesions: a meta-analysis.
      • Shahidi N.
      • Bourke M.J.
      Endoscopic full-thickness resection for invasive colorectal neoplasia: Hype or here to stay?.
      • Vitali F.
      • Naegel A.
      • Siebler J.
      • et al.
      Endoscopic full-thickness resection with an over-the-scope clip device (FTRD) in the colorectum: results from a university tertiary referral center.
      To date, the reported indications for EFTR using the full-thickness resection device (Ovesco Endoscopy, Tuebingen, Germany) include repeat resection (recurrent or incompletely resected), nonlifting lesions caused by fibrosis or deep submucosal invasion, and submucosal tumors. Because of its high R0 resection rate, EFTR is technically applicable for T2 CRC.
      • Liu B.R.
      • Liu D.
      • Ullah S.
      • et al.
      Endoscopic full-thickness resection and endoscopic lymphadenectomy for advanced colonic cancer in an inoperable patient: first human clinical experience.
      In other words, endoscopic removal of T2 CRC is an ambitious but feasible option for future clinical practice if the technology matures. However, lack of an accurate prediction of LNM makes EFTR for T2 CRC less attractive. Preoperative identification of an absence of LNM is the prerequisite for treating T2 CRC with EFTR alone.
      Recently, we developed and validated an algorithm using a neural network, a kind of AI prediction model, to predict the risk of LNM in T1 CRC in a multicenter study.
      • Kudo S.E.
      • Ichimasa K.
      • Villard B.
      • et al.
      Artificial intelligence system to determine risk of T1 colorectal cancer metastasis to lymph node.
      The reported AI system, which outperformed the current guidelines in discriminating LNM positivity and negativity in external validation (AUC of .83 vs .73, P < .001), could determine which patients require additional surgery after endoscopic resection of T1 CRC. Regarding the type of machine learning, we developed another prediction model, namely a nomogram, using the same dataset in addition to the RF model. The AUC of the nomogram in the validation data (n = 100) was .88 versus .93 in the RF model. In addition, compared with the neural network model used in our previous study, we can reduce the calculation time without diminishing the diagnostic performance and visualize the importance of variables used in the RF model. Therefore, we adopted the RF model in this study. Of course, consideration of the number and characteristics of the data when choosing a type of machine learning is necessary.
      The presence of LNM may be overlooked because of the limitations of discriminating power or pathologic evaluation. In fact, 1 of 64 lesions (1.6%) were overlooked in the low-risk group when the threshold was set at 1.18. For LNM assessment after EFTR alone, we performed CT, magnetic resonance imaging, and tumor marker assessments at intervals following the guidelines or shortly after resection and then performed salvage surgery if there was a possibility of metastasis. The curative rate of salvage surgery is not reported to be high, and recurrence increases the likelihood of death.
      • Friel C.M.
      • Cromwell J.W.
      • Marra C.
      • et al.
      Salvage radical surgery after failed local excision for early rectal cancer.
      ,
      • Ikematsu H.
      • Yoda Y.
      • Matsuda T.
      • et al.
      Long-term outcomes after resection for submucosal invasive colorectal cancers.
      Therefore, treatment should be selected after comparing the risk of LNM with the risk of surgical mortality. For example, the postoperative mortality rate was 3.7% in patients aged 75 to 79 years. Conversely, when the AI threshold was set at 1.19, the risk of LNM in the low-risk group, which comprised 66% of the study population, was 3.0%. Therefore, 66% of patients aged 75 to 79 years could have similar risks of recurrence attributable to EFTR alone and surgical mortality. Patients aged ≥80 years may theoretically choose EFTR instead of surgery because the risk of recurrence is lower than the risk of surgical mortality.
      In addition, a sufficient lymph node yield is important for robustly clarifying lymph node status. It has been reported that a minimum of 12 lymph nodes must be examined to accurately predict LNM negativity in CRC.
      • Compton C.C.
      • Greene F.L.
      The staging of colorectal cancer: 2004 and beyond.
      In this validation dataset, the median follow-up period was 55 ± 25 months, and the proportion of patients with <12 harvested lymph nodes was 13% (13/100). The recurrence rate in the LNM-negative group was 4.2% (3/72). Of these, 12 patients had <12 harvested lymph nodes, and all 3 cases with positive recurrence had <12 harvested lymph nodes. Conversely, in the LNM-positive group, the recurrence rate was 10.7% (3/28). Of these, 1 patient had <12 harvested lymph nodes, and all 3 cases with positive recurrence had ≥12 harvested lymph nodes. Therefore, a low lymph node harvest rate could be associated with an underestimation of the lymph node status or an increased recurrence rate in the LNM-negative group.
      For rectal tumors, TAMIS could be a potential option in addition to EFTR. TAMIS was initially described in 2010 by Atallah et al
      • Atallah S.
      • Albert M.
      • Larach S.
      Transanal minimally invasive surgery: a giant leap forward.
      as a hybrid between TEM and single-port laparoscopy. Conventional TEM, as introduced by Buess et al,
      • Buess G.
      • Hutterer F.
      • Theiss J.
      • et al.
      A system for a transanal endoscopic rectum operation [in German].
      is limited by the significant associated initial cost of the operating system and its steep learning curve.
      • Devane L.A.
      • Burke J.P.
      • Kelly J.J.
      • et al.
      Transanal minimally invasive surgery for rectal cancer.
      Conversely, TAMIS, which uses conventional laparoscopic devices with a single incision port, is superior in terms of cost and operability. In addition, TAMIS can obtain a 360-degree view of the rectal lumen and facilitate the resection of lesions >2 cm in size with negative margins, whereas the indication for EFTR is limited by tumor size.
      This study has several limitations. First, the study concept was based on the premise that T2 CRC could be resected en bloc with EFTR. However, currently available EFTR devices allow the removal of tumors up to 2 cm. This will limit generalization of the study results. To overcome this issue, we conducted the subanalyses focused on ≤2-cm tumors and tumors in the rectum and obtained results similar to the primary analysis. However, given recent strong interest in EFTR, we can expect technologic leaps in the near future and may be able to remove all sorts of T2 CRCs with EFTR regardless of their sizes in the near future. Second, this was a retrospective single-center study with internal validation. The clinicopathologic criteria of the respective training and validation datasets were similar, and thus the discriminating power may have been overestimated. In addition, the protocol for histopathologic evaluation differs between Japan and Western countries, and interobserver agreement among pathologists is low.
      • Kojima M.
      • Puppa G.
      • Kirsch R.
      • et al.
      Blood and lymphatic vessel invasion in pT1 colorectal cancer: an international concordance study.
      Therefore, a multicenter study including researchers in Western countries is required as a next step. We believe there is a possibility of using the “transfer learning” method, which permits the efficient adaptation of information learned in the current model to different situations. Third, in this study, tumor budding, which is a risk factor for LNM in T1 CRC, was excluded from the AI system because of a lack of data. Although this AI system only analyzed factors that can be obtained at any hospital, because the addition of tumor budding to the AI system is expected to increase the predictive ability, it is considered a future issue to investigate. Similarly, it is necessary to verify the method of compensating for missing values in the RF model in a prospective or multicenter study.
      In conclusion, this study proposed the resect and analysis approach for T2 CRC after EFTR as a new innovative treatment option. Considering the high prediction capability of the AI system for LNM and the substantial risk of postoperative mortality of CRC, the proposed minimally invasive approach could be an attractive option especially for vulnerable patients for surgical resection.

      Acknowledgment

      We thank Edanz Group (https://en-author-services.edanzgroup.com/ac), for editing a draft of the manuscript.

      Appendix

      Supplementary Table 1Baseline characteristics of lesions with tumor size ≤2 cm (n = 27) and rectum (n = 28) in the validation cohort (n = 100)
      Tumor size ≤2 cm (n = 27)Rectum (n = 28)
      Age, y65 ± 964 ± 11
      Sex
       Male14 (52)13 (46)
       Female13 (48)15 (54)
      Tumor location
       Colon19 (70)0 (0)
       Rectum8 (30)28 (100)
      Tumor size, mm17 ± 431 ± 13
      Lymphatic invasion
       Positive11 (41)11 (39)
       Negative16 (59)17 (61)
      Vascular invasion
       Positive17 (63)19 (68)
       Negative10 (37)9 (32)
      Histologic differentiation
       Poorly differentiated adenocarcinoma/mucinous carcinoma/signet-ring cell carcinoma1 (4)0 (0)
       Tubular differentiated adenocarcinoma/papillary adenocarcinoma26 (96)28 (100)
       Carcinoembryonic antigen1.9 ± 1.62.3 ± 2.4
      Lymph node metastasis
       Positive9 (33)9 (32)
       Negative18 (67)19 (68)
      Values are mean ± standard deviation or n (%).

      References

        • Mori Y.
        • Kudo S.E.
        • Endo S.
        • et al.
        Morphology as a risk factor for the malignant potential of T2 colorectal cancer.
        Mol Clin Oncol. 2016; 5: 223-226
        • Fields A.C.
        • Lu P.
        • Hu F.
        • et al.
        Lymph node positivity in T1/T2 rectal cancer: a word of caution in an era of increased incidence and changing biology for rectal cancer.
        J Gastrointest Surg. 2020; 25: 1029-1035
        • Kajiwara Y.
        • Ueno H.
        • Hashiguchi Y.
        • et al.
        Risk factors of nodal involvement in T2 colorectal cancer.
        Dis Colon Rectum. 2010; 53: 1393-1399
        • Baxter N.N.
        • Garcia-Aguilar J.
        Organ preservation for rectal cancer.
        J Clin Oncol. 2007; 25: 1014-1020
        • Ushigome H.
        • Ohue M.
        • Kitamura M.
        • et al.
        Evaluation of risk factors for lymph node metastasis in T2 lower rectal cancer to perform chemoradiotherapy after local resection.
        Mol Clin Oncol. 2020; 12: 390-394
        • Schmidt A.
        • Beyna T.
        • Schumacher B.
        • et al.
        Colonoscopic full-thickness resection using an over-the-scope device: a prospective multicentre study in various indications.
        Gut. 2018; 67: 1280-1289
        • Zwager L.W.
        • Bastiaansen B.A.J.
        • Bronzwaer M.E.S.
        • et al.
        Endoscopic full-thickness resection (eFTR) of colorectal lesions: results from the Dutch colorectal eFTR registry.
        Endoscopy. 2020; 52: 1014-1023
        • Guillaumot M.A.
        • Barret M.
        • Jacques J.
        • et al.
        Endoscopic full-thickness resection of early colorectal neoplasms using an endoscopic submucosal dissection knife: a retrospective multicenter study.
        Endosc Int Open. 2020; 8: E611-E616
        • Kuellmer A.
        • Behn J.
        • Beyna T.
        • et al.
        Endoscopic full-thickness resection and its treatment alternatives in difficult-to-treat lesions of the lower gastrointestinal tract: a cost-effectiveness analysis.
        BMJ Open Gastroenterol. 2020; : 7:e000449
        • Kuellmer A.
        • Mueller J.
        • Caca K.
        • et al.
        Endoscopic full-thickness resection for early colorectal cancer.
        Gastrointest Endosc. 2019; 89: 1180-1189
        • Meier B.
        • Stritzke B.
        • Kuellmer A.
        • et al.
        Efficacy and safety of endoscopic full-thickness resection in the colorectum: results from the German Colonic FTRD Registry.
        Am J Gastroenterol. 2020; 115: 1998-2006
        • Toyonaga T.
        • Ohara Y.
        • Baba S.
        • et al.
        Peranal endoscopic myectomy (PAEM) for rectal lesions with severe fibrosis and exhibiting the muscle-retracting sign.
        Endoscopy. 2018; 50: 813-817
        • Ono S.
        • Kobayashi R.
        • Ito S.
        • et al.
        A case of successful full-thickness resection using endoscopic submucosal dissection and transanal suturing of rectal cancer.
        Endoscopy. 2019; 51: E301-E302
        • Kudo S.E.
        • Ichimasa K.
        • Villard B.
        • et al.
        Artificial intelligence system to determine risk of T1 colorectal cancer metastasis to lymph node.
        Gastroenterology. 2021; 160: 1075-1084
        • Tomita N.
        • Ishida H.
        • Tanakaya K.
        • et al.
        Japanese Society for Cancer of the Colon and Rectum (JSCCR) guidelines 2020 for the clinical practice of hereditary colorectal cancer.
        Int J Clin Oncol. 2021; 26: 1353-1419
        • Moons K.G.
        • Altman D.G.
        • Reitsma J.B.
        • et al.
        Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD): explanation and elaboration.
        Ann Intern Med. 2015; 162 (W1-73)
        • Breiman L.
        Random forests.
        Machine Learn. 2001; 45: 5-32
        • Japanese Society for Cancer of the Colon and Rectum
        Japanese classification of colorectal, appendiceal, and anal carcinoma: the 3d English edition.
        J Anus Rectum Colon. 2019; 3: 175-195
        • Hashiguchi Y.
        • Muro K.
        • Saito Y.
        • et al.
        Japanese Society for Cancer of the Colon and Rectum (JSCCR) guidelines 2019 for the treatment of colorectal cancer.
        Int J Clin Oncol. 2020; 25: 1-42
        • Jafari M.D.
        • Jafari F.
        • Halabi W.J.
        • et al.
        Colorectal cancer resections in the aging US population: a trend toward decreasing rates and improved outcomes.
        JAMA Surg. 2014; 149: 557-564
        • Brewer Gutierrez O.I.
        • Akshintala V.S.
        • Ichkhanian Y.
        • et al.
        Endoscopic full-thickness resection using a clip non-exposed method for gastrointestinal tract lesions: a meta-analysis.
        Endosc Int Open. 2020; 8: E313-E325
        • Shahidi N.
        • Bourke M.J.
        Endoscopic full-thickness resection for invasive colorectal neoplasia: Hype or here to stay?.
        Gastrointest Endosc. 2019; 89: 1190-1192
        • Vitali F.
        • Naegel A.
        • Siebler J.
        • et al.
        Endoscopic full-thickness resection with an over-the-scope clip device (FTRD) in the colorectum: results from a university tertiary referral center.
        Endosc Int Open. 2018; 6: E98-E103
        • Liu B.R.
        • Liu D.
        • Ullah S.
        • et al.
        Endoscopic full-thickness resection and endoscopic lymphadenectomy for advanced colonic cancer in an inoperable patient: first human clinical experience.
        Gastrointest Endosc. 2020; 91: 451-452
        • Friel C.M.
        • Cromwell J.W.
        • Marra C.
        • et al.
        Salvage radical surgery after failed local excision for early rectal cancer.
        Dis Colon Rectum. 2002; 45: 875-879
        • Ikematsu H.
        • Yoda Y.
        • Matsuda T.
        • et al.
        Long-term outcomes after resection for submucosal invasive colorectal cancers.
        Gastroenterology. 2013; 144 (quiz e14): 551-559
        • Compton C.C.
        • Greene F.L.
        The staging of colorectal cancer: 2004 and beyond.
        CA Cancer J Clin. 2004; 54: 295-308
        • Atallah S.
        • Albert M.
        • Larach S.
        Transanal minimally invasive surgery: a giant leap forward.
        Surg Endosc. 2010; 24: 2200-2205
        • Buess G.
        • Hutterer F.
        • Theiss J.
        • et al.
        A system for a transanal endoscopic rectum operation [in German].
        Chirurg. 1984; 55: 677-680
        • Devane L.A.
        • Burke J.P.
        • Kelly J.J.
        • et al.
        Transanal minimally invasive surgery for rectal cancer.
        Ann Gastroenterol Surg. 2021; 5: 39-45
        • Kojima M.
        • Puppa G.
        • Kirsch R.
        • et al.
        Blood and lymphatic vessel invasion in pT1 colorectal cancer: an international concordance study.
        J Clin Pathol. 2015; 68: 628-632