Wei JM, Chen SX, He T, Wang JF. Application of artificial intelligence-driven three-dimensional imaging in preoperative planning for rectosigmoid colon cancer. World J Gastrointest Surg 2026; 18(5): 119310 [DOI: 10.4240/wjgs.v18.i5.119310]
Corresponding Author of This Article
Jun-Fu Wang, Department of General Surgery, The 1st Affiliated Hospital, Jiangxi Medical College, Nanchang University, No. 17 Yongwai Zheng Street, Nanchang 330006, Jiangxi Province, China. fu5718418@163.com
Research Domain of This Article
Oncology
Article-Type of This Article
research-article
Open-Access Policy of This Article
This article is an open-access article which was selected by an in-house editor and fully peer-reviewed by external reviewers. It is distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/
Baishideng Publishing Group Inc, 7041 Koll Center Parkway, Suite 160, Pleasanton, CA 94566, USA
Share the Article
Wei JM, Chen SX, He T, Wang JF. Application of artificial intelligence-driven three-dimensional imaging in preoperative planning for rectosigmoid colon cancer. World J Gastrointest Surg 2026; 18(5): 119310 [DOI: 10.4240/wjgs.v18.i5.119310]
Jian-Ming Wei, Shang-Xiang Chen, Ting He, Jun-Fu Wang, Department of General Surgery, The 1st Affiliated Hospital, Jiangxi Medical College, Nanchang University, Nanchang 330006, Jiangxi Province, China
Co-first authors: Jian-Ming Wei and Shang-Xiang Chen.
Author contributions: Wei JM and Chen SX designed the study and drafted the paper; He T gathered the data; Wang JF critically revised the manuscript and approved the final version for publication. Wei JM and Chen SX contributed equally to this work as co-first authors.
AI contribution statement: No AI tools were used. We did not use AI tools to write any part of the manuscript body. We did not use AI tools for language polishing, translation, data analysis, or manuscript writing assistance. AI tools are not involved in calculations or research results. No AI tools were used to generate images.
Supported by Nanchang University First Affiliated Hospital Clinical Research Cultivation Fund Project, No. YFYLCYJPY202424; and the Chronic Disease Management Research Project of National Health Commission Capacity Building and Continuing Education Center, No. GWJJMB202510022033.
Institutional review board statement: The study was reviewed and approved by the Medical Ethics Committee of the First Affiliated Hospital of Nanchang University.
Informed consent statement: The Institutional Review Board of the First Affiliated Hospital of Nanchang University waived the requirement for informed consent because of the minimal risk.
Conflict-of-interest statement: The authors of this manuscript declare no relationships with any companies, whose products or services may be related to the subject matter of the article.
STROBE statement: The authors have read the STROBE Statement-checklist of items, and the manuscript was prepared and revised according to the STROBE Statement-checklist of items.
Data sharing statement: No additional data are available.
Corresponding author: Jun-Fu Wang, Department of General Surgery, The 1st Affiliated Hospital, Jiangxi Medical College, Nanchang University, No. 17 Yongwai Zheng Street, Nanchang 330006, Jiangxi Province, China. fu5718418@163.com
Received: January 26, 2026 Revised: February 11, 2026 Accepted: March 4, 2026 Published online: May 27, 2026 Processing time: 124 Days and 3.2 Hours
Abstract
BACKGROUND
Accurate preoperative assessment of tumor burden and the distance from the tumor’s lowest border to the anal verge (DTAV) is essential for planning treatment in rectosigmoid cancer. Conventional imaging modalities offer limited quantitative evaluation of these parameters. Artificial intelligence-driven digital three-dimensional imaging (AI-3D digital imaging) may address this limitation.
AIM
To study the application of AI-3D digital imaging for the preoperative assessment of tumor burden and DTAV.
METHODS
We analyzed patients with rectosigmoid cancer treated in our Department of Gastrointestinal Surgery between July 2024 and January 2026 and collected their clinical data. Tumor burden and DTAV were assessed via AI-3D digital imaging and computed tomography (CT), and compared with the reference standard from pathological specimens. We evaluated the diagnostic accuracy of these modalities for tumor parameters using Bland-Altman plots, scatter plots, receiver operating characteristic curves, and intraclass correlation coefficients (ICC).
RESULTS
We found that the MD in maximum tumor diameter and cross-sectional area between pathological specimens and AI-3D digital imaging were 0.602 cm and 0.150 cm², respectively, indicating high agreement (ICC = 0.921 and ICC = 0.846). This agreement was higher than that achieved with CT. For DTAV measurement, the MD between AI-3D digital imaging and colonoscopy was 2.079 cm, also demonstrating high agreement (R2 = 0.8227, ICC = 0.907). Bland-Altman and scatter plot analyses confirmed superior agreement between AI-3D digital imaging and pathological specimens (R2 = 0.8482 and R2 = 0.7149) compared to CT. In predicting lymph node invasion, AI-3D digital imaging showed a sensitivity of 80% and a specificity of 62.5%, both significantly higher than the corresponding values for CT (60% and 29.2%). The area under the curve (AUC) for AI-3D was 0.713, markedly exceeding that of CT (AUC = 0.446).
CONCLUSION
AI-3D digital imaging demonstrates good efficacy for quantitatively assessing tumor burden and DTAV in rectosigmoid cancer and shows particular utility in predicting lymph node metastasis. This technology can improve the accuracy of preoperative assessment, thereby facilitating individualized surgical planning.
Core Tip: In this study, Artificial intelligence-driven digital three-dimensional imaging demonstrated significant advantages in the preoperative evaluation of rectosigmoid cancer. Compared with pathological specimens, it showed high consistency in measuring the maximum tumor diameter and cross-sectional area. Moreover, the sensitivity of this technique in predicting lymph node metastasis was significantly higher than that of computed tomography. This technology can enhance the accuracy of preoperative assessment and assist in individualized surgical planning.
Citation: Wei JM, Chen SX, He T, Wang JF. Application of artificial intelligence-driven three-dimensional imaging in preoperative planning for rectosigmoid colon cancer. World J Gastrointest Surg 2026; 18(5): 119310
Colorectal cancer is a leading digestive tract malignancy worldwide, with high incidence and mortality. Global cancer statistics for 2022 indicate approximately 1.9 million new cases, ranking it third in incidence, and 900000 deaths, placing it second in cancer mortality[1]. Rectosigmoid cancer represents a major subtype of colorectal cancer. In recent years, rectosigmoid cancer has shown a trend of increasing incidence among younger individuals. This shift in epidemiology not only elevates increases the individual risk of disease, but also places a significant burden on the healthcare system and society’s economy[2]. In the treatment system for rectosigmoid cancer, accurately assessing tumor burden and the distance from the tumor’s lowest border to the anal verge (DTAV) is essential for determining individualized therapy, surgical planning, and prognosis[3]. Tumor burden, which includes primary tumor size and location along with regional lymph node status, reflects biological aggressiveness and is central to TNM staging and prognostic evaluation[4,5]. The DTAV is a decisive parameter for selecting the surgical approach, the need for a prophylactic stoma, and the feasibility of sphincter preservation[6,7]. Following rectosigmoid cancer resection, deep pelvic anastomoses are prone to leakage due to poor blood supply or high tension, potentially resulting in pelvic infection and anal dysfunction[8]. Consequently, surgeons routinely recommend a prophylactic stoma for high-risk patients, such as those receiving neoadjuvant chemoradiotherapy, with a low anastomosis, or with poor nutritional status, to mitigate leakage risk. Based on DTAV, rectal cancer is classified as low (DTAV < 5 cm), mid (DTAV 5-10 cm), or high (DTAV > 10 cm)[9]. Achieving curative intent for low rectal cancer typically necessitates abdominoperineal resection and the formation of a permanent stoma after anal excision. Therefore, a combined assessment of tumor burden and DTAV integrates the tumor’s biological behavior with its precise anatomical location, which is crucial for preserving physiological function and improving quality of life.
Computed tomography (CT), magnetic resonance imaging (MRI), and colonoscopy are conventional modalities for evaluating rectosigmoid cancer. Colonoscopy, the gold standard for preoperative diagnosis of colorectal tumors, provides direct visualization of lesion morphology and permits pathological biopsy. However, this invasive intraluminal examination cannot assess extraluminal conditions such as bowel wall involvement or invasion of surrounding lymph nodes[10]. CT scanning can detect distant metastases and evaluate the overall disease status, but it offers limited discrimination of lymph node invasion in the rectosigmoid region and of the layered structures of the intestinal wall. MRI clearly delineates the relationship between the tumor and the intestinal wall layers, yet it is susceptible to motion artifacts, requires long acquisition times, and incurs higher costs[11-13]. Consequently, developing novel, non-invasive imaging techniques is important for advancing precision treatment in rectosigmoid cancer. Three-dimensional (3D) visualization technology employs computer graphics algorithms to render complex data, displaying the morphological characteristics of anatomical structures. In gastrointestinal tumor surgery, this technology is commonly used to delineate tumor location and the variations and distribution of surrounding vascular structures, thereby aiding precise surgical planning[14]. However, its reliance on manual operation results in relatively low workflow efficiency, and its practical adoption rate in surgical settings does not exceed 25%[15]. Therefore, integrating artificial intelligence algorithms has become an important strategy for 3D imaging to overcome the above limitations.
Artificial intelligence-driven digital 3D imaging (AI-3D digital imaging) integrates multi-layered, multi-modal clinical data through artificial intelligence to intelligently segment, render, and construct high-quality 3D models. Studies show that this technology can improve a surgeon’s accuracy in identifying pulmonary lesion structures, increasing the identification accuracy to 85%, compared to traditional CT imaging[16]. Yang et al[17] further demonstrated that preoperative planning for direct anterior approach total hip arthroplasty using an AI-assisted 3D system enhances prosthesis size matching precision, reduces operative time, and decreases intraoperative blood loss. Despite its significant application value in cardiothoracic surgery and orthopedics, research employing AI-3D digital imaging for the quantitative assessment of preoperative tumor parameters in rectosigmoid diseases remains limited. Consequently, this study systematically evaluates the comprehensive analytical capability of AI-3D digital imaging in assessing tumor burden and DTAV to validate the technology’s effectiveness and accuracy for the preoperative evaluation of rectosigmoid cancer. The findings may offer a new methodology and theoretical foundation for developing precise, personalized clinical treatment plans.
MATERIALS AND METHODS
Study subjects
This study initially enrolled 65 patients clinically diagnosed with rectosigmoid tumors at our hospital between July 2024 and January 2026. Inclusion criteria mandated the completion of preoperative CT, colonoscopy, and AI-3D digital imaging, as well as a pathological diagnosis of sigmoid colon or rectal cancer. Patients were excluded if they had intestinal obstruction due to an excessively large tumor volume, or if distant metastasis was detected preoperatively or intraoperatively. According to these criteria, we excluded 23 patients who lacked AI-3D digital imaging and 3 patients whose pathological diagnosis was rectosigmoid adenoma. The final analysis therefore included 39 cases (Figure 1). The hospital's ethics committee approved this study (IITS2025898). Given its single-center, observational, non-randomized design and minimal risk, the requirement for informed consent was waived.
Figure 1 Clinical cohort establishment flowchart.
DTAV: Distance from the tumor’s lowest border to the anal verge; AI-3D: Artificial intelligence-driven digital three-dimensional imaging; ICC: Intraclass correlation coefficients; CT: Computed tomography.
Baseline data collection
General data were obtained from the electronic medical record system, pathology information system, picture archiving and communication system, and surgical anesthesia records at our center. A complete patient information chain was constructed by linking and matching records across these systems. Patient age and sex were extracted from admission records. Preoperative and postoperative pathological reports provided the histological type, differentiation grade, TNM stage, and lymphovascular and lymph node invasion status of the rectosigmoid tumors. Surgical records or discharge summaries confirmed whether sphincter-preserving surgery and/or a prophylactic stoma was performed. Multidisciplinary team records together with chemotherapy and radiotherapy records confirmed the administration of postoperative adjuvant therapy.
CT imaging data
Patients underwent spiral CT scanning after admission. After fasting, they were placed supine and received an intravenous injection of iopamidol (370 mgI/mL) at a flow rate of 3.5 mL/s. Enhanced scanning was performed using a 28-row/256-slice CT scanner, and the data were saved. The scanning range extended from the xiphoid process superiorly to the lower border of the pubic symphysis inferiorly. Scanning parameters were as follows: Pitch 0.96, slice interval 0.5 mm, matrix 512 × 512, and a scanning threshold of 150 HU. Two associate chief radiologists, each with over 10 years of experience, reviewed the CT imaging reports to assess tumor size and location, lymph node invasion, and/or enlargement (Figure 2A). All imaging data were archived in the hospital information system.
Figure 2 Three techniques for detecting tumor burden and the distance from the tumor’s lowest border to the anal verge.
A: Detect tumor burden in colorectal cancer using computed tomography equipment; B: Detect distance from the tumor’s lowest border to the anal verge (DTAV) using colonoscopy equipment; C: Detect tumor burden in colorectal cancer and DTAV using artificial intelligence-driven digital three-dimensional. DTAV: Distance from the tumor’s lowest border to the anal verge; AI-3D: Artificial intelligence-driven digital three-dimensional imaging; CT: Computed tomography.
Colonoscopy imaging data
All patients underwent preoperative colonoscopy. A senior gastroenterologist performed the procedure using a high-definition electronic colonoscopy system to evaluate the colorectal lumen and tumor. After identifying a rectal or sigmoid colon tumor, the endoscopist positioned the scope tip at the tumor’s lower border and slowly withdrew it to the clearly visualized anorectal ring (dentate line). Alternatively, an assistant performed a digital rectal examination at the external anal verge to assist localization. The built-in measurement function of the endoscopy system or a marked catheter recorded the straight-line distance from the tumor’s lowest point to the anal verge, defined as the colonoscopically measured DTAV (Figure 2B). This DTAV value was collected from the patient medical records.
AI-3D digital imaging data
A high-dimensional database was constructed to store and manage multi-modal feature vectors, with each imported patient record comprising a unique identifier alongside imaging and biomarker features. Visualization and interactive analysis of these high-dimensional data were implemented using programming languages such as R and Python and their associated toolkits (Figure 2C). For feature fusion, an attention mechanism assigned contribution weights to features from each modality to enhance the model's extraction of key information and its generalization performance. The resulting fused multi-modal imaging and biomarker features were then integrated into the high-dimensional database. For the automated segmentation and 3D reconstruction of tumors, we employed the nnU-net framework and automatically configured the network topology and hyperparameters based on the dataset’s “fingerprint characteristics” (spacing distribution, intensity distribution). Using an encoder-decoder-based U-Net architecture, the convolutional neural network compresses the input image and extracts deep features, then progressively restores spatial details through upsampling and skip connections to achieve precise segmentation of tumors and organs. In terms of training strategy, 5-fold cross-validation was adopted along with ensemble selection to enhance generalization capability (Figure 3A). For lymph node invasion screening, a 3D convolutional neural network was constructed for feature extraction and probability map generation (Figure 3B). Regarding tumor parameter assessment, particularly the tumor distance from the anal verge, we applied skeletonization and centerline computation algorithms on the segmentation masks instead of relying solely on deep learning predictions to ensure geometric measurement accuracy (Figure 3C). The system automatically calculated critical clinical indicators, including the tumor's maximum diameter and volume, the distance from its distal margin to the anal verge, the status of peritumoral lymph node enlargement, and the tumor's spatial orientation relative to adjacent bony landmarks. A peritumoral lymph node with a shortest diameter exceeding 8 mm was defined as enlarged, suggesting nodal invasion[18].
Figure 3 The flowchart of artificial intelligence-driven digital three-dimensional imaging.
A: Computed tomography-based automatic segmentation and three-dimensional reconstruction; B: Automated screening and segmentation of enlarged lymph nodes; C: Tumor-to-anal-verge distance estimation. CT: Computed tomography.
Statistical analysis
The data were processed and statistical graphs were generated using SPSS 20.0 and GraphPad Prism 9. Categorical variables were compared with the χ2 test, and receiver operating characteristic curves were plotted to evaluate the utility of different techniques. For continuous variables, normality was assessed with the Shapiro-Wilk test. Data conforming to a normal distribution were analyzed using an independent or paired t test, whereas non-normally distributed data were analyzed with the Mann-Whitney U test or Wilcoxon signed-rank test. Normally distributed data are presented as mean ± SD, and non-normally distributed data as median (interquartile range). P < 0.05 was considered to indicate statistical significance.
RESULTS
Baseline characteristics
The clinical data of the 39 patients are summarized in Table 1. The cohort comprised 23 males (59%) and 16 females (41%), with a male-to-female ratio of 1.44:1. The median age was 62 years (range 40-71). Most tumors were rectal cancers (76.9%), with sigmoid colon cancers comprising the remaining 23.1%. The majority of tumors were moderately differentiated. Intermediate- and high-grade tumor budding was present in 53.8% of cases. Lymphovascular and perineural invasion were identified in 41% and 51.3% of patients, respectively. A stoma was created in 35.9% of patients. Anal sphincter preservation was achieved in 97.4% of cases, and 56.4% of patients received postoperative adjuvant therapy.
Figure 4 presents the imaging for a rectal cancer case. The AI-3D reconstruction accurately depicted the tumor location (yellow module), consistent with the CT findings (Figure 4A-C). The system automatically renders and color-codes anatomical modules, enabling surgeons, patients, and families to observe the spatial relationships between the tumor and adjacent structures, including the pancreas, spleen, liver, stomach, jejunum and ileum, colon, kidneys, psoas major muscle, ureters, prostate, uterus, and major arteries and veins (Figure 4B). This visualization facilitates safer surgical planning and helps minimize iatrogenic injury. Furthermore, the technology can intelligently quantify tumor parameters and predict lymph node invasion (Figure 4D-F).
Figure 4 Artificial intelligence-driven digital three-dimensional imaging digital imaging render.
A: Cross-sectional imaging view of the tumor; B: Sagittal imaging view of the tumor; C: Location of the rectal tumor (yellow area); D: Tumor size (a: 3.39 cm, b: 3.03 cm); E: Distance from the tumor’s lowest border to the anal verge measurement result (a: 15.63 cm); F: Enlarged lymph nodes.
Quantitative assessment of tumor burden characteristics and DTAV in rectosigmoid cancer using different techniques
Table 2 presents the measurement results and statistical comparisons among AI-3D digital imaging, CT, and pathological specimens. For maximum tumor diameter, the difference between AI-3D and CT measurements was not significant (MD: -0.106 cm; t = -0.471, P = 0.640). In contrast, pathological measurements differed significantly from both CT (MD: -0.708 cm; t = -3.332, P = 0.002) and AI-3D imaging (MD: -0.602 cm; t = -7.367, P < 0.001). In the assessment of the maximum cross-sectional area of the tumor, the measurement difference between AI-3D and CT was approximately 5.430 cm², and the difference between pathological specimens and CT was 5.280 cm², both of which were statistically significant. In contrast, the measurement difference between pathological specimens and AI-3D was only -0.150 cm², indicating a high degree of consistency between the two methods. For DTAV, the difference between AI-3D and colonoscopy was 2.079 cm (t = 2.024, P = 0.05), demonstrating comparable performance between the two techniques. These findings indicate that AI-3D digital imaging may improve the quantitative assessment of tumor burden.
Table 2 Analysis of differences in tumor burden and the distance from the tumor’s lowest border to the anal verge measurements by different methods.
Consistency analysis of tumor burden and DTAV in rectosigmoid cancer based on Bland-Altman and Scatter plots
This study assessed the consistency of measurements from AI-3D digital imaging, pathological specimens and CT scans using Bland-Altman and scatter plots. Measurements from pathological specimens and AI-3D imaging showed high consistency for both maximum tumor diameter and maximum cross-sectional area (R2 = 0.8482, intraclass correlation coefficients (ICC) = 0.921 and R2 = 0.7149, ICC = 0.846) (Figures 5A, 5B, 6A, and 6B). The consistency between pathological specimens and CT measurements was weaker (R2 = 0.2684, ICC = 0.518; R2 = 0.1654, ICC = 0.407) (Figures 5C, 5D, 6C, and 6D). Similarly, AI-3D and CT also demonstrated weak agreement (R2 = 0.2324, ICC = 0.482; R2 = 0.2677, ICC = 0.517) (Figures 5E, 5F, 6E, and 6F). For DTAV, AI-3D and colonoscopy findings were highly consistent (R2 = 0.8227 and ICC = 0.907). These results indicate that AI-3D and pathological specimens are highly consistent for measuring tumor burden (Figure 7A-C). The findings confirm the accuracy and reliability of AI-3D digital imaging for the quantitative clinical assessment of tumor burden.
Figure 5 Bland-Altman and scatter plots evaluating the pairwise agreement in measurements of the maximum tumor diameter among the three methods.
A: Scatter plots of agreement between measurements from pathological specimens and artificial intelligence-driven digital three-dimensional imaging (AI-3D digital imaging); B: Bland-Altman of agreement between measurements from pathological specimens and AI-3D digital imaging; C: Scatter plots of agreement between measurements from pathological specimens and computed tomography (CT); D: Bland-Altman of agreement between measurements from pathological specimens and CT; E: Scatter plots of agreement between measurements from AI-3D digital imaging and CT; F: Bland-Altman of agreement between measurements from AI-3D digital imaging and CT. AI-3D: Artificial intelligence-driven digital three-dimensional imaging; CT: Computed tomography.
Figure 6 Bland-Altman and scatter plots evaluating the pairwise agreement in measurements of the maximum tumor cross-sectional area among the three methods.
A: Scatter plots of agreement between measurements from pathological specimens and artificial intelligence-driven digital three-dimensional imaging (AI-3D digital imaging); B: Bland-Altman of agreement between measurements from pathological specimens and AI-3D digital imaging; C: Scatter plots of agreement between measurements from pathological specimens and computed tomography (CT); D: Bland-Altman of agreement between measurements from pathological specimens and CT; E: Scatter plots of agreement between measurements from AI-3D digital imaging and CT; F: Bland-Altman of agreement between measurements from AI-3D digital imaging and CT. AI-3D: Artificial intelligence-driven digital three-dimensional imaging; CT: Computed tomography.
Figure 7 Evaluation of the distance from the tumor's lowest border to the anal verge and lymph node involvement using artificial intelligence-driven digital three-dimensional imaging compared with colonoscopy and computed tomography.
A and B: Bland-Altman plot and Scatter plot of agreement between measurements from artificial intelligence-driven digital three-dimensional imaging (AI-3D digital imaging) and colonoscopy; C: Intraclass correlation coefficient cluster diagram (1: The maximum tumor diameter; 2: The maximum tumor cross-sectional area; 3: Distance from the tumor’s lowest border to the anal verge); D: Predictive evaluation of lymph node involvement using AI-3D and computed tomography (1: Sensitivity; 2: Specificity; 3: Positive predictive value; 4: Negative predictive value); E: Receiver operating characteristic curves of the two techniques. AI-3D: Artificial intelligence-driven digital three-dimensional imaging; CT: Computed tomography.
Preoperative assessment of lymph node invasion in rectosigmoid cancer using AI-3D digital imaging technology vs CT
In Table 3 and Figure 7D, the sensitivity, specificity, positive predictive value, and negative predictive value of AI-3D digital imaging for predicting lymph node invasion were 80%, 62.5%, 57.1%, and 83.3%, respectively, compared to corresponding CT values of 60%, 29.2%, 34.6%, and 53.8%. The area under the curve (AUC) for AI-3D was 0.713, compared with 0.446 for CT (Figure 7E). These results indicate that AI-3D digital imaging provides a more accurate prediction of lymph node invasion than conventional CT.
Table 3 Assessment of perirectal lymph node invasion by artificial intelligence-driven digital three-dimensional imaging and computed tomography.
Accurate preoperative assessment is essential for the individualized and radical treatment of rectosigmoid cancer. Tumor size, DTAV, and regional lymph node status are key parameters that guide therapeutic decisions[19]. Measurements of the maximum tumor diameter and cross-sectional area from high-resolution MRI provide information on the extent of local invasion. When combined with an evaluation of the tumor’s relationship to the mesentery and surrounding organs, these measurements aid in T-staging and predicting circumferential resection margin status. The maximum tumor cross-sectional area offers a more comprehensive representation of both the intramural tumor volume and the degree of luminal stenosis[20,21]. For patients with a large primary tumor or extensive lymph node metastasis, surgery alone is often insufficient for radical treatment and is associated with a high risk of local recurrence. In such cases, neoadjuvant chemoradiotherapy is required to reduce tumor volume and eliminate metastatic lymph nodes, thereby achieving downstaging. Conversely, local excision or radical surgery is the standard approach for patients with a small tumor diameter and imaging findings suggesting no lymph node involvement. Maximizing anal function preservation in these patients significantly improves postoperative quality of life[22,23]. In high rectal cancer, low anterior resection is typically performed to achieve both radical resection and anal preservation. For low rectal cancer, abdominoperineal resection is usually necessary to ensure an adequate distal margin, resulting in a permanent stoma[24,25]. CT and colonoscopy are the primary tools for assessing these parameters, yet both have inherent limitations in imaging principles and measurement conditions that can lead to discrepancies with postoperative pathology. Although 3D imaging can reduce such discrepancies, its accuracy is often suboptimal, and the modeling process is time-consuming[26]. To address these limitations, the application value of AI-driven digital 3D imaging in the surgical treatment of colorectal cancer is gradually gaining attention. This study systematically evaluated the efficacy of AI-3D digital imaging for measuring key tumor parameters and assessing lymph node invasion, demonstrating its advantages in the preoperative assessment of rectosigmoid cancer.
Pathological specimens, the gold standard for tumor diagnosis, provide ex vivo and dehydrated tumor information. The process of tissue soaking and parameter measurement, however, can cause tumor shrinkage, introducing systematic errors between pathological and imaging measurements[27]. In contrast, CT imaging, which relies on continuous two-dimensional slices, can underestimate or overestimate the dimensions of morphologically irregular tumors. These inherent differences in imaging principles and dimensionality inevitably lead to statistical discrepancies. Henschke et al[28] noted that CT detection of tumor boundaries in low-contrast tissues may introduce substantial errors, which are further exacerbated by adjacent blood vessels or organs. AI-3D digital imaging automatically generates rotatable, sectionable, and measurable 3D models. These stereoscopic images restore the tumor's true spatial relationship to surrounding vessels, mesentery, and organs while substantially reducing reconstruction time and improving workflow efficiency compared to traditional manual methods[29]. In this study, measurements of maximum tumor diameter and cross-sectional area reflecting tumor infiltration volume showed high concordance between pathological specimens and AI-3D digital imaging (R2 = 0.8482, ICC = 0.921; R2 = 0.7149, ICC = 0.846). The MD between techniques were -0.602 cm and -0.150 cm², respectively, with AI-3D measurements being consistently slightly lower. Only weak agreement was observed between CT and pathological specimens and between AI-3D and CT. These findings indicate that AI-3D digital imaging offers a distinct advantage for the preoperative quantification of tumor volume burden, providing a reliable imaging basis for assessing local invasion in rectosigmoid tumors.
For patients with rectosigmoid cancer who present with high-risk factors, precise measurement of the DTAV is critical for determining the feasibility of sphincter-preserving surgery. During colonoscopy, air insufflation distends the colonic lumen and can introduce minor errors in DTAV measurement[30]. In contrast, AI-3D digital imaging technology for DTAV measurement simulates the colon's natural state, generating data that closely approximate actual anatomical conditions. Previous research reported an ICC of 0.7 and a MD of 2.5 cm between MRI and colonoscopy for DTAV measurement[31]. In our study, DTAV measurements from AI-3D digital imaging and colonoscopy showed no statistically significant difference (P = 0.05), indicating their comparability. The MD between the two techniques was 2.079 cm, demonstrating a high level of agreement (R2 = 0.8227, ICC = 0.907). Consequently, for patients with severe luminal stenosis that precludes complete colonoscopy, or for those requiring multi-angle assessment of the distal resection margin, AI-3D digital imaging can serve as an effective non-invasive supplement or alternative. This technology assists surgeons in more accurately planning the extent of surgical resection preoperatively, thereby enhancing the precision of sphincter-preservation decisions.
Accurate assessment of regional lymph node involvement remains a significant challenge for precise preoperative staging and intraoperative dissection in rectosigmoid cancer. While CT is essential for evaluating distant metastasis, its performance in assessing local lymph node status is inconsistent, with reported specificities and sensitivities ranging from approximately 55%-95% and 13%-90%, respectively[32,33]. In our cohort, CT demonstrated limited utility for predicting regional lymph node involvement (AUC = 0.446), with a sensitivity of 60% and a specificity of 29.2%, aligning with documented limitations. Conversely, AI-3D digital imaging showed improved predictive performance. Its sensitivity (80%) and specificity (62.5%) exceeded those of CT, and the AUC reached 0.713. Notably, the negative predictive value of AI-3D imaging was 83.3%, suggesting it can reliably identify patients at low risk for lymph node metastasis. This capability could help select candidates for local excision or limited lymphadenectomy, potentially reducing surgical morbidity and preserving quality of life. Therefore, AI-3D digital imaging may improve the accuracy of predicting peritumoral lymph node involvement in this context. The underlying reasons are likely multifactorial: The 3D model allows clearer visualization of spatial relationships between lymph nodes, the primary tumor, and adjacent vasculature, offering critical topological data. Additionally, artificial intelligence algorithms can synthesize multiparametric features, including density homogeneity, margin characteristics, and enhancement patterns.
While Bland-Altman and scatter plots demonstrated the consistency between methods, the AI-3D digital imaging technology was developed and validated on internal data. This study has limitations, including its single-center design and relatively small sample size, which may restrict the diversity of patient demographics and potentially lead to selection bias. Furthermore, to ensure measurement accuracy, although all CT or colonoscopy examinations were performed by senior physicians (with over 10 years of experience), some slight inter-observer variability in measurements was possible. Therefore, future studies will involve the collection of multicenter data and the inclusion of multiple physicians for independent blinded assessments. However, the ongoing development of AI-3D digital imaging technology shows considerable clinical potential. For example, integration into intraoperative navigation or virtual reality systems could offer surgeons real-time, 3D anatomical guidance.
CONCLUSION
In summary, this study demonstrates that AI-driven 3D digital imaging is a valuable preoperative assessment tool for rectosigmoid cancer. The technology shows high concordance with pathology in measuring tumor burden and compensates for the limited sensitivity of CT in assessing regional lymph node invasion, thereby supplying critical information for accurate preoperative planning and staging. In future practice, AI-3D digital imaging may become a powerful tool for supporting individualized, precise treatment of rectosigmoid cancer.
Freeman CL, Noble J, Menges M, Villanueva R, Nakashima JY, Figura NB, Tonseth RP, Werner Idiaquez D, Skelson L, Smith E, Abraham-Miranda J, Corallo S, De Avila G, Castaneda Puglianini OA, Liu H, Alsina M, Nishihori T, Shain KH, Baz R, Blue B, Grajales-Cruz A, Koomen JM, Atkins RM, Hansen DK, S Silva A, Kim J, Balagurunathan Y, Locke FL. Tumor burden quantified by soluble B-cell maturation antigen and metabolic tumor volume determines myeloma CAR-T outcomes.Blood. 2025;145:1645-1657.
[RCA] [PubMed] [DOI] [Full Text][Cited by in Crossref: 2][Cited by in RCA: 24][Article Influence: 24.0][Reference Citation Analysis (0)]
Duan X, Zhao Q, Li F. Association of CT features with TNM stage and pathology of patients with rectal cancer and their significance in evaluation of efficacy and prognosis.J BUON. 2020;25:1430-1435.
[PubMed] [DOI]
Sighinolfi MC, Menezes AD, Patel V, Moschovas M, Assumma S, Calcagnile T, Panio E, Sangalli M, Turri F, Sarchi L, Micali S, Varca V, Annino F, Leonardo C, Bozzini G, Cacciamani G, Gregori A, Morini E, Terzoni S, Eissa A, Rocco B. Three-Dimensional Customized Imaging Reconstruction for Urological Surgery: Diffusion and Role in Real-Life Practice from an International Survey.J Pers Med. 2023;13:1435.
[RCA] [PubMed] [DOI] [Full Text][Cited by in RCA: 6][Reference Citation Analysis (0)]
Chen X, Xu H, Qi Q, Sun C, Jin J, Zhao H, Wang X, Weng W, Wang S, Sui X, Wang Z, Dai C, Peng M, Wang D, Hao Z, Huang Y, Wang X, Duan L, Zhu Y, Hong N, Yang F. AI-based chest CT semantic segmentation algorithm enables semi-automated lung cancer surgery planning by recognizing anatomical variants of pulmonary vessels.Front Oncol. 2022;12:1021084.
[RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)][Cited by in RCA: 17][Reference Citation Analysis (0)]
Li X, Zhang S, Luo X, Gao G, Luo X, Wang S, Li S, Zhao D, Wang Y, Cui X, Liu B, Tao Y, Xiao B, Tang L, Yan S, Wu N. Accuracy and efficiency of an artificial intelligence-based pulmonary broncho-vascular three-dimensional reconstruction system supporting thoracic surgery: retrospective and prospective validation study.EBioMedicine. 2023;87:104422.
[RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)][Cited by in RCA: 20][Reference Citation Analysis (0)]
Wang L, Bai C, Wu Q, Ma Y, Li W, Nan X, Zhao X, Wang S, Cheng X. Study on the evaluative value of abdominal multi-slice spiral CT examination for distant metastasis and regional lymph node metastasis in colorectal cancer.Minerva Surg. 2024;79:674-677.
[RCA] [PubMed] [DOI] [Full Text][Cited by in RCA: 1][Reference Citation Analysis (0)]
