A comparative study of magnetic resonance imaging image-based deep learning and conventional turbo spin echo sequence in tumor-node staging of rectal cancer

Introduction

Rectal cancer is a common malignant tumor of the digestive system worldwide. In recent years, the incidence of rectal cancer has been increasing, with more prevalent occurrence in middle-aged and elderly people, and it is highly invasive, recurrent, and metastatic in nature. Early rectal cancer is generally asymptomatic, but with the continuous development of the disease, patients will have changes in bowel habits, abdominal pain, and other symptoms, and most patients have entered the middle and late stage by the time they are diagnosed, which is not conducive to a favorable prognosis (1-4). The 2021 release of the World Health Organization (WHO) classification criteria integrates the histological characteristics and molecular phenotypes of rectal cancer and proposes a new grading standard for the classification of tumors (5,6). This standard offers more precise guidance for diagnosing and treating rectal cancer, accurately reflecting the biological behavior of the tumor and its response to treatment, and helping clinicians formulate personalized treatment plans. Tumor-node (TN) staging is essential for treatment option selection and assessment of prognostic outcomes in rectal cancer. The T stage is employed to describe tumor invasion depth. Specifically, it reflects the extent to which the tumor has spread outward from the rectal wall, and the N stage focuses on the status of lymph node metastasis, demonstrating whether the tumor has spread to regional lymph nodes and its extent. These two parameters determine treatment strategy selection and prognosis. The release of the rectal cancer classification standard and the application of the TN have marked the entry of rectal cancer diagnosis and treatment into a more precise and individualized era. Tumor infiltration degrees and the presence or absence of pelvic lymph node metastasis affect surgical plans. Mesorectal excision is recommended for patients with T1–2 stage rectal cancer without lymph nodes or distant metastases. At the same time, for locally advanced rectal cancer (cT3NxM0) with a clinical stage of T3, neoadjuvant chemoradiotherapy (nCRT) should be recommended regardless of whether there is regional lymph node metastasis (cN ±) or suspected distant metastasis (cM0/1a) as evaluated by preoperative imaging (7). Different types and grades of rectal cancer have varying prognoses and treatment options. Therefore, accurately assessing the TN stage of rectal cancer is of great significance for determining treatment plans and prognosis (8).

Currently, magnetic resonance imaging (MRI) is the most common preoperative diagnostic method for rectal cancer patients. It has the advantages of being noninvasive, multiparameter, and multidimensional and can show different signal intensity differences in tumor hemorrhage, necrosis, and edematous tissues, offering clinicians abundant and accurate imaging information. However, in terms of clarifying the extent of tumor infiltration, traditional MRI technology still has specific limitations.

Conventional MRI imaging techniques, including turbo spin echo (TSE), have been widely used in the TN staging of rectal cancer. With its high signal-to-noise ratio (SNR) and high resolution, TSE sequences can display the structure of each rectal wall layer and the relationship to the surrounding organs, offering a foundation for TN staging of rectal cancer (9).

However, the acquisition time of conventional TSE (conv TSE) sequence is prolonged, and conv TSE is susceptible to motion and magnetic susceptibility artifacts, as well as other factors, limiting image quality. This impacts the accurate assessment of the extent of tumor infiltration. Conventional acceleration of conv TSE sequences is routinely a factor of two, but further increases in acceleration cause a significant decrease in SNR. Existing techniques compensate for the SNR loss by averaging the signals several times, resulting in a linear increase over scan time (10).

With the progress of deep learning (DL) technology, image reconstruction algorithms (DL Recon) have a high potential to improve image quality. DL Recon can accurately capture and extract key image features through DL and training on a large amount of MRI image data. Moreover, it enables collaborative optimization of parallel acquisition techniques and noise reduction algorithms, thereby effectively suppressing image noise and artifacts, enhancing image resolution, and compensating for the SNR loss associated with high acceleration factors [GeneRalized Autocalibrating Partially Parallel Acquisitions (GRAPPA) =4]. This allows the adoption of higher acceleration factors in DL TSE sequences, achieving a 36.6% reduction in scanning time, which collectively demonstrates the comprehensive advantages of this technology in clinical workflow optimization (11,12).

Recently, studies have shown that DL Recon has attained remarkable results in various medical image reconstruction tasks, including improving image clarity, reducing acquisition time, and revolutionizing the medical imaging field (13). Although DL Recon has been widely used in imaging studies of various cancers, such as nasopharyngeal carcinoma, breast cancer, and cervical cancer, the effectiveness of this technique in the clinical application of rectal cancer has not been investigated.

This study aimed to systematically compare the diagnostic efficacy of two comprehensive clinical imaging acquisition and reconstruction protocols, rather than evaluating a single image reconstruction algorithm in isolation. Specifically, we evaluated the integrated performance of a “fast spin-echo imaging protocol incorporating DL-based reconstruction with a higher acceleration factor” versus a “conventional spin-echo imaging protocol employing standard reconstruction and acceleration factors” in clinical applications. The effectiveness of the DL-based TSE sequence (DL TSE) compared to the conv TSE sequence was assessed for TN staging of rectal cancer. By collecting MRI image data from rectal cancer patients, the accuracy, reliability, and clinical utility of conv TSE sequence and DL TSE sequence in TN staging of rectal cancer were comprehensively examined. We anticipate that this study will provide more reliable and efficient imaging support for accurate diagnosis and treatment of rectal cancer, thus improving the prognosis and quality of life of patients. We present this article in accordance with the STARD reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1884/rc).

Methods

General information

The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of the Cancer Hospital, the Chinese Academy of Medical Sciences (approval No. NCC4435). Informed consent was provided by all the patients. A total of 60 patients who were observed at the Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College and diagnosed with rectal cancer were prospectively enrolled from October 2023 to October 2024. A total of 6 patients were excluded, and 54 patients were finally included. There were 32 males and 22 females aged 35–69 (mean ± standard deviation: 53±10) years. The inclusion criteria were as follows: (I) patients diagnosed with rectal cancer by clinical colonoscopy pathology and without treatment; (II) no contraindications to MRI scanning; and (III) over 18 years old. The exclusion criteria were as follows: (I) patients failed to complete the examination; and (II) the lesion was too small to visualize the lesion in the MRI image (Figure 1).

Figure 1 Flow chart of inclusion and exclusion of patients with rectal cancer. MRI, magnetic resonance imaging.

Scanning methods

All 60 patients underwent routine rectal MRI scans using a Siemens 3T MRI scanner (MAGNETOM Prisma, Siemens Healthcare, Erlangen, Germany), including sagittal and coronal T2-weighted imaging (SAG T2WI, COR T2WI), axial T2WI with fat suppression (TRA FS T2WI), axial T2WI (TRA T2WI), axial T1-weighted imaging (TRA T1WI) conv TSE sequence, and a corresponding DL TSE sequence, as well as diffusion-weighted imaging (DWI). The research DL reconstruction sequence (WIP1093 of VE11E; Siemens Healthcare) was provided. The localization of DL TSE sequences was obtained from conv TSE sequences to ensure fair comparison. Detailed imaging parameters are provided in Table 1.

Patients were placed supine in the center of the examination bed with arms raised above the head. To ensure image quality, all metal jewelry and clothing were removed from the patient before scanning, standardized breathing training was performed to reduce respiratory motion artifacts, and anticholinergic drugs (Racemic scopolamine hydrochloride injection, Hangzhou Minsheng Pharmaceutical Co., Ltd., China National Pharmaceutical License No. H33021707, specification 1 mL: 10 mg, adults each intramuscular injection 10 mg) to limit intestinal motility artifacts.

Pathologic diagnosis

Figure 2 illustrates the clinical and pathological staging criteria for rectal cancer evaluation. In T stage (T), the depth of tumor invasion was classified as: T0, no evidence of primary tumor; T1, tumor invades the submucosal layer without invading the muscularis propria; T2, tumor has invaded the muscularis propria and has not yet breached the muscularis propria; T3, tumor has breached the muscularis propria layer invading the rectal mesenteric fat; T4, tumor invades the peripheral organs and structures is further subdivided into T4a (tumor invades the peritoneal reflexes) and T4b (tumor invades other organs and structures). Regarding N stage (N), the evaluation criteria for suspicious lymph nodes include short diameter ≥5 mm, irregular morphology and heterogeneous signal can be considered suspicious if the lymph nodes are between 5 and 9 mm in size and need to meet two morphological criteria. The specific stages were as follows: N0, no lymph node metastasis was found; N1, 1–3 suspicious lymph nodes were found; N2, ≥4 suspicious lymph nodes were found. The intrinsic muscular layer [mesorectal fascia (MRF)] was also evaluated. This study compared the conv TSE TN staging, DL TSE TN staging, and colonoscopy pathological staging to calculate the diagnostic efficacy of conv TSE and DL TSE TN staging (14,15).

Figure 2 Diagram of T staging for rectal cancer. T, tumor.

Image analysis

The images were independently analyzed by two radiologists, each with over 5 years of experience in MRI diagnosis of colorectal cancer, including conv TSE and DL TSE, in a random order. To minimize recall bias, two radiologists independently reviewed the images at separate time points, with randomized sequences. The two scan sequences from the same participant were evaluated in a segregated manner. The reviewer was unaware of the type of scan sequence, the imaging report, and the assessment of the other reviewer (Figure 3). The image quality of two sequences in terms of lesion contour sharpness, image artifacts, lesion structural clarity, and diagnostic confidence were independently evaluated by two physicians (Table 2). The assessments were conducted using the Likert scale with a 5-point grading system, where 5 denotes the highest quality and 1 represents the lowest quality.

Figure 3 Representative axial high-resolution T2-weighted images of rectal cancer patients comparing DL TSE and conv TSE sequences. (A) DL TSE image of a 53-year-old male patient, colonoscopy pathology confirmed moderately differentiated adenocarcinoma of the rectum. The pathological TN stage was T2N1. The white arrow indicates an area with clear and well-defined texture. (B) Conv TSE image of the same patient as in (A), showing more blurred tumor margins and heterogeneous internal signal. The conv TSE image showed more blurred edges and uneven internal signals. The white arrow indicates the corresponding area, which appears blurry with loss of details. (C) DL TSE image of a 55-year-old female patient with pathologically confirmed T3N0 stage moderately differentiated adenocarcinoma. The muscularis propria (white box) demonstrates a more uniform low signal intensity. (D) Conv TSE image of the same patient as in (C), showing uneven signal intensity within the muscularis propria (white box), making it difficult to delineate tumor infiltration. conv TSE, conventional TSE; DL TSE, deep learning TSE; TN, tumor-node; TSE, turbo spin echo.

Statistical analysis

The software SPSS 26.0 (IBM Corp., Armonk, NY, USA) was employed for statistical analysis. The Kappa test for agreement between TN stage assignment by two radiologists was performed, with a Kappa value greater than 0.75 indicating good agreement between the data (Table 3) (16).

The diagnostic accuracy, specificity, and sensitivity of conv TSE sequences and DL TSE sequences on different T-stages were calculated. McNemar’s test and Kappa test analyses compared the TN staging of patients’ conv TSE sequences and DL TSE sequences with the TN staging verified by colonoscopy pathological analysis.

Results

General information of enrolled patients

In this study, 60 patients attending the Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College and diagnosed with rectal cancer by pathology were prospectively included, and 6 patients were excluded (2 patients did not complete the examination and 4 patients’ lesions were too small to observe the lesions in MRI images). A total of 54 patients were finally included; among them, 32 cases were male, 22 were female, and their ages ranged from 35 to 69 (53±10) years.

Consistency results

The image quality of conv TSE sequences and DL TSE sequences of all patients could meet the radiologists’ needs, the inter-observer agreement between two radiologists in subjectively evaluating the four image quality assessment metrics demonstrated substantial consistency, with Kappa values ranging from 0.803 to 0.922, indicating strong concordance, and the inter-observer agreement of TN staging across conv TSE and DL TSE was strong, with a Kappa of 0.904–0.957 (Table 3).

TN staging results of MRI and pathologic diagnosis

Diagnostic accuracy is defined as the proportion of cases in which the imaging staging results are completely consistent with the pathological staging results from colonoscopy, the gold standard.

Pathologic diagnosis of T1 stage patients included 1 case, T2 stage included 6 cases, T3 stage included 40 cases, T4 stage included 7 cases, conv TSE correctly diagnosed staging of 42 cases, incorrectly diagnosed 12 cases, 5 overestimated and 7 underestimated. DL TSE correctly diagnosed the staging of 49 cases, incorrectly diagnosed 5 cases, 0 overestimated, and 5 underestimated. The accuracy of the conv TSE sequence for the T-stage was 0.778, and the accuracy of the DL TSE sequence for the T-stage was 0.907 (Tables 4-6).

Pathologic diagnosis of the N0 stage included 19 cases, the N1 stage included 21 cases, and the N2 stage included 14 cases. Conv TSE correctly diagnosed the staging of 38 cases, incorrectly diagnosed 16 cases, 10 overestimated, and 6 underestimated. DL TSE correctly diagnosed 50 cases, incorrectly diagnosed 4 cases, 1 overestimated, and 3 underestimated. The diagnostic conformity rate of conv TSE sequence N staging was 0.703, and the diagnostic conformity rate of DL TSE sequence N was 0.926 (Tables 4-6).

McNemar’s test and Kappa test analyses compared the TN staging of patients with conv TSE sequences and DL TSE sequences with TN staging confirmed by colonoscopy pathological analysis.

In the T-stage assessment, McNemar’s test showed that for conv TSE (χ² =0.667, P=0.717) and DL TSE (χ² =5.000, P=0.082), there was no statistically significant difference between the staging results (correct or incorrect) and the pathological gold standard. In the N-stage assessment, McNemar’s test indicated that for conv TSE (χ² =2.486, P=0.478) and DL TSE (χ² =4.000, P=0.261), there was no statistically significant difference between the staging results and the pathological gold standard. The results suggest that neither conv TSE nor DL TSE showed a statistically significant difference in their staging results compared to the pathological gold standard (all P>0.05) (Table 7).

Furthermore, we analyzed the consistency between the staging results of Conv TSE and DL TSE sequences (Table 8). The Kappa test showed that there was a moderate consistency between the two in T staging (Kappa =0.591, P<0.001), and a weak to moderate consistency in N staging (Kappa =0.413, P<0.001). This suggests that the DL TSE sequence produced different staging judgments from the conventional sequence, and this difference was statistically significant.

This result indicates that the DL TSE sequence has a numerical improvement in accuracy. Compared with the conv TSE sequence, the DL TSE sequence has a significant advantage in the accuracy of N staging diagnosis.

Discussion

MRI, with its high soft tissue resolution, has become a critical approach for assessing TN staging of tumors. Conv TSE sequences, especially high-resolution T2WI and T1WI TSE sequences, are crucial in TN staging of rectal cancer. This study aimed to compare DL TSE with conv TSE regarding TN staging and the acquisition time of T2WI and T1WI in rectal cancer.

Clinical value and limitations of TSE sequences in TN staging of rectal cancer

With its high SNR and high spatial resolution as the core sequence for MRI evaluation of rectal cancer, conv TSE sequences have significant value in TN staging (17). In T staging, the conv TSE sequence can clearly show the structure of each layer of the rectal wall, especially the continuity of the intrinsic muscular layer, which helps radiologists accurately determine the depth of tumor infiltration. In N staging, the conv TSE sequence can indicate the morphology and size of the lymph nodes in the pelvis, offering an important foundation for clinicians to determine whether the lymph nodes are metastatic.

Although the results of this study show that the TN staging accuracy of the conv TSE sequence is relatively high, it is slightly lower than that of the DL TSE sequence. This accuracy rate is very important in clinical practice and can provide more accurate treatment plans for patients, enhancing treatment outcomes and survival rates. Sikkenk et al. [2024] investigated the diagnostic performance of primary colon cancer by MRI before resection of the tumor without neoadjuvant therapy, using histopathology as the reference standard. It was found that for differentiating T1–2 from T3–4 diseases, the positive predictive value (PPV) was 64.8% [95% confidence interval (CI): 52.9–75.5%], and the negative predictive value (NPV) was 88.9% (95% CI: 82.7–93.7%). The conclusion was drawn that MRI had the strongest predictive performance for T1–2 and T4 diseases (18). Liu et al. [2024] evaluated the predictive ability of the combination of radiology based on nuclear magnetic resonance and tumor markers to predict the TN stage of patients with rectal cancer. This study found that radiological characteristics based on nuclear magnetic resonance were independent predictors of T stage and N stage. When combined with tumor markers, they have a good predictive effect on the TN stage of rectal cancer (19).

Although conv TSE sequences have significant application value in TN staging of rectal cancer, their limitations are clear. In T staging, the conv TSE sequence has limited sensitivity to deep infiltration of the rectal wall, and it is difficult to distinguish whether the tumor has penetrated the muscularis propria due to partial volume effects or motion artifacts, particularly because it is challenging to accurately determine the infiltration between the muscularis propria and the muscularis propria layer.

A lower SNR limits the contrast of bowel wall stratification and blurs the demarcation between submucosal and muscular layers. The rectal-air interface is prone to magnetic susceptibility artifacts, which also blur this demarcation. This may lead to over- or under-estimation of T-staging, which impacts treatment planning and patient prognosis. Moreover, due to perirectal tissue complexity, conv TSE sequences present difficulties in distinguishing boundaries between tumors and the surrounding tissues.

In terms of N staging, the conv TSE sequence is less sensitive to small lymph nodes (below 5 mm in diameter), meaning that many potential lymph node metastases may be overlooked, affecting the accuracy and efficacy of treatment. In contrast, the conv TSE sequence relies on morphological criteria (e.g., short diameter growth, margin irregularity) to characterize metastatic lymph nodes (14). However, metastatic lymph nodes with a short diameter of <5 mm cannot be reliably identified by traditional morphological criteria. Lymph node enlargement caused by reactive hyperplasia or inflammation may be difficult to differentiate between necrosis and fibrosis due to low SNR, and may be misclassified as metastasis. As these morphologic criteria are influenced by image quality, they are not reliable. This increases the uncertainty of N staging and the chances of misdiagnosis (20,21).

Technical breakthroughs and advantages of DL reconstruction

To address the limitations of conv TSE sequences in TN staging of rectal cancer, DL reconstruction offers the possibility of technological innovation. The DL reconstruction in this study significantly enhances the accuracy of TN staging by integrating multi-sequence MRI features (T2WI, T1WI) (11).

In T staging, DL reconstruction can more accurately characterize the depth of tumor infiltration by analyzing the differences between the tumor and rectal wall layers. A convolutional neural network (CNN) is applied to eliminate noise, improve the SNR, and increase the contrast of the bowel wall. The stratification-DL-assisted fat suppression algorithm reduces the signal interference due to the inhomogeneity of the magnetic field and improves the contrast between peri-bowel fat and tumor infiltration. DL TSE sequence improves the identification of micro-infiltration and muscularis propria invasion compared to the conv TSE sequence, reducing the risk of underestimation of T staging, improving the accuracy and effectiveness of the treatment plan, and providing a more reliable basis for the development of the surgical plan (e.g., the need for neoadjuvant therapy) and the prognostic assessment. The accuracy of identifying the T staging by DL TSE reaches 0.907 (95% CI: 0.639–0.989), an improvement of approximately 12.9% over the conv TSE sequence (Table 6).

Regarding N staging, given the central prognostic role of nodal status in rectal cancer, multiple technology-enabled strategies—including refined imaging workflows and adjunct approaches to nodal assessment—are being explored to reduce understaging (22), the DL TSE sequence improves image quality by motion correction, reducing noise to increase the SNR and spatial resolution. It can utilize multi-dimensional features such as morphology, size, density, and other characteristics of lymph nodes to make comprehensive judgments, improving the sensitivity and specificity of microscopic lymph nodes. The DL TSE sequence enhances the detection rate of microscopic metastatic lymph nodes compared to the conv TSE sequence and reduces the chances of N staging underestimation. The experimental data showed that the sensitivity of DL TSE for N staging reached 93.4% (50/54), which was 17.2% (P<0.001) higher than that of 76.2% (38/54) of conv TSE. In terms of N staging accuracy, the accuracy of DL TSE was 92.6%, which was 22.3% higher than that of 70.3% of conv TSE. Notably, DL TSE lowered the N staging underestimation rate from 11.1% (6/54) to 5.6% (3/54) with conv TSE. This assists radiologists in detecting potential lymph node metastases earlier, taking timely and effective interventions, and reducing the misclassification of inflammatory lymph nodes to avoid over-treatment. The identification accuracy of the DL TSE sequence was improved to 0.926 (compared with 0.703 for conv TSE sequence), especially for micrometastatic lymph nodes with a short diameter of 3–5 mm. This is consistent with recent research trends, as Zheng et al. (23) found that a DL model based on ResNet50 could effectively differentiate between T2 and T3 stages by capturing microscopic heterogeneity of the tumor-membrane fat interface. Across minimally invasive care pathways, technological advances in image guidance and algorithmic support are increasingly used to improve procedural accuracy and efficiency, highlighting the clinical value of methods that enhance image quality while reducing acquisition time (24).

Rectal cancer patients often move due to bowel movement, respiratory motion, or pain, and the conv TSE sequence is susceptible to motion artifacts due to prolonged scanning. Compared to the conv TSE sequence, the DL TSE sequence achieves improved scanning speed without losing any key diagnostic information through the synergistic optimization of DL-driven noise suppression mechanism and efficient high-frequency parallel acquisition and reconstruction strategy. In this study, the scanning time of the DL TSE sequence was 6 minutes 31 seconds, and the scanning time of the conv TSE sequence was 10 minutes 17 seconds, differing by 36.6% (12). This technological innovation dramatically shortens the scanning time and ensures the stability of the image quality and diagnostic accuracy by effectively suppressing noise through the DL algorithm. The application of high-frequency parallel acquisition and reconstruction technology and compression further improves the acceleration factor, making the scanning process more rapid and efficient. Scanning time is reduced, and the probability of patient movement is reduced, producing clearer images. This finding is consistent with recent research results, which indicate that DL-based accelerated reconstruction techniques can significantly shorten the acquisition time while improving image quality (25). This reduces the underestimation of the T-stage due to the blurring of rectal wall stratification and limits the risk of N-stage underestimation by making the morphology and borders of microscopic lymph nodes more easily recognizable. Additionally, it optimizes the clinical workflow and improves diagnosis and treatment efficiency. The results of this study are highly consistent with the current cutting-edge paradigm in the diagnosis and treatment of rectal tumors, which is to conduct rigorous clinical validation and evaluation of artificial intelligence (AI)-based systems under the framework of technology-assisted decision-making [AI-based decision support system (AI-DSS)], with the aim of continuously improving diagnostic efficacy and promoting the standardization of the diagnosis and treatment process.

Limitations

This study has some limitations. First, the primary limitation lies in the uneven distribution of T-stages within the sample cohort, coupled with a relatively small sample size. This is particularly evident in cases of localized rectal cancer, where the number of T1 and T2 stage cases is notably limited (only 1 case for T1 and 6 cases for T2). This limitation may compromise the statistical power for evaluating the performance of DL TSE in discriminating early-stage invasion depth and could restrict the generalizability of the study findings. Second, the inclusion criteria may have produced selection bias, as patients with smaller, fewer observable swellings were excluded. Third, the single-center 3T non-contrast imaging limits insights into broader field strengths or contrast-enhanced protocols. As with other image-guided and algorithm-supported technologies, multicenter validation across heterogeneous hardware and acquisition protocols is essential before broad clinical adoption. Although DL TSE T1WI was included in this study, its application of contrast-enhanced T1WI remains to be assessed.

Conclusions

This study demonstrates that DL TSE MRI for rectal cancer TN staging, including T1WI and T2WI, improves diagnostic accuracy and shortens scan time by 36.6% compared to conv TSE. The DL accelerated protocol enhances scanner efficiency and patient throughput, affording higher patient comfort and improving rectal TN staging by minimizing motion artifacts. Future studies could explore multi-center validation across different MRI scanners alongside evaluation in larger cohorts to confirm DL reconstruction in rectal cancer MRI clinical utility, supporting its integration into typical clinical practice.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the STARD reporting checklist. Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1884/rc

Data Sharing Statement: Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1884/dss

Funding: This research was supported by the National Natural Science Foundation of China (grant No. 82402229) and CAMS Innovation Fund for Medical Sciences (CIFMS) (grant No. 2024-I2M-C&T-A-007).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1884/coif). Y.J. and N.M. are current employees of Siemens Healthineers. The other authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of the Cancer Hospital, the Chinese Academy of Medical Sciences (approval No. NCC4435). Informed consent was taken from all the patients.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Siegel RL, Miller KD, Fuchs HE, Jemal A. Cancer statistics, 2022. CA Cancer J Clin 2022;72:7-33. [Crossref] [PubMed]
2. Siegel RL, Miller KD, Wagle NS, Jemal A. Cancer statistics, 2023. CA Cancer J Clin 2023;73:17-48. [Crossref] [PubMed]
3. Siegel RL, Giaquinto AN, Jemal A. Cancer statistics, 2024. CA Cancer J Clin 2024;74:12-49. [Crossref] [PubMed]
4. Glynne-Jones R, Wyrwicz L, Tiret E, Brown G, Rödel C, Cervantes A, Arnold D. Rectal cancer: ESMO Clinical Practice Guidelines for diagnosis, treatment and follow-up. Ann Oncol 2017;28:iv22-40. [Crossref] [PubMed]
5. Al-Sukhni E, Milot L, Fruitman M, Beyene J, Victor JC, Schmocker S, Brown G, McLeod R, Kennedy E. Diagnostic accuracy of MRI for assessment of T category, lymph node metastases, and circumferential resection margin involvement in patients with rectal cancer: a systematic review and meta-analysis. Ann Surg Oncol 2012;19:2212-23. [Crossref] [PubMed]
6. Benson AB, Venook AP, Al-Hawary MM, Azad N, Chen YJ, Ciombor KK, et al. Rectal Cancer, Version 2.2022, NCCN Clinical Practice Guidelines in Oncology. J Natl Compr Canc Netw 2022;20:1139-67. [Crossref] [PubMed]
7. Tsai KY, You JF, Huang SH, Tsai TY, Hsieh PS, Lai CC, Tsai WS, Hung HY. Comparison of clinical outcomes of stoma reversal during versus after chemotherapy for rectal cancer patients. Langenbecks Arch Surg 2023;408:274. [Crossref] [PubMed]
8. Wang G, Li J, Huang Y, Guo Y. A dynamic nomogram for predicting pathologic complete response to neoadjuvant chemotherapy in locally advanced rectal cancer. Cancer Med 2024;13:e7251. [Crossref] [PubMed]
9. Beets-Tan RGH, Lambregts DMJ, Maas M, Bipat S, Barbaro B, Curvo-Semedo L, Fenlon HM, Gollub MJ, Gourtsoyianni S, Halligan S, Hoeffel C, Kim SH, Laghi A, Maier A, Rafaelsen SR, Stoker J, Taylor SA, Torkzad MR, Blomqvist L. Correction to: Magnetic resonance imaging for clinical management of rectal cancer: Updated recommendations from the 2016 European Society of Gastrointestinal and Abdominal Radiology (ESGAR) consensus meeting. Eur Radiol 2018;28:2711. [Crossref] [PubMed]
10. Lustig M, Pauly JM. SPIRiT: iterative self-consistent parallel imaging reconstruction from arbitrary k-space. Magn Reson Med 2010;64:457-71. [Crossref] [PubMed]
11. Hammernik K, Klatzer T, Kobler E, Recht MP, Sodickson DK, Pock T, Knoll F. Learning a variational network for reconstruction of accelerated MRI data. Magn Reson Med 2018;79:3055-71. [Crossref] [PubMed]
12. Hu S, Fan W, Teng Z, Liu K, Tong X, Jiang Y, Liu P, Lang Y, Nickel MD, Zhang H. Study on the value of deep reconstruction technique in improving the image quality of magnetic resonance rectal cancer. Chin J Magn Reson Imaging 2024;15:30-5.
13. Tamada D, Kromrey ML, Ichikawa S, Onishi H, Motosugi U. Motion Artifact Reduction Using a Convolutional Neural Network for Dynamic Contrast Enhanced MR Imaging of the Liver. Magn Reson Med Sci 2020;19:64-76. [Crossref] [PubMed]
14. Brown G, Richards CJ, Bourne MW, Newcombe RG, Radcliffe AG, Dallimore NS, Williams GT. Morphologic predictors of lymph node status in rectal cancer with use of high-spatial-resolution MR imaging with histopathologic comparison. Radiology 2003;227:371-7. [Crossref] [PubMed]
15. Karamchandani DM, Gonzalez RS, Lee H, Westerhoff M, Cox B, Pai RK. Interobserver agreement and practice patterns for grading of colorectal carcinoma: World Health Organization (WHO) classification of tumours 5th edition versus American Joint Committee on Cancer (AJCC) staging manual. Histopathology 2025;86:1101-11.
16. McHugh ML. Interrater reliability: the kappa statistic. Biochem Med (Zagreb) 2012;22:276-82.
17. Diagnostic accuracy of preoperative magnetic resonance imaging in predicting curative resection of rectal cancer: prospective observational study. BMJ 2006;333:779. [Crossref] [PubMed]
18. Sikkenk DJ, Henskens IJ, van de Laar B, Burghgraef TA, da Costa DW, Somers I, Verheijen PM, Nederend J, Nagengast WB, Tanis PJ, Consten ECJ. Diagnostic Performance of MRI and FDG PET/CT for Preoperative Locoregional Staging of Colon Cancer: Systematic Review and Meta-Analysis. AJR Am J Roentgenol 2024;223:e2431440. [Crossref] [PubMed]
19. Liu Z, Zhang J, Wang H, Chen X, Song J, Xu D, Li J, Zheng M. MRI-based radiomics feature combined with tumor markers to predict TN staging of rectal cancer. J Robot Surg 2024;18:229. [Crossref] [PubMed]
20. Taylor FG, Quirke P, Heald RJ, Moran B, Blomqvist L, Swift I, Sebag-Montefiore DJ, Tekkis P, Brown G. Preoperative high-resolution magnetic resonance imaging can identify good prognosis stage I, II, and III rectal cancer best managed by surgery alone: a prospective, multicenter, European study. Ann Surg 2011;253:711-9. [Crossref] [PubMed]
21. Horvat N, Veeraraghavan H, Khan M, Blazic I, Zheng J, Capanu M, Sala E, Garcia-Aguilar J, Gollub MJ, Petkovska I. MR Imaging of Rectal Cancer: Radiomics Analysis to Assess Treatment Response after Neoadjuvant Therapy. Radiology 2018;287:833-43. [Crossref] [PubMed]
22. Boland PA, McEntee PD, Cucek J, Erzen S, Niemiec E, Galligan M, Petropoulou T, Burke JB, Knol J, Hompes R, Tuynman J, Aigner F, Arezzo A, Cahill RA. Protocol for CLASSICA software as medical device trial. Minim Invasive Ther Allied Technol 2025;34:441-6. [Crossref] [PubMed]
23. Zheng X, Yao Z, Huang Y, Yu Y, Wang Y, Liu Y, Mao R, Li F, Xiao Y, Wang Y, Hu Y, Yu J, Zhou J. Deep learning radiomics can predict axillary lymph node status in early-stage breast cancer. Nat Commun 2020;11:1236. [Crossref] [PubMed]
24. Boretto L, Pelanis E, Regensburger A, Fretland ÅA, Edwin B, Elle OJ. Hybrid optical-vision tracking in laparoscopy: accuracy of navigation and ultrasound reconstruction. Minim Invasive Ther Allied Technol 2024;33:176-83. [Crossref] [PubMed]
25. Jurka M, Macova I, Wagnerova M, Capoun O, Jakubicek R, Ourednicek P, Lambert L, Burgetova A. Deep-learning-based reconstruction of T2-weighted magnetic resonance imaging of the prostate accelerated by compressed sensing provides improved image quality at half the acquisition time. Quant Imaging Med Surg 2024;14:3534-43. [Crossref] [PubMed]