Characteristics of small early colorectal cancer in the non-pedunculated lesions and its relationship with the depth of invasion Authors: Yu-Bei Li, Wen Song, Wen-Jun Zhao, Hua Liu, Xiao-Yu Li, Xiao-Yan Yin Published: 2026 ABSTRACT BackgroundColorectal cancer (CRC) exhibits high morbidity and mortality globally among malignant tumors. Early diagnosis and treatment can improve prognosis. Previously, tumor size was considered correlated with colorectal cancer invasion. However, whether small CRCs exhibit lower aggressiveness remains uncertain. To investigate the endoscopic features and risk factors for deep submucosal invasion in early, non-pedunculated CRCs with a diameter < 2 cm, and to establish a preliminary nomogram to inform clinical diagnosis and treatment.MethodsWe retrospectively analyzed clinical, endoscopic, and pathological data from 289 patients who underwent endoscopic submucosal dissection (ESD) or surgical resection for non-pedunculated polyps or early-stage colorectal cancer with a diameter < 2 cm at the Affiliated Hospital of Qingdao University between January 2015 and November 2024. Patients were grouped based on the nature and invasion depth of the lesion. Statistical descriptions were performed for each group, and a preliminary nomogram was constructed to evaluate its predictive value for deep submucosal invasion in colorectal lesions.ResultsThe mean age of patients was 63 years, with a higher proportion of males (60.9%). Lesions measuring 10–20 mm were predominant (83.7%). The majority were located in the left colon. Flat lesions were the most common endoscopic appearance (57.8%). The final pathology results showed precancerous lesions, intramucosal cancers, superficial submucosal invasion carcinoma and deep submucosal invasion carcinoma were 40.1%, 38.4%, 8.3% and 13.2% respectively. ESD was directly performed in 95.2% of cases (275 lesions). Surface fullness (OR = 13.81), dilated blood vessels in tumor periphery (OR = 6.44), JNET2B (OR = 6.43), JNET3 (OR = 22.94), and de novo carcinoma pathway (OR = 21.61) were independent risk factors for deep submucosal invasion in early colorectal cancer. Chicken skin mucosa (CSM) (OR = 5.71), surface depression (OR = 12.38), and invasive pit pattern (OR = 161.18) were independent risk factors for malignancy in non-pedunculated colorectal polyps less than 2 cm. The preliminary nomogram constructed based on these independent risk factors demonstrated predictive value for deep submucosal invasion in early-stage colorectal cancer.ConclusionsNon-pedunculated colorectal lesions smaller than 2 cm carry risks of malignancy and deep submucosal invasion. Surface fullness, dilated blood vessels in tumor periphery, type JNET2B/3, and de novo carcinoma pathway are independent risk factors for deep submucosal invasion. CSM, surface depression, and invasive pit pattern are independent risk factors for malignancy. The nomogram prediction model constructed based on these findings provides a basis for clinical decision-making. The preliminary nomogram developed from these findings could offer some guidance for clinical decision-making. Background Colorectal cancer (CRC) ranks as the third most common malignancy worldwide [1]. According to 2022 data from the International Agency for Research on Cancer (IARC), It is estimated that there will be more than 1.9 million new cases of colorectal cancer (including anal cancer) and 904,000 deaths. The morbidity and mortality of CRC are among the highest of all cancers, accounting for 9.6% of total cancer cases and 9.3% of cancer deaths, respectively [2]. Early diagnosis and treatment are crucial for improving the survival rates of CRC. The five-year survival rate for early colorectal cancer (ECC) is 90%, whereas it falls below 15% for advanced CRC. Small colorectal lesions are defined as those with a diameter ≤ 2 cm observed endoscopically, with those ≤ 1 cm considered diminutive colorectal lesions. larger lesions are more likely to harbor advanced histology or deeper invasion; however, small lesions may also exhibit substantial malignant potential. Research has confirmed [3] that small invasive ECCs exhibit malignant potential comparable to large invasive ECCs. A study conducted by Zhang et al [4] included 233 ECCs with diameters less than 2 cm, among which 37 were sub-mucosal invasive colorectal cancer (SM-CRC). Another study [5] included 64 ECCs measuring < 2 cm, of which 25 lesions (39%) showed submucosal invasion. Furthermore, the current understanding of such lesions is insufficient, particularly those with a diameter < 2 cm. Before the pathological tissue evaluation, it is challenging to judge the nature and invasion depth of lesions under conventional white light endoscopy, which may lead to inappropriate treatment. Wan Park et al [5] have reported that traditional endoscopic features cannot distinguish between superficial and deep submucosal invasion in small CRCs. Endoscopic treatment has become the primary approach for early-stage CRC. Pedunculated lesions differ from non-pedunculated lesions in endoscopic management and histopathologic assessment; therefore, this study focused on non-pedunculated lesions. For larger flat lesions, endoscopic submucosal dissection (ESD) is the preferred treatment due to its postoperative pathology facilitating assessment of nature and invasion depth. Multiple therapeutic options exist for small non-pedunculated colorectal lesions, including cold snare polypectomy, EMR, ESD, and surgical resection. However, inappropriate treatment selection still leads to excessive or supplemental surgeries [6], which consequently increases the patient's burden. The primary aim of this study was to identify endoscopic factors associated with deep submucosal invasion (≥ 1000 μm) in early, non-pedunculated colorectal cancers measuring < 2 cm. As secondary exploratory objectives, we also described the clinicopathological and endoscopic differences between benign lesions and ECC, and compared endoscopic characteristics across different invasion-depth categories. We further developed a preliminary nomogram to estimate the risk of deep submucosal invasion. Methods Ethics Ethical approval was obtained from the Institutional Ethics Committee of the Affiliated Hospital of Qingdao University (approval number: QYFYWZLL29671). The requirement for informed consent to participate was waived by the Institutional Ethics Committee of the Affiliated Hospital of Qingdao University. The study adhered to the Declaration of Helsinki. Participants A total of 695 patients who underwent endoscopic submucosal dissection (ESD) or surgical treatment for non-pedunculated polyps with a diameter < 2 cm at the Digestive Endoscopy Center of the Shinan Campus of the Affiliated Hospital of Qingdao University between January 2015 and November 2024 were collected. Ultimately, 289 patients meeting inclusion and exclusion criteria were enrolled. Inclusion criteria ① Non-Pedunculated polyps and ECC treated with ESD and surgical procedures at our hospital. ② Lesion with a diameter < 2 cm. ③ Complete clinical, pathological, and endoscopic data available. Exclusion criteria ① Patients with a confirmed diagnosis of advanced colorectal cancer. ②Patients with familial adenomatous polyposis or hereditary polyposis syndrome. ③Cases with incomplete documentation or inadequate endoscopic imaging. Study design Pathological diagnoses were categorized according to the Modified Vienna classification of gastrointestinal epithelial neoplasia [7]. Low-grade adenoma/dysplasia was classified as Vienna category 3. High-grade adenoma/dysplasia, noninvasive carcinoma, suspicion for invasive carcinoma, and intramucosal carcinoma were classified as Vienna category 4. Submucosal invasive carcinoma was classified as Vienna category 5 (Fig. 1). For the primary analysis of deep submucosal invasion, lesions with Vienna category 4 or 5 pathology were included in the ECC group. The non-deep invasion group included category 4 lesions and superficial submucosal invasive carcinoma, whereas the deep invasion group included lesions with submucosal invasion depth ≥ 1000 μm.Fig. 1White Light Endoscopy features: (a) Surface redness (b) Surface depression (c) Nodules (d)Dilated blood vessels in tumor periphery (e) CSM(f) Surface fullness (g) Fold convergency White Light Endoscopy features: (a) Surface redness (b) Surface depression (c) Nodules (d)Dilated blood vessels in tumor periphery (e) CSM(f) Surface fullness (g) Fold convergency The primary analysis was restricted to ECC lesions and aimed to identify factors associated with deep submucosal invasion, defined as a pathological submucosal invasion depth of ≥ 1000 μm. The comparison between benign lesions and ECC was treated as a secondary exploratory analysis, while the three-category comparison according to invasion depth was retained as a descriptive subgroup analysis. Data collection Clinical data, endoscopic findings, and pathological information for enrolled patients were collected through the His system and endoscopic image management system at the Affiliated Hospital of Qingdao University. Each case was observed using an Olympus endoscope with NBI and magnifying endoscopy. Endoscopic images were independently assessed by three experienced endoscopists, the kappa value for inter-observer agreement among the three endoscopists was 0.85. The resected specimens were assessed based on morphological criteria and immunohistochemical markers, using predefined interpretation thresholds. The findings were independently reviewed by two pathologists, and any disagreements were resolved by referral to a third pathologist. Clinical data The clinical data included sex and age. Endoscopic data White light endoscopy aTumor location, sizeLocation: We classified tumor location as right-sided (including the cecum, ascending colon, transverse colon, or splenic flexure) or left-sided (including the descending colon, sigmoid colon, and rectum) [4].bMacroscopic morphologyBased on the Paris classification [8], tumors were subclassified into three main endoscopic morphologic types, including protruded (0-I), flat (0-II), and depressed (0-III) types [9].cOther White Light Endoscopy features, as shown in Fig. 2. aTumor location, sizeLocation: We classified tumor location as right-sided (including the cecum, ascending colon, transverse colon, or splenic flexure) or left-sided (including the descending colon, sigmoid colon, and rectum) [4]. Tumor location, size Location: We classified tumor location as right-sided (including the cecum, ascending colon, transverse colon, or splenic flexure) or left-sided (including the descending colon, sigmoid colon, and rectum) [4]. bMacroscopic morphologyBased on the Paris classification [8], tumors were subclassified into three main endoscopic morphologic types, including protruded (0-I), flat (0-II), and depressed (0-III) types [9]. Macroscopic morphology Based on the Paris classification [8], tumors were subclassified into three main endoscopic morphologic types, including protruded (0-I), flat (0-II), and depressed (0-III) types [9]. cOther White Light Endoscopy features, as shown in Fig. 2. Other White Light Endoscopy features, as shown in Fig. 2. Fig. 2Modified Vienna classification of gastrointestinal epithelial neoplasia Modified Vienna classification of gastrointestinal epithelial neoplasia Additional White Light Endoscopy features include CSM (pale yellow-speckled mucosa surrounding colon polyps) [4], Surface fullness, fold convergency, surface depression [10], surface redness, dilated blood vessels in tumor periphery, and nodules. ② magnified chromoendoscopy Pit pattern classification [11]: Based on the appearance, structure and staining patterns, lesions are categorized into Pit pattern I, II, III-s, III-L, IV, VI, and VN. Non-invasive pit patterns include Types I-IV, while invasive pit patterns include Types VI and VN [12]. CRCs have four recognized pathways of development: The adenoma–carcinoma sequence (aka chromosomal instability pathway), the de novo carcinoma pathway, the serrated lesion-carcinoma pathway, and the inflammatory-carcinoma pathway [13]; Inflammatory-carcinoma pathway refers to dysplasia or carcinogenesis arising on a background of chronic inflammation, without an adenomatous component [14]; Adenoma-carcinoma sequence is a continuous pathological process in which normal epithelial cells, through the accumulation of genetic mutations, first develop into benign adenomas (dysplasia) and then further evolve into invasive carcinoma [15]; A de novo colorectal cancer (CRC) originates from cancer cells that do not arise from the progression of adenomatous changes, and the exact mechanism underlying its development remains unclear [16]. The classification is primarily based on the pit pattern and vascular morphology observed under magnifying endoscopy (ME) combined with narrow-band imaging (NBI) or chromoendoscopy (using indigo carmine), as well as the macroscopic morphology of the lesion and the characteristics of the surrounding background mucosa. The endoscopic features of the adenoma-carcinoma sequence include: type IIIL or IV pit pattern (Kudo classification), network or dense vascular pattern (Sano classification/JNET type 2 A), and absence of active inflammation in the surrounding mucosa [17–19]. The endoscopic features of the inflammatory-carcinoma pathway include: type I/II pit pattern accompanied by redness or depression, or the presence of pine-cone/villi patterns; NBI findings of irregular, interrupted, or absent vessels; and the absence of an adenomatous component [20]. Normal I type around tumor and lack of III or IV type pits is a hallmark of de novo CRC under endoscopy [21–23]. The four pathways are determined through endoscopic observation. NBI magnifying endoscopy The JNET classification, established by the Japan National Cancer Center Research and Development Fund (Yutaka Saito Group), is used to classify colorectal tumors under NBI magnifying endoscopy [19]. The classification includes Type 1, Type 2 A, Type 2B, and Type 3. Pathological data ESD specimen assessment includes horizontal margins, vertical margins, lymphatic invasion, vascular invasion, and tumor budding [24]; Statistical analysis Statistical analysis was performed using SPSS 25.0 and R 4.1.3. Continuous variables are expressed as mean ± SD or median (interquartile range); categorical variables are expressed as percentages or rates (%). Univariate analysis employed Fisher’s exact test, the χ2 test, and Student’s t-test, and Mann–Whitney U tests. Multivariate analysis was performed using stepwise regression. A preliminary nomogram was constructed based on the aforementioned independent risk factors. The predictive capability of this model was evaluated using the area under the ROC curve (AUC) after plotting the ROC curve. A calibration curve was generated to evaluate the accuracy and discrimination of the nomogram. The DCA curve was used to evaluate the clinical utility of the prediction model. Statistical significance was considered at P < 0.05. Results Patient characteristics Among the 289 patients included in the study (Fig. 3), the median age was 63 years, with a higher proportion of males (n = 176, 60.9%). Most lesions measured between 10–20 mm in diameter (n = 242, 83.7%) and the majority were located in the left colon and rectum (n = 83, 28.7%, n = 100, 34.6%). Regarding macroscopic morphology, 167 cases (57.8%) were flat, 99 cases (34.3%) were protruded, and 23 cases (8.0%) were depressed. A total of 173 cases (59.9%) were ECC, including 38 cases (22.0%) with deep submucosal invasion. Most lesions followed the adenoma–carcinoma sequence (n = 215, 74.4%), while a smaller proportion followed the de novo carcinoma pathway (n = 21, 7.3%). Patient data are summarized in Table 1.Fig. 3Flowchart of study patientsTable 1Patient's basic informationTotal n = 289(%)Age63 (55,69)Sex Male176 (60.9) Female113 (39.1) Size ≤ 10 mm47 (16.3) 10–20 mm242 (83.7)Location Left colon83 (28.7) Right colon106 (36.7) Rectum100 (34.6)Morphology Protruded type99 (34.3) Flat type167 (57.8) Depressed type23 (8.0)Pathway of development Adenoma–carcinoma sequence215 (74.4) Inflammatory-carcinoma pathway53 (18.3) De-novo carcinoma pathway21 (7.3)Pathology Vienna category 3: low-grade adenoma/dysplasia116 (40.1) Vienna category 4: high-grade neoplasia/intramucosal carcinoma111(38.4) Vienna category 5: submucosal invasive carcinoma62(21.5)Depth of invasion Non-submucosal invasive carcinoma111 (64.1) Superficial submucosal invasive carcinoma24 (13.9) Deep submucosal invasive mucosal carcinoma38 (22.0) Lymphatic invasion8 (2.8%) Vascular invasion10 (3.5%) Tumor budding2 (0.7%)Treatment ESD263 (91.0) surgical resection14 (4.8) ESD + surgical resection12 (4.2) Flowchart of study patients Table 1Patient's basic informationTotal n = 289(%)Age63 (55,69)Sex Male176 (60.9) Female113 (39.1) Size ≤ 10 mm47 (16.3) 10–20 mm242 (83.7)Location Left colon83 (28.7) Right colon106 (36.7) Rectum100 (34.6)Morphology Protruded type99 (34.3) Flat type167 (57.8) Depressed type23 (8.0)Pathway of development Adenoma–carcinoma sequence215 (74.4) Inflammatory-carcinoma pathway53 (18.3) De-novo carcinoma pathway21 (7.3)Pathology Vienna category 3: low-grade adenoma/dysplasia116 (40.1) Vienna category 4: high-grade neoplasia/intramucosal carcinoma111(38.4) Vienna category 5: submucosal invasive carcinoma62(21.5)Depth of invasion Non-submucosal invasive carcinoma111 (64.1) Superficial submucosal invasive carcinoma24 (13.9) Deep submucosal invasive mucosal carcinoma38 (22.0) Lymphatic invasion8 (2.8%) Vascular invasion10 (3.5%) Tumor budding2 (0.7%)Treatment ESD263 (91.0) surgical resection14 (4.8) ESD + surgical resection12 (4.2) Patient's basic information Analysis of risk factors for submucosal deep invasion in non-pedunculated ECC less than 2 cm Among the characteristics observed with conventional white light endoscopy, protruded and depressed type, CSM, surface fullness, fold convergency, surface redness, and dilated peripheral blood vessels were more prevalent in the deep submucosal invasive carcinoma group compared to the superficial submucosal invasive carcinoma group (P < 0.05); JNET classification and malignant transformation pathways showed significant differences between the non-deep and deep invasion groups (P < 0.001), with a higher proportion of de novo carcinoma pathway in the deep invasion group; as shown in Table 2. Multivariate logistic regression analysis revealed that surface fullness (OR = 13.81, CI: 3.92–48.63, P < 0.001), dilated blood vessels in tumor periphery (OR = 6.44, CI: 1.92–21.60, P = 0.003), JNET2B type (OR = 6.43, CI: 1.63–25.41, P = 0.008), JNET3 type (OR = 22.94, CI: 1.97–266.61, P = 0.012), and de novo carcinoma pathway (OR = 21.61, CI: 2.81–165.99, P = 0.003) were significantly associated with the depth of invasion; as shown in Table 3.Table 2Analysis of risk factors for submucosal deep invasion in non-pedunculated ECC less than 2 cmVariablesnon-deep invasion groupn = 135(%)deep invasion groupn = 38(%)χ2或fisherPAge, Mean ± SD62.12 ± 9.6660.55 ± 9.88t = 0.880.381Sex0.020.878 Male87 (64.44)25 (65.79) Female48 (35.56)13 (34.21)Sized2.740.098 ≤ 1 cm20 (14.81)10 (26.32)> 1 cm115 (85.19)28 (73.68)Location5.440.066 Right colon39 (28.89)4 (10.53) Left colon37 (27.41)14 (36.84) Rectum59(43.70)20(52.63)Morphology13.220.001* Flat73 (54.07)8 (21.05) protruded48 (35.56)22 (57.89) depressed type14 (10.37)8 (21.05)CSM7.540.006* No90 (66.67)16 (42.11) Yes45 (33.33)22 (57.89)Surface fullness22.33 <.001* No116 (85.93)19 (50.00) Yes19 (14.07)19 (50.00)Fold convergency13.64 <.001*No134 (99.26)32 (84.21)Yes1 (0.74)6 (15.79)Surface redness11.79 <.001* No63 (46.67)6 (15.79) Yes72 (53.33)32 (84.21)Dilated blood vessels in tumor periphery19.95 <.001*No87 (64.44)9 (23.68)Yes48 (35.56)29 (76.32)Nodules0.130.720 No86 (63.70)23 (60.53) Yes49 (36.30)15 (39.47)Surface depression2.860.091 No52 (38.52)9 (23.68) Yes83 (61.48)29 (76.32)Pit pattern4.900.083 Non-invasive24 (17.78)2 (5.26) Type VI110 (81.48)35 (92.11) Type VN1 (0.74)1 (2.63)JNET35.56 <.001* JNET2A80 (59.26)3 (7.89) JNET2B53 (39.26)33 (86.84) 32 (1.48)2 (5.26)Pathway of development35.00 <.001* Inflammatory-carcinoma pathway23 (17.04)2 (5.26) Adenoma-carcinoma sequence106 (78.52)21 (55.26) De novo carcinoma pathway6 (4.44)15 (39.47)*P < 0.05Table 3Simple and multivariate analysis of deep submucosal invasionVariablesSimple analysisMultivariate analysisβS.EZPOR (95%CI)βS.EZPOR (95%CI)Morphology Flat1.00 (Reference) Protruded1.430.453.160.002*4.18 (1.72 ~ 10.16) Depressed1.650.582.850.004*5.21(1.68 ~ 16.22)CSM No1.00 (Reference) Yes1.010.382.690.007*2.75 (1.32 ~ 5.74)Surface fullness No1.00 (Reference)1.00 (Reference) Yes1.810.414.43 <.001*6.11 (2.74 ~ 13.58)2.630.644.09 <.001*13.81 (3.92 ~ 48.63)Fold convergency No1.00 (Reference) Yes3.221.102.940.003*25.12 (2.92 ~ 216.09)Surface redness No1.00 (Reference) Yes1.540.483.230.001*4.67 (1.83 ~ 11.89)Dilated blood vessels in tumor periphery No1.00 (Reference)1.00 (Reference) Yes1.760.424.18 <.001*5.84 (2.56 ~ 13.35)1.860.623.010.003*6.44 (1.92 ~ 21.60)JNET JNET2A1.00 (Reference)1.00 (Reference) JNET2B2.810.634.47 <.001*16.60 (4.84 ~ 56.91)1.860.702.650.008*6.43 (1.63 ~ 25.41) 33.281.162.830.005*26.67 (2.74 ~ 259.09)3.131.252.500.012*22.935 (1.97 ~ 266.61)Pathway of development Inflammatory-carcinoma pathway1.00 (Reference)1.00 (Reference) Adenoma-carcinoma sequence0.820.771.060.2882.28 (0.50 ~ 10.40)0.470.950.490.6221.60 (0.25 ~ 10.22) De novo carcinoma pathway3.360.883.81 <.001*28.75 (5.11 ~ 161.76)3.071.042.960.003*21.61 (2.81 ~ 165.99)*P < 0.05 Table 2Analysis of risk factors for submucosal deep invasion in non-pedunculated ECC less than 2 cmVariablesnon-deep invasion groupn = 135(%)deep invasion groupn = 38(%)χ2或fisherPAge, Mean ± SD62.12 ± 9.6660.55 ± 9.88t = 0.880.381Sex0.020.878 Male87 (64.44)25 (65.79) Female48 (35.56)13 (34.21)Sized2.740.098 ≤ 1 cm20 (14.81)10 (26.32)> 1 cm115 (85.19)28 (73.68)Location5.440.066 Right colon39 (28.89)4 (10.53) Left colon37 (27.41)14 (36.84) Rectum59(43.70)20(52.63)Morphology13.220.001* Flat73 (54.07)8 (21.05) protruded48 (35.56)22 (57.89) depressed type14 (10.37)8 (21.05)CSM7.540.006* No90 (66.67)16 (42.11) Yes45 (33.33)22 (57.89)Surface fullness22.33 <.001* No116 (85.93)19 (50.00) Yes19 (14.07)19 (50.00)Fold convergency13.64 <.001*No134 (99.26)32 (84.21)Yes1 (0.74)6 (15.79)Surface redness11.79 <.001* No63 (46.67)6 (15.79) Yes72 (53.33)32 (84.21)Dilated blood vessels in tumor periphery19.95 <.001*No87 (64.44)9 (23.68)Yes48 (35.56)29 (76.32)Nodules0.130.720 No86 (63.70)23 (60.53) Yes49 (36.30)15 (39.47)Surface depression2.860.091 No52 (38.52)9 (23.68) Yes83 (61.48)29 (76.32)Pit pattern4.900.083 Non-invasive24 (17.78)2 (5.26) Type VI110 (81.48)35 (92.11) Type VN1 (0.74)1 (2.63)JNET35.56 <.001* JNET2A80 (59.26)3 (7.89) JNET2B53 (39.26)33 (86.84) 32 (1.48)2 (5.26)Pathway of development35.00 <.001* Inflammatory-carcinoma pathway23 (17.04)2 (5.26) Adenoma-carcinoma sequence106 (78.52)21 (55.26) De novo carcinoma pathway6 (4.44)15 (39.47)*P < 0.05 Analysis of risk factors for submucosal deep invasion in non-pedunculated ECC less than 2 cm *P < 0.05 Table 3Simple and multivariate analysis of deep submucosal invasionVariablesSimple analysisMultivariate analysisβS.EZPOR (95%CI)βS.EZPOR (95%CI)Morphology Flat1.00 (Reference) Protruded1.430.453.160.002*4.18 (1.72 ~ 10.16) Depressed1.650.582.850.004*5.21(1.68 ~ 16.22)CSM No1.00 (Reference) Yes1.010.382.690.007*2.75 (1.32 ~ 5.74)Surface fullness No1.00 (Reference)1.00 (Reference) Yes1.810.414.43 <.001*6.11 (2.74 ~ 13.58)2.630.644.09 <.001*13.81 (3.92 ~ 48.63)Fold convergency No1.00 (Reference) Yes3.221.102.940.003*25.12 (2.92 ~ 216.09)Surface redness No1.00 (Reference) Yes1.540.483.230.001*4.67 (1.83 ~ 11.89)Dilated blood vessels in tumor periphery No1.00 (Reference)1.00 (Reference) Yes1.760.424.18 <.001*5.84 (2.56 ~ 13.35)1.860.623.010.003*6.44 (1.92 ~ 21.60)JNET JNET2A1.00 (Reference)1.00 (Reference) JNET2B2.810.634.47 <.001*16.60 (4.84 ~ 56.91)1.860.702.650.008*6.43 (1.63 ~ 25.41) 33.281.162.830.005*26.67 (2.74 ~ 259.09)3.131.252.500.012*22.935 (1.97 ~ 266.61)Pathway of development Inflammatory-carcinoma pathway1.00 (Reference)1.00 (Reference) Adenoma-carcinoma sequence0.820.771.060.2882.28 (0.50 ~ 10.40)0.470.950.490.6221.60 (0.25 ~ 10.22) De novo carcinoma pathway3.360.883.81 <.001*28.75 (5.11 ~ 161.76)3.071.042.960.003*21.61 (2.81 ~ 165.99)*P < 0.05 Simple and multivariate analysis of deep submucosal invasion *P < 0.05 Analysis of characteristics of early colorectal cancer less than 2 CM in non-pedunculated lesions We found that lesions located in the rectum, CSM, surface fullness, surface redness, dilated blood vessels in tumor periphery, surface depression, flat type, depressed type, JNET2B type, and invasive pit patterns are risk factors for malignant transformation of colorectal lesions (P < 0.05) (Table 4); Multivariate logistic regression analysis revealed that CSM (OR = 5.71, CI: 1.46–22.39, P = 0.012), surface depression (OR = 12.38, CI: 3.76–40.74, P < 0.001), and invasive pit pattern(OR = 161.18, CI: 31.38–827.71, P < 0.001) were significantly associated with malignant transformation of lesions (Table 5).Table 4Analysis of characteristics of early colorectal cancer less than 2 CM in non-pedunculated lesionsVariablesPrecancerouslesion groupn = 116(%)ECC groupn = 173(%)χ2或fisherPAge, M (Q₁, Q₃)63.00 (56.00, 68.00)63.00 (55.00, 69.00)0.928Sex2.670.102 Male64 (55.17)112 (64.74) Female52 (44.83)61 (35.26)Size0.370.544 ≤ 1 cm17 (14.66)30 (17.34) ≤ 2 cm99 (85.34)143 (82.66)Location31.76 <.001* Right colon63 (54.31)43 (24.86) Left colon32 (27.59)51 (29.48) Rectum21(18.10)79(45.66)CSM38.96 <.001* No109 (93.97)106 (61.27) Yes7 (6.03)67 (38.73)Surface fullness26.49 <.001* No115 (99.14)135 (78.03) Yes1 (0.86)38 (21.97)Fold convergency1.570.211 No115 (99.14)166 (95.95) Yes1 (0.86)7 (4.05)Surface redness47.82 <.001* No94 (81.03)69 (39.88) Yes22 (18.97)104 (60.12)Nodules0.800.370 No67 (57.76)109 (63.01) Yes49 (42.24)64 (36.99)Dilated blood vessels in tumor periphery52.49 <.001* No110 (94.83)96 (55.49) Yes6 (5.17)77 (44.51)Surface depression86.73 <.001* No105 (90.52)61 (35.26) Yes11 (9.48)112 (64.74)Morphology26.08 <.001* Flat86 (74.14)81 (46.82) Protruded29 (25.00)70 (40.46) Depressed type1 (0.86)22 (12.72)Pit pattern184.97 <.001* Non-invasive112 (96.55)26 (15.03) Invasive4 (3.45)147 (84.97)JNET80.11 <.001* JNET2A110 (94.83)83 (47.98) JNET2B5 (4.31)86 (49.71) JNET31 (0.86)4 (2.31)*P < 0.05Table 5Simple and multivariate analysis of malignant transformation of colorectal lesionsVariablesSimple analysisMultivariate analysisβS.EZPOR (95%CI)βS.EZPOR (95%CI)Location Right colon1.00 (Reference)1.00 (Reference) Left colon0.850.302.830.005*2.34 (1.30 ~ 4.20)0.350.600.570.5651.41 (0.44 ~ 4.58) Rectum1.710.325.41 <.001*5.51 (2.97 ~ 10.22)1.070.661.630.1022.92 (0.81 ~ 10.53)CSM No1.00 (Reference)1.00 (Reference) Yes2.290.425.45 <.001*9.84 (4.32 ~ 22.42)1.740.702.500.012*5.71 (1.46 ~ 22.39)Surface fullness No1.00 (Reference)1.00 (Reference) Yes3.481.023.41 <.001*32.37 (4.38 ~ 239.45)2.531.391.820.06912.51 (0.82 ~ 190.69)Surface redness No1.00 (Reference)1.00 (Reference) Yes1.860.286.58 <.001*6.44 (3.70 ~ 11.22)−0.490.66−0.740.4600.61 (0.17 ~ 2.25)Dilated blood vessels in tumor periphery No1.00 (Reference)1.00 (Reference) Yes2.690.456.02 <.001*14.70 (6.13 ~ 35.26)−0.330.84−0.400.6920.72 (0.14 ~ 3.71)Surface Depression No1.00 (Reference)1.00 (Reference) Yes2.860.358.08 <.001*17.53 (8.75 ~ 35.12)2.520.614.14 <.001*12.38 (3.76 ~ 40.74)Morphology Flat1.00 (Reference)1.00 (Reference) Protruded0.940.273.49 <.0012.56 (1.51 ~ 4.35)−0.590.69−0.860.3910.55 (0.14 ~ 2.14) Depressed type3.151.033.050.00223.36 (3.08 ~ 177.24)1.951.391.410.1607.02 (0.46 ~ 106.49)Pit pattern Non-invasive1.00 (Reference)1.00 (Reference) Invasive5.060.559.18 <.001158.31 (53.71 ~ 466.63)5.080.836.09 <.001*161.18 (31.38 ~ 827.71)JNET JNET2A1.00 (Reference)1.00 (Reference) JNET2B3.130.486.48 <.00122.80 (8.85 ~ 58.68)−0.820.90−0.910.3630.44 (0.08 ~ 2.57) 31.671.131.480.1395.30 (0.58 ~ 48.31)−1.682.09−0.800.4230.19 (0.00 ~ 11.33) Table 4Analysis of characteristics of early colorectal cancer less than 2 CM in non-pedunculated lesionsVariablesPrecancerouslesion groupn = 116(%)ECC groupn = 173(%)χ2或fisherPAge, M (Q₁, Q₃)63.00 (56.00, 68.00)63.00 (55.00, 69.00)0.928Sex2.670.102 Male64 (55.17)112 (64.74) Female52 (44.83)61 (35.26)Size0.370.544 ≤ 1 cm17 (14.66)30 (17.34) ≤ 2 cm99 (85.34)143 (82.66)Location31.76 <.001* Right colon63 (54.31)43 (24.86) Left colon32 (27.59)51 (29.48) Rectum21(18.10)79(45.66)CSM38.96 <.001* No109 (93.97)106 (61.27) Yes7 (6.03)67 (38.73)Surface fullness26.49 <.001* No115 (99.14)135 (78.03) Yes1 (0.86)38 (21.97)Fold convergency1.570.211 No115 (99.14)166 (95.95) Yes1 (0.86)7 (4.05)Surface redness47.82 <.001* No94 (81.03)69 (39.88) Yes22 (18.97)104 (60.12)Nodules0.800.370 No67 (57.76)109 (63.01) Yes49 (42.24)64 (36.99)Dilated blood vessels in tumor periphery52.49 <.001* No110 (94.83)96 (55.49) Yes6 (5.17)77 (44.51)Surface depression86.73 <.001* No105 (90.52)61 (35.26) Yes11 (9.48)112 (64.74)Morphology26.08 <.001* Flat86 (74.14)81 (46.82) Protruded29 (25.00)70 (40.46) Depressed type1 (0.86)22 (12.72)Pit pattern184.97 <.001* Non-invasive112 (96.55)26 (15.03) Invasive4 (3.45)147 (84.97)JNET80.11 <.001* JNET2A110 (94.83)83 (47.98) JNET2B5 (4.31)86 (49.71) JNET31 (0.86)4 (2.31)*P < 0.05 Analysis of characteristics of early colorectal cancer less than 2 CM in non-pedunculated lesions *P < 0.05 Table 5Simple and multivariate analysis of malignant transformation of colorectal lesionsVariablesSimple analysisMultivariate analysisβS.EZPOR (95%CI)βS.EZPOR (95%CI)Location Right colon1.00 (Reference)1.00 (Reference) Left colon0.850.302.830.005*2.34 (1.30 ~ 4.20)0.350.600.570.5651.41 (0.44 ~ 4.58) Rectum1.710.325.41 <.001*5.51 (2.97 ~ 10.22)1.070.661.630.1022.92 (0.81 ~ 10.53)CSM No1.00 (Reference)1.00 (Reference) Yes2.290.425.45 <.001*9.84 (4.32 ~ 22.42)1.740.702.500.012*5.71 (1.46 ~ 22.39)Surface fullness No1.00 (Reference)1.00 (Reference) Yes3.481.023.41 <.001*32.37 (4.38 ~ 239.45)2.531.391.820.06912.51 (0.82 ~ 190.69)Surface redness No1.00 (Reference)1.00 (Reference) Yes1.860.286.58 <.001*6.44 (3.70 ~ 11.22)−0.490.66−0.740.4600.61 (0.17 ~ 2.25)Dilated blood vessels in tumor periphery No1.00 (Reference)1.00 (Reference) Yes2.690.456.02 <.001*14.70 (6.13 ~ 35.26)−0.330.84−0.400.6920.72 (0.14 ~ 3.71)Surface Depression No1.00 (Reference)1.00 (Reference) Yes2.860.358.08 <.001*17.53 (8.75 ~ 35.12)2.520.614.14 <.001*12.38 (3.76 ~ 40.74)Morphology Flat1.00 (Reference)1.00 (Reference) Protruded0.940.273.49 <.0012.56 (1.51 ~ 4.35)−0.590.69−0.860.3910.55 (0.14 ~ 2.14) Depressed type3.151.033.050.00223.36 (3.08 ~ 177.24)1.951.391.410.1607.02 (0.46 ~ 106.49)Pit pattern Non-invasive1.00 (Reference)1.00 (Reference) Invasive5.060.559.18 <.001158.31 (53.71 ~ 466.63)5.080.836.09 <.001*161.18 (31.38 ~ 827.71)JNET JNET2A1.00 (Reference)1.00 (Reference) JNET2B3.130.486.48 <.00122.80 (8.85 ~ 58.68)−0.820.90−0.910.3630.44 (0.08 ~ 2.57) 31.671.131.480.1395.30 (0.58 ~ 48.31)−1.682.09−0.800.4230.19 (0.00 ~ 11.33) Simple and multivariate analysis of malignant transformation of colorectal lesions Endoscopic and pathological characteristics of ECC with different depths of invasion: a descriptive subgroup analysis Intergroup comparative analysis revealed that among the non-submucosal invasive carcinoma group, superficial submucosal invasive carcinoma group, and the deep submucosal invasive carcinoma group, the prevalence of CSM (31.5% vs 41.7% vs 57.9%), surface redness (53.2% vs 54.2% vs 84.2%), and surface fullness (9.9% vs 29.2% vs 50%) gradually increased with advancing invasion depth. Lesions in all three groups were predominantly characterized by surface depression and mostly lacked fold convergency. The deep submucosal invasive group had the highest incidence of fold convergency at 15.8%. dilated blood vessels in tumor periphery were significantly more prevalent in the non-submucosal invasive carcinoma group (36.9%) and deep submucosal invasion group (76.3%) compared to the superficial submucosal invasion group (29.2%). All these endoscopic features showed statistically significant differences among the three groups (p < 0.001). Under magnifying endoscopy, the predominant type of the pit patterns in all three groups was Type VI and its prevalence gradually increased with deeper invasion: 80.2%, 87.5%, and 92.1%, respectively. The non-submucosal invasive carcinoma group predominantly exhibited JNET2A type (66.8%), while the superficial submucosal invasion group and deep submucosal invasion group predominantly showed JNET2B type, accounting for 70.8% and 71.1%, respectively. The deep submucosal invasion group also included a significant proportion of JNET 3 type (21%). Regarding the pathway of malignant transformation, the proportion of de novo carcinoma pathway progressively increased with greater depth of invasion (p < 0.001), as shown in Table 6.Table 6Endoscopic and pathological characteristics of ECC with different depths of invasionEndoscopic Featuresnon-submucosal invasive carcinoma group n = 111(%)superficial submucosal invasive carcinoma group n = 24(%)deep submucosal invasive carcinoma group n = 38(%)PCSM0.015* Yes n = 6735(52.2%)10(15.0%)22(32.8%) No n = 10676(71.7%)14(13.2%)16(15.1%)Surface fullness < 0.001* Yes n = 3711(29.7%)7(18.9%)19(51.4%) No n = 136100(73.5%)17(12.5%)19(14.0%)Fold convergency < 0.001* Yes n = 70(0.0%)1(14.3%)6(85.7%) No n = 166111(66.8%)23(13.9%)32(19.3%)Surface redness0.003* Yes n = 10459(56.7%)13(12.5%)32(30.8%) No n = 6952(75.4%)11(15.9%)6(8.7%)Nodules0.888 Yes n = 6341(65.1%)7(11.1%)15(23.8%) No n = 11070(63.6%)17(15.5%)23(20.9%)Surface depression0.032* Yes n = 11264(57.1%)19(17.0%)29(25.9%) No n = 6147(77.0%)5(8.2%)9(14.8%)Dilated blood vessels in tumor periphery < 0.001* Yes n = 7741(53.2%)7(9.1%)29(37.7%) No n = 9670(72.9%)17(17.7%)9(9.4%)Pit pattern0.027* Non-invasive n = 2622(84.6%)2(7.7%)2(7.7%) Type VI n = 14589(61.4%)21(14.5%)35(24.1%) Type VN n = 20(0.0%)1(50.0%)1(50.0%)JNET < 0.001* JNET2A n = 8373(88.0%)7(8.4%)3(3.6%) JNET2B n = 8238(46.3%)17(20.7%)27(33.0%) JNET3 n = 80(0.0%)0(0.0%)8(100.0%)Pathway of development < 0.001* Adenoma-carcinoma sequence n = 12786(67.7%)20(15.7%)21(16.6%) Inflammatory-carcinoma pathway n = 2522(88.0%)1(4.0%)2(8.0%) De novo carcinoma pathway n = 213(14.3%)3(14.3%)15(71.4%)*P < 0.05 Table 6Endoscopic and pathological characteristics of ECC with different depths of invasionEndoscopic Featuresnon-submucosal invasive carcinoma group n = 111(%)superficial submucosal invasive carcinoma group n = 24(%)deep submucosal invasive carcinoma group n = 38(%)PCSM0.015* Yes n = 6735(52.2%)10(15.0%)22(32.8%) No n = 10676(71.7%)14(13.2%)16(15.1%)Surface fullness < 0.001* Yes n = 3711(29.7%)7(18.9%)19(51.4%) No n = 136100(73.5%)17(12.5%)19(14.0%)Fold convergency < 0.001* Yes n = 70(0.0%)1(14.3%)6(85.7%) No n = 166111(66.8%)23(13.9%)32(19.3%)Surface redness0.003* Yes n = 10459(56.7%)13(12.5%)32(30.8%) No n = 6952(75.4%)11(15.9%)6(8.7%)Nodules0.888 Yes n = 6341(65.1%)7(11.1%)15(23.8%) No n = 11070(63.6%)17(15.5%)23(20.9%)Surface depression0.032* Yes n = 11264(57.1%)19(17.0%)29(25.9%) No n = 6147(77.0%)5(8.2%)9(14.8%)Dilated blood vessels in tumor periphery < 0.001* Yes n = 7741(53.2%)7(9.1%)29(37.7%) No n = 9670(72.9%)17(17.7%)9(9.4%)Pit pattern0.027* Non-invasive n = 2622(84.6%)2(7.7%)2(7.7%) Type VI n = 14589(61.4%)21(14.5%)35(24.1%) Type VN n = 20(0.0%)1(50.0%)1(50.0%)JNET < 0.001* JNET2A n = 8373(88.0%)7(8.4%)3(3.6%) JNET2B n = 8238(46.3%)17(20.7%)27(33.0%) JNET3 n = 80(0.0%)0(0.0%)8(100.0%)Pathway of development < 0.001* Adenoma-carcinoma sequence n = 12786(67.7%)20(15.7%)21(16.6%) Inflammatory-carcinoma pathway n = 2522(88.0%)1(4.0%)2(8.0%) De novo carcinoma pathway n = 213(14.3%)3(14.3%)15(71.4%)*P < 0.05 Endoscopic and pathological characteristics of ECC with different depths of invasion *P < 0.05 Development and validation of a preliminary nomogram for predicting deep submucosal invasion in non-pedunculated ECC less than 2 cm Based on the multivariate logistic regression analysis, we confirmed that surface fullness, dilated blood vessels in tumor periphery, JNET 2B type, JNET 3 type, and the de novo carcinoma pathway are independent risk factors for increased invasion depth. A preliminary nomogram was constructed (Fig. 4). The ROC curve for the training set was generated using all data, and the ROC curve for the test set was generated using cross-validation. The areas under two ROC curves (AUC) were 0.916 (95% CI: 0.87–0.96) and 0.881 (95% CI: 0.82–0.95) (Fig. 5). The generated calibration curve demonstrated consistency and goodness-of-fit (Fig. 6). Additionally, the Hosmer-Leme show goodness-of-fit test yielded a P-value of 0.058 (> 0.05). Furthermore, we constructed a DCA curve and found that the prediction model has certain clinical utility (Fig. 7). Based on the limited existing research findings, the model constructed in this study demonstrates reliability in predicting the depth of invasion in ECC. However, In the future, standardized criteria and larger sample sizes may be necessary to confirm the long-term reliability of the model.Fig. 4Prediction Model for Deep Submucosal Invasion in ECC: A NomogramFig. 5ROC curveFig. 6Calibration curveFig. 7DCA curve Prediction Model for Deep Submucosal Invasion in ECC: A Nomogram ROC curve Calibration curve DCA curve Discussion Individual endoscopic features cannot be used to predict whether a lesion is malignant or to estimate its depth of invasion. Instead, a comprehensive assessment or additional features is required. In our study, we summarized the factors affecting the malignant degree of lesions from previous studies. Among 47 cases measuring less than 1 cm in diameter, 30 cases were confirmed as cancerous, with 10 cases showing deep submucosal invasion. Previous studies [4] indicate that the most common sites of both diminutive and small were the left colon, inconsistent with our findings. This may be because their study classified the rectum as part of the left colon. This may also suggest that endoscopists should pay more attention to lesions located in the left colon and rectum during endoscopic examinations. We found that deep submucosal invasive CRC predominantly showed a protruded morphology (p = 0.001). However, this cannot be ruled out as being due to the lack of more detailed stratification and comparison in this study, where some flat-depressed lesions (II c) were classified as flat lesions. And we found protruded and flat lesions were equally prevalent in the ECC group in our study. Superficial submucosal invasive CRC predominantly exhibited a flat morphology. We found that lesions with CSM were associated with carcinogenesis (OR = 5.66, CI: 1.44–22.11, P = 0.013), the endoscopic identification of CSM should raise suspicion for this risk. In our study, surface depression can be used to some extent to predict whether a lesion is malignant, whereas fold convergency, surface fullness and surface nodules hold little value in predicting malignancy in colorectal lesions. We investigated the relationship between these endoscopic features and the depth of invasion. Multivariate analysis revealed that CSM, fold convergency, surface depression and surface redness were not significantly associated with deep submucosal invasion, whereas surface fullness and dilated blood vessels were correlated with invasion depth. Especially in the diminutive lesion group, the proportion of dilated blood vessels was higher. This indicates that the diagnostic value of white-light endoscopic features for both lesion nature and depth of invasion remains limited. Some endoscopic features are useful for determining the nature of the lesion but are insufficient for depth assessment, whereas others can predict the depth of invasion but offer limited value for characterizing the lesion. Therefore, relying solely on white-light endoscopy cannot provide a complete diagnosis for such lesions. Our study found that invasive pit pattern is valuable for determining the nature of colorectal lesions smaller than 2 cm. However, type VI may not guide therapeutic decisions. Hiroshi Kawano et al [25] indicate that Type VI glandular opening patterns encompass a wide range of irregularities, varying from mild to severe. In our study data, most of early-stage non-pedunculated CRCs smaller than 2 cm exhibited Type VI pit patterns regardless of deep submucosal invasion and Type VN pit patterns were rare. Thus, pit pattern classification may help in predicting malignant transformation of lesions, but invasive pit patterns alone cannot reliably predict deep submucosal invasion. Our study found that JNET2B was not an independent risk factor for malignant transformation of colorectal lesions. This may be attributed to the large proportion of JNET2A type in the ECC group. JNET2B can serve as an independent predictor of deep submucosal invasion in lesions. However, our study demonstrated that JNET2B predicted the depth of invasion in ECC with an accuracy of 0.67 and a sensitivity of 0.60. Besides, a significant number of cases in the superficial invasion group also exhibited JNET2B in our study. This means that using JNET classification alone may still lead to misdiagnosis. In our study, 21 cases met the diagnostic criteria for de novo carcinoma. The proportion of de novo carcinoma increased progressively across intramucosal carcinoma group, submucosal superficial invasion group, and submucosal deep invasion group. Furthermore, the de novo carcinogenic pathway was an independent risk factor for submucosal deep invasion. Due to the limited sample size in this study, serrated adenomas were not analyzed separately. We plan to expand the sample size for further investigation. This study does not aim to replace established magnifying endoscopic classifications such as JNET or pit pattern analysis. Instead, it focuses on small non-pedunculated lesions < 2 cm, in which conventional size-based assumptions may underestimate invasive potential. In this specific subgroup, selected white-light features and the endoscopically suspected de novo phenotype may provide additional contextual information for identifying lesions at high risk of deep submucosal invasion. We introduced a nomogram to enable readers to more intuitively understand the impact of each predictor on outcomes [26]. Due to our small sample size, we are concerned about overfitting in the predictive model. However, the small difference between AUC of the training set and the test set demonstrates that the nomogram prediction model obtained from this study does not suffer from this problem and the model has a predictive value for deep submucosal invasion in ECC. These results indicate that the model possesses preliminary predictive value and can assist in guiding personalized clinical treatment decisions. This study had some limitations: First, the sample size was limited, and the study’s single-center nature may have caused bias. Another important limitation is the number of deep submucosal invasion events was limited relative to the number of candidate predictors. Therefore, the nomogram should be interpreted cautiously and validated in larger independent cohorts before routine clinical application. Finally, it should be noted that our research involved a retrospective study of static endoscopic images, which may introduce assessment bias. Conclusions Non-pedunculated colorectal lesions smaller than 2 cm are at risk for carcinoma and deep submucosal invasion. Surface fullness, dilated blood vessels in tumor periphery, JNET2B/3 type, and de novo carcinoma pathway are significant indicators of invasive depth for submucosal deep invasion in early, non-pedunculated CRC less than 2 cm in diameter. CSM, surface depression, and invasive pit patterns may help predict whether non-pedunculated colorectal lesions less than 2 cm in diameter are malignant. The prediction model constructed according to the above independent risk factors had preliminary predictive value for the depth of submucosal invasion in ECC of non-pedunculated lesions smaller than 2 cm. However, further studies and more data are needed for validation.