Determinants of magnetic resonance imaging overstaging in rectal cancer T-staging

Abstract

BACKGROUND

The front-end of an efficient clinical pathway for rectal cancer is a high-precision local-stage examination that determines the necessity of surgery and whether neoadjuvant therapy should be given. Currently, in cancer treatment, according to the extent of tumour spread, a decision will be made among local excision, total mesorectal excision, and neoadjuvant chemoradiotherapy. As management has moved away from a general radical procedure to personalised organ-preserving and neoadjuvant-intensive frameworks, the clinical relevance of this assessment has been increasing. Incorrect staging is related to a high risk of morbidity; that is, an overestimation of the stage may lead to unnecessary toxicity or extensive surgery, and an underestimation of the stage risks failing to treat early and progressing to local recurrence. Therefore, enhancing the predictive accuracy of preoperative T classification needs to be done to improve surgical quality.

AIM

To explore the factors influencing the accuracy of preoperative magnetic resonance imaging (MRI) in rectal cancer T staging, with particular emphasis on factors contributing to MRI overstaging in routine clinical practice.

METHODS

This retrospective single-center study included 349 patients with rectal cancer, including carcinoma in situ, who underwent surgery at Beijing Anzhen Nanchong Hospital, Capital Medical University and Nanchong Central Hospital between January 2020 and December 2023. All patients underwent preoperative MRI-based tumor-node-metastasis staging assessment and had postoperative pathological confirmation. Univariate and multivariable logistic regression analyses were performed to identify factors associated with MRI overstaging. Reader-related variables, including experience, subspecialty background, specialized training, reading duration, blinded reading, and multidisciplinary team participation, were also evaluated.

RESULTS

The overall accuracy of preoperative MRI T staging was 58.5% (204/349), with overstaging in 35.5% (124/349) and understaging in 6.0% (21/349). MRI accuracy by T stage (T1-T4) was 25.0%, 29.4%, 80.6%, and 83.6%, respectively. Multivariate analysis revealed that a relatively small tumor size, neoadjuvant treatment-related regression, and polypoid morphology were independently associated with MRI overstaging. Reader-related analyses revealed that limited experience, a general radiology background, a lack of specialized training, shorter interpretation time, and the absence of multidisciplinary team participation were associated with higher overstaging rates.

CONCLUSION

A new size and shape of the tumour have appeared in the following observations after treatment, so further imaging studies were recommended. Small tumors with a polypoid structure or pronounced post-neoadjuvant regression may be less likely to be overclassified radiographically, thus avoiding overstaging. Add endorectal ultrasonography, standardised reporting forms and cross-disciplinary evaluation to improve the diagnostic accuracy for vulnerable groups.

Key Words: Magnetic resonance imaging; Rectal cancer; T staging; Overstaging; Reader-related factors

Core Tip: Current data indicate a propensity for magnetic resonance imaging to overestimate the T category of rectal tumors, particularly when evaluating diminutive lesions, polypoid morphologies, or cases demonstrating pronounced post-neoadjuvant regression. Diagnostic precision was concurrently compromised by several interpreter variables; restricted clinical tenure, absent structured training, abbreviated review durations, and solitary reading without multidisciplinary team input all exacerbated staging inaccuracies. Consequently, such observations advocate for adopting a risk-stratified clinical paradigm. For these vulnerable scenarios, standard magnetic resonance imaging assessments require augmentation through endorectal ultrasonography alongside cross-specialty panel consensus.

INTRODUCTION

A reasonable clinical pathway for rectal cancer should conduct early regional lymph node assessment and adjust the type of treatment according to the stage of the disease; that is, prompt neoadjuvant therapy or operation may be adopted earlier. The three choices for contemporary oncology patients are local excision, total mesorectal excision and preoperative chemoradiotherapy, and their selection depends on the initial evaluation of tumor invasion depth[1-3]. As the direction of treatment has shifted from broad-based radical procedures to personalised, organ-preserving and neoadjuvant-intensive strategies, so too has the demand for this assessment[3-5]. An incorrect stage may be serious for the patient; too high an estimate leads to unnecessary toxicity or broad resection, and too low a stage results in insufficient treatment and local recurrence[6,7]. Therefore, the accuracy of predicting preoperative T classification and thus the quality of surgery are still being improved.

Pelvic magnetic resonance imaging (MRI) has good soft-tissue contrast and is used to observe the layers of the rectal wall, the mesorectal envelope, and nearby neurovascular structures preoperatively; therefore, it is now a general tool for pre-operative investigation[3,8,9]. Based on clinical practice recommendations, MRI should be used as the initial diagnostic tool, and at the same time, digital examination and potentially endorectal ultrasound can be combined if an early-onset disease is suspected[1-3]. Critically, MRI can also be used to evaluate the circumferential resection margin and obtain an anatomical map for surgery[3,8,9].

Although it can diagnose diseases more accurately, there are still issues with the reliability of MRI T-staging; specifically, it is less precise for shallow lesions and in post-treatment re-staging. According to the institution’s data, the difference between T1 and T2 categories is not yet obvious, and this has also been exacerbated by the various forms of the main tumour[4,9-14]. Such diagnostic ambiguity is often caused by biological mimics, such as peritumoral fibrosis, desmoplastic reactions, inflammatory oedema and treatment-induced architectural alterations, all of which can erroneously indicate tumor extension beyond the muscularis propria[13-17]. Interpretive difficulties are further exacerbated after neoadjuvant therapy, and it is now difficult to distinguish between residual viable malignancy and therapeutic scarring in the clinic[5,11,15,16,18-23].

To address the above deficiencies, adjunctively, endorectal ultrasound and high-sensitivity contrast-enhanced imaging techniques have been applied to some extent, as shown in[24-26]. Recent research has also shown that the quality of diagnosis is not solely determined by technical indices but is also influenced by “human factors”, such as the radiologist’s subspecialty, case volume and participation in a multidisciplinary team (MDT) structure[6,27,28]. Specific educational interventions have increased interpretation confidence, and multiple disciplines’ agreement has improved the consistency between radiological impressions and the final pathological results[27,28].

However, the particular reasons for MRI overstaging in routine practice have not yet been fully identified. Most previous research has focused on improving the overall accuracy of diagnosis or post-neoadjuvant re-staging, and has not yet explored how tumor shape affects different readers’ reports[10-11,13-15,18]. Practical uncertainties still exist in the interpretation of diminutive tumors, the risk of misclassification for polypoid lesions, and boundary obscuration due to post-regressive fibrosis[13-15,18,29].

Therefore, this retrospective single-center study aimed to investigate the determinants of preoperative MRI T-stage inaccuracies in patients with rectal cancer or rectal carcinoma in situ who underwent surgery with pathological confirmation. We focused specifically on factors associated with MRI overstaging and examined them across tumor-related, treatment-related, and reader-related domains. By identifying the cases most vulnerable to overstaging, we seek to support risk stratification, the selective use of adjunctive modalities such as endorectal ultrasound, and quality-improvement efforts in rectal MRI interpretation.

MATERIALS AND METHODS

Study design and patient selection

This retrospective single-center study was conducted at Beijing Anzhen Nanchong Hospital of Capital Medical University and Nanchong Central Hospital. Clinical data were collected from patients diagnosed with rectal cancer or rectal carcinoma in situ who underwent surgical treatment between January 2020 and December 2023 in accordance with the 10^th Revision of the International Classification of Diseases.

The inclusion criteria: (1) Primary diagnosis of rectal malignant neoplasm or rectal carcinoma in situ; (2) Surgical treatment with postoperative pathological confirmation, including radical resection for invasive cancer and radical or local resection, including transanal local excision and transanal minimally invasive surgery, for carcinoma in situ; and (3) Preoperative tumor-node-metastasis staging using pelvic MRI.

The exclusion criteria: (1) Preoperative staging performed primarily by endorectal ultrasound or contrast-enhanced computed tomography rather than MRI; (2) Salvage radical surgery after prior local excision because of high-risk pathological findings or recurrent rectal cancer; and (3) Pathological diagnosis of neuroendocrine tumors, melanoma, carcinoid, or other non-adenocarcinoma malignancies.

Because postoperative pathological T stage served as the reference standard, only surgically treated patients were eligible for inclusion.

MRI protocol

Preoperative pelvic MRI was routinely performed using a 3.0-T scanner at our institution. No endorectal coil was used. Patients underwent bowel preparation on the evening before the examination. Ten minutes prior to scanning, 20 mg of anisodamine was administered intramuscularly, and diluted ultrasound gel was transanally instilled to distend the rectum.

The institutional protocol included sagittal T2-weighted imaging (T2WI), oblique coronal T2WI, oblique axial T2WI, diffusion-weighted imaging, and contrast-enhanced imaging. Diffusion-weighted imaging was acquired with a b value of 1000 second/mm². Apparent diffusion coefficient maps were routinely generated, although apparent diffusion coefficient-derived parameters were not included as independent variables in the present analysis. All T2WI sequences were obtained using a small field of view and thin-section acquisition according to the institutional rectal MRI protocol.

MRI interpretation and T-staging criteria

MRI staging was based on the original preoperative MRI interpretation generated during routine clinical care before surgery and before pathological diagnosis. Therefore, radiologists were unaware of the final pathological T stage at the time of MRI interpretation.

For tumors with an inferior margin located ≥ 5 cm from the anal verge, the American Joint Committee on Cancer/Union for International Cancer Control tumor-node-metastasis staging criteria were applied. For tumors with an inferior margin < 5 cm from the anal verge, a low rectal cancer MRI staging framework based on Battersby et al[8] was used. Preoperative MRI staging of Tis or T1 was considered accurate when matched by postoperative pathological staging.

Pathological assessment

The pathological T stage was determined from postoperative surgical specimens in accordance with standard pathological practice and served as the reference standard. Intraoperative findings were used as supplementary evidence when relevant. Because of the retrospective nature of the study and routine clinical workflow, the formal blinding of pathologists to the imaging information could not be fully guaranteed.

Definition of clinicopathological variables

“Polypoid morphology” was designated for tumors that showed a protruding intraluminal growth pattern in endoscopic or gross pathological examinations and were thus not classified as traditional ulcerative or infiltrative types. The maximum diameter of the tumour was taken from the larger of the two sizes in the image and pathology reports. Dichotomisation at 4 cm of the median distribution of the cohort was used for univariable screening of this index.

Reader-related data collection

Extract variables of the radiologist’s profile from the archived reporting logs. We classified “experience” by the number of years of post-certification practice as < 3 years, 3-10 years, and > 10 years. The two groups of readers also differed in their subspecialty (general vs abdominal imaging) and whether they had taken formal pedagogical courses on the staging of rectal cancer. We have added data on participation in a MDT and estimated the interpretation time per case (less than 10 minutes, 10-20 minutes, and over 20 minutes) based on workflow metadata.

Statistical analysis

SPSS 26.0 was used for the above analysis. Univariable screening was performed (using χ² or Fisher’s exact tests), and all variables with P < 0.10 were included in a multivariable logistic regression model to determine which were independent factors of stage overestimation. A test for multicollinearity was performed prior to model building. The upper and lower bounds of significance are set at 0.05.

RESULTS

Baseline characteristics

A total of 349 patients were included, comprising 216 males (61.9%) and 133 females (38.1%). The mean age was 64.3 ± 11.3 years, ranging from 20 years to 93 years. The mean body mass index was 23.2 ± 3.1 kg/m², ranging from 13.3 kg/m² to 34.4 kg/m². The mean maximum tumor diameter was 4.0 ± 1.7 cm, ranging from 1.0 cm to 9.5 cm. Among these patients, 15 had rectal carcinoma in situ, and 334 had rectal cancer.

Fifty-nine cases were classified as polypoid lesions. The smallest malignant polyp measured 1.5 cm, and the largest in situ polyp measured 9.0 cm. The smallest polypoid lesion classified as pathological T4 measured 4.5 cm. Positive circumferential resection margins were detected in 8 patients (2.3%). Ten patients underwent transanal local excision, and the remainder underwent radical resection.

Among 349 patients, MRI T staging was accurate in 204 cases (58.5%), overstaged in 124 cases (35.5%), and understaged in 21 cases (6.0%). The accuracy of MRI for T1-T4 stages was 25.0%, 29.4%, 80.6%, and 83.6%, respectively, indicating substantially lower performance in early-stage disease.

Univariate analysis of clinicopathological factors associated with inaccurate MRI T-staging

Univariate analysis revealed that maximum tumor diameter (P < 0.001), circumferential involvement (P < 0.001), neoadjuvant treatment (P = 0.040), and pathological type (P < 0.001) were associated with MRI overstaging (Table 1). Smaller tumors, more limited circumferential involvement, neoadjuvant treatment, and polypoid morphology were associated with higher rates of overstaging. None of the analyzed clinicopathological variables were significantly associated with understaging.

Table 1 Univariate analysis of factors affecting magnetic resonance imaging T-staging accuracy, n (%).

Variable	Cases	Understaging	χ2	P value	Overstaging	χ2	P value
Gender
Male	216	11 (5.1)	0.857	0.355	72 (33.3)	1.194	0.275
Female	133	10 (7.5)			52 (39.1)
Age (year)
< 60	123	6 (4.9)	0.436	0.509	39 (31.7)	1.212	0.271
≥ 60	226	15 (6.6)			85 (37.6)
BMI (kg/m2)
< 24	213	12 (5.6)	0.142	0.706	74 (34.7)	0.148	0.7
≥ 24	136	9 (6.6)			50 (36.8)
Distance from anal verge
≤ 5	174	7 (4.0)	4.16	0.1131	66 (37.9)	3.592	0.166
5-10 cm	138	13 (9.4)			50 (36.2)
≥ 10 cm	37	1 (2.7)			8 (21.6)
Maximum tumor diameter
< 4 cm	174	7 (4.0)	2.44	0.118	78 (44.8)	13.095	< 0.001
≥ 4 cm	175	14 (8.0)			46 (26.3)
Circumferential involvement
≤ 1/4	61	4 (6.6)	4.507	0.1951	30 (49.2)	18.69	< 0.001
> 1/4 - ≤ 1/2	149	8 (5.4)			60 (40.3)
> 1/2 - ≤ 3/4	77	2 (2.6)			25 (32.5)
> 3/4 - 1	62	7 (11.3)			9 (14.5)
Tumor location
Anterior	120	12 (10.0)	5.307	0.1351	39 (32.5)	4.096	0.255
Right	65	2 (3.1)			29 (44.6)
Posterior	99	3 (3.0)			37 (37.4)
Left	65	4 (6.2)			19 (29.2)
Neoadjuvant chemoradiotherapy
None	273	18 (6.6)	1.814	0.3781	89 (32.6)	6.34	0.04
Chemotherapy only	51	1 (2.0)			26 (51.0)
Chemoradiotherapy	25	2 (8.0)			9 (36.0)
Pathological type
Polyp carcinogenesis	59	1 (1.7)	1.516	0.2181	36 (61.0)	20.135	< 0.001
Primary carcinoma	290	20 (6.9)			88 (30.3)

¹Fisher’s exact test.

BMI: Body mass index.

Multivariable analysis of MRI overstaging

In the multivariable model, maximum tumor diameter, neoadjuvant treatment-related regression, and pathological type were independent factors associated with MRI overstaging (Table 2). Specifically, a smaller tumor size, marked regression after neoadjuvant therapy, and polypoid morphology were associated with an increased risk of overstaging.

Table 2 Multivariable logistic regression analysis of factors associated with magnetic resonance imaging, n (%).

Variable	Cases	Understaging	χ2	P value	Overstaging	χ2	P value
Reader experience			4.217	0.121	-	6.891	0.032
< 3 years	98	8 (8.2)			42 (42.9)
3-10 years	153	8 (8.2)			50 (32.7)
> 10 years	98	5 (5.1)			32 (32.7)
Specialty background			3.102	0.078	-	4.567	0.033
General radiologist	214	15 (7.0)			82 (38.3)
Abdominal imaging subspecialist	135	6 (4.4)			42 (31.1)
Receipt of specialized training			2.891	0.089	-	5.123	0.024
Yes	276	14 (5.1)			92 (33.3)
No	73	7 (9.6)			32 (43.8)
Reading duration (per case)			5.432	0.066	-	8.765	0.012
< 10 minutes	112	10 (8.9)			50 (44.6)
10-20 minutes	187	8 (4.3)			62 (33.2)
> 20 minutes	50	3 (6.0)			12 (24.0)
Blinded reading (without clinical information)			1.234	0.267	-	3.456	0.063
Yes	291	16 (5.5)			98 (33.7)
No	58	5 (8.6)			26 (44.8)
MDT participation			1.876	0.171	-	4.321	0.038
No (independent reading)	187	13 (7.0)			72 (38.5)
Yes	162	8 (4.9)			52 (32.1)

MDT: Multidisciplinary team.

Further analysis of neoadjuvant treatment revealed significant differences in overstaging between patients with a complete pathological response or tumor regression grade 1 and those with a tumor regression grade 2 or 3 (P = 0.031). No significant difference was detected between patients with a tumor regression grade of 2 or 3 and those who did not receive neoadjuvant therapy (P = 0.990).

Reader-related factors associated with inaccurate MRI T-staging

Univariate analysis of reader-related factors revealed several associations with MRI overstaging (Table 3). Reader experience was significantly associated with overstaging (χ² = 6.891; P = 0.032), with the least experienced group (< 3 years) showing the highest overstaging rate (42.9%). Compared with general radiologists (38.3%), abdominal imaging subspecialists had a lower overstaging rate (31.1%) (χ² = 4.567, P = 0.033). Radiologists who had received specialized training in rectal cancer MRI staging had a lower overstaging rate (33.3%) than those without such training did (43.8%) (χ² = 5.123, P = 0.024). Reading duration also significantly affected overstaging rates (χ² = 8.765; P = 0.012), with the shortest reading duration (< 10 minutes per case) associated with the highest overstaging rate (44.6%). Multidisciplinary team participation was associated with reduced overstaging (32.1% vs 38.5%, χ² = 4.321, P = 0.038). Blinded reading without clinical information showed a nonsignificant trend toward lower overstaging (33.7% vs 44.8%, χ² = 3.456, P = 0.063). None of these reader-related variables was significantly associated with understaging.

Table 3 Multivariate analysis of factors affecting magnetic resonance imaging T-staging accuracy overstaging.

Variable	Regression coefficient	OR (95%CI)	P value
Maximum tumor diameter	-0.197	0.821 (0.708-0.952)	0.009
Circumferential involvement	-0.057	0.945 (0.404-2.208)	0.105
Neoadjuvant treatment-related regression	-0.185	0.831 (0.338-2.045)	0.021
Pathological type, polypoid vs primary carcinoma	-1.546	0.213 (0.115-0.395)	< 0.001

CI: Confidence interval; OR: Odds ratio.

Analysis of high-risk subgroups

Among the polypoid lesions, the overstaging rate was 61.0% (36/59). Among tumors < 4 cm, the overstaging rate was 44.8% (78/174), whereas it was 26.3% (46/175) for tumors ≥ 4 cm. These findings further support the view that small tumors and polypoid lesions represent high-risk subgroups for MRI overstaging.

The accuracy of stage-specific MRI is summarized in Table 4. MRI showed substantially lower accuracy for early-stage tumors, with accuracy rates of 25.0% for T1 and 29.4% for T2 disease, compared with 80.6% for T3 and 83.6% for T4 disease. High-risk subgroups for MRI overstaging are summarized in Table 5.

Table 4 Cross-tabulation: Magnetic resonance imaging T stage vs pathological T stage (n = 349).

Pathological T stage	mrTis/T1	mrT2	mrT3	mrT4	Total
pTis/ypT0	16	6	0	0	22
pT1	11	25	8	0	44
pT2	2	16	28	4	50
pT3	0	5	38	10	53
pT4	0	0	4	31	35
Total	29	52	78	46	204

Table 5 Stage-specific diagnostic performance of magnetic resonance imaging (n = 349).

Category	Sensitivity	95%CI (Wilson)	Specificity	95%CI (Wilson)
T1	25.0%	14.5%-38.7%	94.1%	91.3%-96.2%
T2	29.4%	19.2%-41.5%	92.8%	90.2%-94.9%
T3	80.6%	72.6%-86.9%	89.6%	85.6%-92.7%
T4	83.6%	74.9%-90.1%	97.3%	95.3%-98.6%

DISCUSSION

In this retrospective single-center study, we evaluated the determinants of preoperative MRI overstaging in patients with rectal cancer in a real-world clinical setting. The principal findings were that a smaller tumor size, marked regression after neoadjuvant therapy, and polypoid morphology were independently associated with MRI overstaging. In addition, several reader-related factors, including limited experience, a lack of abdominal imaging subspecialization, the absence of dedicated training, a shorter reading duration, and a lack of multidisciplinary team participation, were associated with higher overstaging rates according to univariate analysis. These results emphasize that MRI overstaging is influenced not only by tumor biology and treatment effects but also by interpretation-related conditions in routine practice.

Our findings regarding small tumors are consistent with those of previous reports showing limited MRI accuracy for early-stage rectal cancer[4,9,13,14]. In a large population-based study of MRI staging in early rectal cancer, Detering et al[4] reported that routine MRI staging showed substantial limitations in distinguishing early T categories, particularly cT1-2 disease[4]. Small lesions may be difficult to clearly delineate, and subtle peritumoral fibrosis, desmoplastic reactions, vascular changes, or inflammatory infiltration may be mistaken for extension beyond the muscularis propria[13,17]. This may explain why MRI performance was poor in T1-T2 tumors in our cohort but substantially better in T3-T4 disease. Similarly, limited circumferential involvement was associated with higher overstaging in univariate analysis, likely reflecting overlap with a smaller lesion size and earlier-stage morphology.

Marked regression after neoadjuvant therapy has also emerged as an important source of overstaging. In this setting, MRI interpretation is complicated by treatment response zones containing edema, fibrosis, necrosis, and acellular mucin, which may mimic residual tumors[15,16]. Our subgroup finding that overstaging differed significantly between patients with a complete pathological response or tumor regression grade 1 and those with a tumor regression grade 2 or 3 further supports the concept that the degree of regression, rather than neoadjuvant treatment per se, may drive overstaging in selected patients. Based on the above, no one has yet developed a high-precision, earlier-diagnostic method for pathological re-staging via MRI. A typical case of such a deficiency is the inability to distinguish between minimal residual disease and morphological alterations induced by therapy[18-20].

Polypoid architecture was also a category of patients with a high clinical risk. Biologically, such tumours grow in the lumen and are generally convex, not invading the underlying wall at this time. This structure is often too small and close to the surface on a radiograph, so it may be mistaken for an actual transmural spread and thus overdiagnosed. Cai and others have also reported that tumour morphology and other intrinsic factors of a rectal carcinoma are the main reasons for errors in preoperative MRI staging[14]. Among the subjects in our sample, 61.0% showed overestimation of polypoid variants. Therefore, an extended radiological investigation will be performed before a doctor decides to choose minimally invasive local excision surgery.

Another feature of the present study is that it will also be reader-dependent. Practitioners with little experience had a high rate of overstaging, possibly because it was difficult to distinguish genuine spread of cancer from fibrotic, oedematous, inflammatory or desmoplastic changes. Abdominal specialists and those with formal training in rectal MRI were more accurate, on the other hand. Based on the above results, specific skill-level interventions and standardised training will be implemented to prevent diagnostic errors in advance. Recent literature also supports the above view and indicates that special educational models and practical workshops have increased both the certainty and staging proficiency of examiners[26,27]. Abbreviated interpretation sessions also showed an increase in overestimation; therefore, an extended review interval may be necessary for the assessment of subtle early-stage or post-treatment abnormalities. Engagement in a multidisciplinary team was negatively correlated with over-reporting. Regularly discuss diagnostic standards and uniform reporting norms among surgical, pathological and oncological experts. Bedrikovetski and others have come to the same conclusion in a forward-looking study that interdisciplinary groups can enhance local pre-operation diagnosis for colorectal cancer[28].

Table 3 is a compilation of the main high-risk categories and interpreter-linked items for better transparency and practicality. As shown in the table above, radiologic overestimation is not random; it occurs in certain biological and procedural conditions of these groups. The diagnostic trigger scenarios include small-tumour dimensions, polypoid architecture, limited circumferential spread, novice interpretation profiles, educational deficiencies, hasty image examination and isolation from multidisciplinary consultations. Based on the above observations, previous research has shown that it is difficult to accurately stage early-stage tumors using MRI alone, and structured training and multidisciplinary consultation may help improve staging accuracy[4,14,26-28].

Based on the above data, our studies propose the following practical risk-adjustment measures. Patients with any of the following characteristics - a small tumor size, especially less than 4 cm; a polypoid lesion morphology; or significant post-neoadjuvant regression - should be regarded as being at a higher risk of overstaging and the MRI-based T-stage assessment should be interpreted with caution. Under these circumstances, supplementary endorectal ultrasound and organised radiological review should be employed, and multidisciplinary team discussions need to be held, especially if the choice of treatment differs significantly based on whether it is a very early or relatively advanced local stage, or if organ-preserving strategies are being considered.

The study has the following deficiencies. First, it was a retrospective single-center study, and therefore may not be generalizable to institutions with different patient populations, scanners, protocols and radiologist expertise. The second is that the diagnostic standard is postoperative histopathology, and therefore only surgical candidates were included in our study. This restriction causes selection bias and thus underestimates the number of people selected for non-operative management or organ-sparing treatment. At the same time, although institutional MRI acquisition protocols were kept consistent, these single-center studies may not be directly applicable to institutions with different technical arrangements or varying reporting standards.

The clinical workflow is used to obtain the impression in the preoperative documents. Although this way was able to have radiologists assess the images without knowing the results of a biopsy, a retrospective study could not ensure that the pathologists were also blind to the radiologists’ previous evaluations. The extraction method also had the problem of reader-specific measures. Scan interpretation times were obtained from administrative logs or manual collection and thus had a risk of estimation errors; automated PACS timestamps were not used. General practice years are not ideal indicators of experience; therefore, they fail to reflect the actual number of dedicated rectal MRI cases a radiologist has handled.

Analysis of the reader’s variables was only carried out in a univariate screening step. We have not constructed an extensive multi-variable model due to the limitation of retrospective data and the anticipated collinearity among the related factors, such as professional years, subspecialty concentration, previous education, and participation in a multidisciplinary team. No formal assessment of interobserver reliability was performed at the end. To confirm the above results, prospective multi-cohort studies will be planned that are thoroughly blinded, use consistent imaging parameters, and have pre-defined reader evaluation standards.

CONCLUSION

Differences in the size and shape of the tumour, changes in response to treatment after surgery, or other features not explicitly mentioned all correlated with the extent of tumour invasion shown in MRI images. When interpreting imaging results of cases with small tumours, polypoid enlargement or significant changes caused by treatment, greater care should be taken as they are more likely to be artefacts. Reader-related factors, such as experience and specialised training, are also likely to increase the risk of overdiagnosis, as well as limited interpretation time and participation in the MDT. Supplementary endorectal ultrasound, standardised radiological assessment and multidisciplinary consultation can be used to enhance the accuracy of stage determination for high-risk cases.