Introduction

Colorectal cancer (CRC) is one of the most common digestive tumors and ranks the third cause in cancer mortality worldwide [1]. Rectal cancer (RC) accounts for approximately one-third of all CRC cases [2]. There is an increasing attention to the diagnosis, treatment, and prognosis of patients with RC. The appropriate treatment plan depends on the tumor-node-metastasis (TNM) staging system. However, patients with the same RC stage at initial diagnosis may have markedly different clinical outcomes [3,4,5]. Novel biomarkers are needed to explore and better stratify the clinical outcomes in RC patients.

Tumor budding (TB) presents as a single cell or a cell cluster of up to four tumor cells at the invasive margins of CRCs [6]. It is assessed by pathologists using the International Tumor Budding Consensus Conference (ITBCC) recommendations on hematoxylin and eosin (H&E) staining and scored by the budding count. TB is an independent adverse prognostic factor in CRC [7,8,9,10]. High-grade TB has been correlated with adverse clinicopathological features, including high TNM stage and poor overall and disease-free survival [6]. TB is included as a supplemental prognostic factor for CRC in the TNM (2017) and WHO (2019) classification schemes [11,12,13]，and as a recommended element in the College of American Pathologists and International Collaboration on Cancer Reporting protocols for CRC histopathology [14]. TB potentially affects clinical decision-making for patients with locally advanced rectal cancer (LARC). High-grade TB is an adverse prognostic factor and is an indication for adjuvant chemotherapy in patients with stage II RC [7, 15]. However, clinical characteristics and tumor stages are poor predictors of tumor response to neoadjuvant chemoradiotherapy and overall prognosis [16, 17]. Moreover, TB is an indicator of metastasis and indicates a lack of response to neoadjuvant therapy if detected in pre-treatment biopsies [18,19,20]. Management and treatment strategies could be tailored for LARC patients who would benefit from adjuvant therapy or who are not likely to exhibit a complete pathological tumor response to neoadjuvant chemoradiotherapy. LARC biopsies are traumatic, and samples usually yield a small amount of tumor material. Thus, preoperative biomarkers of TB are of immense importance and would provide a non-invasive approach for the evaluation of TB.

Magnetic resonance imaging (MRI) is a non-invasive imaging technique and is widely used to evaluate preoperative staging and treatment response in patients with RC [21,22,23,24]. However, conventional MR images primarily provide qualitatively diagnostic information of RC lesions rather than quantitative metrics that reflect clinicopathologic information. Diffusion-weighted imaging (DWI) could evaluate the microscopic mobility of water molecules in lesions, and the derived apparent diffusion coefficient (ADC) values can be quantitatively evaluated and used to diagnose rectal tumors and evaluate the treatment response. With the increased amplitude of b-values, DWI images and ADC values are more sensitive to tumors than to normal tissues [24,25,26]. However, DWI images acquired with very high b-values result in significant image distortion and lower signal-to-noise ratios (SNR) [27, 28].

b-value threshold (bThreshold) map, derived from DWI images, has been proposed as a novel diffusion contrast method [29]. bThreshold map provides improved lesion visualization for prostate, breast, and rectal tumors than conventional DWI images [23, 28] and improves the signal contrast between lesions and normal tissues. It might also compliment DWI and ADC in the evaluation of the pathologic features of RC [23].

The use of bThreshold map for the preoperative evaluation of TB has not been assessed. This study investigated the preoperative use of bThreshold maps to evaluate TB in LARC patients and to compare the diagnostic performance of bThreshold maps with ADC maps in patients with different TB grades.

Materials and methods

Patients

The present study was approved by the local Institutional Review Board (Committee on Ethics of Biomedicine, Changhai Hospital of Shanghai). Informed consent was waived for this retrospective study. Between January 2018 and December 2020, 113 patients with LARC who underwent rectal MRI before surgical resection were considered for the study. Inclusion criteria were (1) pathologically confirmed rectal adenocarcinoma; (2) complete postoperative pathological data, including the TB grade; (3) and single focal lesion. The exclusion criteria were (1) any treatment (radiotherapy, chemotherapy, or chemoradiotherapy) prior to surgery; (2) more than 2 weeks between MRI and surgery; (3) poor quality DWI image; (4) simultaneous distant metastases; (5) and invasion of the circumferential resection margin. In all, 51 patients with LARC were included in the final study (Fig. 1). Clinical data and patient information were retrospectively retrieved from clinical and pathological databases, including gender, age, body mass index, pathological stage, TN stage, tumor location, differentiation, tumor deposit, lymphovascular invasion, perineural invasion, molecular biomarkers of the Ras signaling pathway (KRAS, NRAS, and BRAF types), CEA, CA19-9, and mismatch repair (MMR) status (deficient MMR and proficient MMR).

Fig. 1
figure 1

Patient selection flow diagram

Full size image

Magnetic resonance imaging

All rectal MRI was performed on a 3-Tesla MRI system (MAGNETOM Skyra, Siemens Healthcare GmbH) using an 18-channel phased-array body coil and an integrated spine coil. Before scanning, intestinal cleaning was performed by enema administration using 20 mL of glycerin. The imaging protocol included sagittal T2-weighted imaging (T2WI), oblique axial high-resolution T2WI, axial DWI (the optimal b-value combination of 0 and 1000 s/mm2 was recommended for RC based on previous studies [23, 30]), and axial T1-weighted imaging (T1WI). The main imaging parameters of the MRI protocol are summarized in supplemental Table 1. Gadolinium contrast-enhanced T1WI of the pelvis was obtained in the sagittal, coronal, and axial planes. The total scan time was approximately 10 min.

Image analysis

DWI images were independently evaluated by two experienced pelvic radiologists using a prototype post-processing software (Body Diffusion Toolbox, Siemens Healthcare GmbH). ADC maps were derived from DW images using the mono-exponential model

$$ \mathrm{ADC}=1/b\ast \log \left({S}_0/{S}_{\mathrm{b}}\right) $$

where S0 and Sb represent the signal intensity with b = 0 s/mm2 and with b > 0 s/mm2 diffusion weighting, respectively. bThreshold map was calculated using the formula

$$ {b}_{\mathrm{Threshold}}=-1/\mathrm{ADC}\ast \log \left(\mathrm{Threshold}/{S}_0\right) $$

with Threshold defined as 50 au for RC, the intensities of bThreshold map indicate the b-values at which the diffusion signal drops under a given threshold, and its unit is s/mm2 [23, 28]. Single slices with the maximum cross-sectional tumor size were used to delineate the regions of interest (ROIs) and were manually outlined on ADC and bThreshold maps of the lesions by two independent observers. The mean values of ADC and bThreshold were recorded for each lesion (supplemental Fig. 1). In addition, the areas of ROI were also recorded. The contrast-to-noise ratios (CNRs) of the DWI images with b = 1000 s/mm2 and bThreshold maps were determined using

$$ \mathrm{CNR}=\left|{SI}_{\mathrm{lesion}-}{SI}_{\mathrm{gluteus}\ \mathrm{maximus}}\right|/{\left({\sigma_{\mathrm{lesion}}}^2+{\sigma_{\mathrm{gluteus}\ \mathrm{maximus}}}^2\right)}^{1/2} $$

where SI and σ refer to the mean signal intensity and standard deviation of the ROI, respectively, of the lesion or gluteus maximus (same size = 100 voxels).

Pathological evaluation

All tissue sections underwent H&E staining. Histopathology results included tumor TN staging, histological grade, presence of perineural invasion, presence of lymph-vascular invasion (LVI), tumor deposits, and descriptions of the circumferential resection margins [21].

H&E-stained sections were scanned at medium power (10 × magnification), and an area with maximal budding was identified at the invasive front. TB was counted and scored by two experienced pathologists with consensus, according to the ITBCC 2016 Recommendations [31]. TB was counted in a selected area at 20 × magnification. The bud count was divided by a normalization factor to determine the tumor bud count per 0.785 mm2 [31]. TB categories were based on the bud count and defined as follows: Bd 1 (low-grade): 0–4 buds; Bd 2 (intermediate-grade): 5–9 buds; Bd 3 (high-grade): 10 buds or more. Patients were divided into two groups, low-intermediate (Bd 1+2) and high grade (Bd 3) for analysis.

Statistical analysis

SPSS (version 22.0, Inc.) and MedCalc Statistical software (version 13.0.0.0, MedCalc Software) were used to perform statistical analyses. The Kolmogorov-Smirnov test was used to check for the normality of continuous variables. Continuous variables are presented as mean and standard deviation or median and quartile according to the normal distribution of data, and categorical variables are expressed as percentages. Categorical variables were assessed using the chi-square test or Fisher’s exact test. Interobserver reproducibility for the ADC, bThreshold values, and ROI sizes was assessed using intraclass correlation coefficients (ICCs), coefficients of variability (CVs), and Bland-Altman plots. ICC values > 0.75 indicated excellent agreement, 0.4 to 0.75 indicated good agreement, and < 0.4 indicated poor agreement. Levene’s test was used to test for equality of error variances. Significant differences in CNR between DWI (b = 1000 s/mm2) images and bThreshold maps were assessed using paired-sample t-tests. The ADC and bThreshold values of the two radiologists were averaged. The correlations of the mean ADC and bThreshold values with TB category were determined using Spearman’s rank correlation test. Differences in ADC and bThreshold values among Bd grades 1, 2, and 3 were evaluated using Kruskal-Wallis one-way ANOVA with a pairwise multiple comparisons test. Differences in ADC and bThreshold values between budding groups Bd 1+2 and Bd 3 were evaluated using the Mann-Whitney U test. The diagnostic performance of the ADC and bThreshold values for group Bd 1+2 vs group Bd 3 was assessed using area under the curve (AUC) and compared using the DeLong test. The decision curve analysis (DCA) was performed by estimating the net benefit with probability thresholds to confirm the clinical benefit. A p value < 0.05 indicated statistical significance.

Results

Patient characteristics

A total of 51 LARC patients with rectal adenocarcinomas were enrolled in the final analysis, including 32 males with a mean age of 56.3 ± 8.8 years (range 32–74 years). Total mesorectal excision was performed at a time interval of 7.8 ± 4.3 days (range 3–14 days) after MR imaging. There were 13 Bd 1, 13 Bd 2, and 25 Bd 3 patients. No case had a positive circumferential resection margin. Characteristics and pathological outcomes were not significantly different among the patients (Table 1).

Table 1 Patient demographics and clinicopathologic findings
Full size table

All 51 cases of LARC had a single lesion, of which 29 cases had space-occupying masses, 16 cases had an irregular thickening of the local intestinal wall, and 6 cases had abnormal local nodular signals. All lesions demonstrated high signals on DWI images with b = 1000 s/mm2 and on bThreshold maps (Fig. 2). Significant differences were observed in CNR between DWI images and bThreshold maps (7.779 ± 3.508 vs 9.807 ± 4.811, p = 0.005).

Fig. 2
figure 2

Images of a rectal cancer lesion from a patient with poorly differentiated adenocarcinoma confirmed as stage IIa (pT3N0). a T2WI showing abnormal signals on the posterior of the rectal wall (arrow). b Axial DWI at b = 0 s/mm2 showing abnormal signals on the posterior of the rectal wall (arrow). c DWI at b = 1000 s/mm2 showing the lesion with high-signal intensity (arrow). d ADC map showing the lesion with low-signal intensity (arrow). e The bThreshold map showing the lesion with high-signal intensity (arrow). f Hematoxylin & eosin (H&E)–stained high-grade histopathological section showing more than 10 buds at the invasive front (black arrows)

Full size image

Interobserver variability of ADC, b Threshold values, and ROI sizes

There was an excellent reproducibility for ADC (ICC, 0.933; CV, 8.807%) and bThreshold measurements (ICC, 0.958; CV, 7.399%). In addition, the bias and limits of agreement for ADC (−2.655%; −21.666 to 16.356) and bThreshold (3.880%; −16.788 to 24.549) were relatively low (Table 2, Fig. 3). No significant difference was observed between the two observers in ROI size delineation (365.2 ± 159.5 mm2 vs 375.1 ± 168.5 mm2, p = 0.491), which also had excellent agreement, with ICC and CV values of 0.934 and 8.425%, respectively (Table 2).

Table 2 Interobserver variability of ADC and bThreshold values
Full size table
Fig. 3
figure 3

Bland-Altman plots. a ADC plot. b bThreshold plot. The solid blue line in each plot indicates the mean difference in reads between two radiologists. The dashed red lines indicate the limits of the agreements

Full size image

Correlation and comparison of mean ADC and b Threshold values among and between different TB grades

For patients with LARC, a significant negative correlation was observed between mean ADC values and TB grades, and a positive correlation was found between mean bThreshold values and TB grades. Categories Bd 1, 2, and 3 had Spearman correlation coefficients of −0.392 and 0.675 (p < 0.05) for ADC and bThreshold, respectively. Significant differences were observed in mean ADC and bThreshold values among Bd categories 1, 2, and 3 and between the groups Bd 1+2 and Bd 3, respectively (Table 3). Multiple pairwise comparisons showed that significant differences were found in ADC and bThreshold values between categories Bd 1 and Bd 3 and between categories Bd 2 and Bd 3. No significant differences were found in ADC and bThreshold values between categories Bd 1 and Bd 2. The mean bThreshold value of category Bd3 was significantly higher than the Bd 1+2 group (Fig. 4, supplemental Table 2) (p < 0.05).

Table 3 Comparison of tumor budding category in RC patients using DWI parameters
Full size table
Fig. 4
figure 4

Box plots of (a) apparent diffusion coefficient (ADC) and (b) bThreshold values in Bd groups 1, 2, and 3. Box plots of (c) apparent diffusion coefficient (ADC) and (d) bThreshold values in low-intermediate grade and high-grade TB

Full size image

Diagnostic performance of ADC and b Threshold

The AUC, sensitivity, and specificity of the mean ADC and bThreshold values for differentiating groups Bd 1+2 vs Bd 3 were 0.726, 69.2%, and 84.0% and 0.914, 88.5%, and 92.0%, respectively (Table 4). The diagnostic performance of bThreshold maps was greater than that of ADC values for group Bd 1+2 vs group 3 (p = 0.048) and the optimal cut-off threshold of bThreshold values was 1.773 × 103 s/mm2 (Fig. 5 and supplemental Fig. 2).

Table 4 Diagnostic performance of ADC and bThreshold values in differentiating TB grades (Bd 1+2 vs Bd 3)
Full size table
Fig. 5
figure 5

Receiver operating characteristic curves of the ADC and bThreshold values for differentiating tumor budding (Bds 1+2 and 3). Areas under the curves of ADC and bThreshold values are 0.726 and 0.914, respectively (p = 0.048)

Full size image

Decision curve analysis

The DCA showed an adequate performance for ADC and bThreshold values in distinguishing group Bd 1+2 from group Bd 3 (Fig. 6). When the threshold probability was between 0.15 and 1.0, the bThreshold map for predicting category Bd 3 TB showed a greater advantage than either the “all” or “none” scheme.

Fig. 6
figure 6

Decision curve analysis of the apparent diffusion coefficient (ADC) and bThreshold values. The light gray line represents the assumption that all patients had high-grade TB. The dark gray line represents the hypothesis that no patients were high-grade TB. The red decision curve shows that when the threshold probability was between 0.15 and 1.0, the bThreshold value was better in predicting high-grade TB than the ADC value

Full size image

Discussion

TB is used as a prognostic biomarker for solid tumors such as RC and has the potential to stratify patients for different therapeutic options [32, 33]. TB is closely related to many different clinical and histological parameters, including histological grade, lymph node involvement, and lymphovascular invasion and metastasis in RC [34,35,36,37,38,39]. High-grade TB predicts adverse outcomes in LARC, including higher TNM stages, higher recurrence rates, and increased risk of mortality [34,35,36,37,38,39]. Identifying individuals with high-grade TB in a preoperative assessment could be helpful for guiding clinical practice.

We explored the values of ADC and bThreshold in the preoperative diagnosis of TB. We found that significant correlations and differences were observed in mean ADC and bThreshold values between different categories of TB grades, particularly group Bd 1+2 vs group Bd 3. The diagnostic performance of bThreshold maps was higher than those of ADC and may be applicable in the preoperative TB evaluation in LARC patients.

The present study showed that the ADC value for patients in the Bd 1+2 group was significantly higher than that for patients in the Bd 3 group. In contrast, we found that the mean bThreshold value for patients in the Bd 3 group was significantly higher than that in Bd 1+2 group. This finding could be explained by Bd 3 tumors having more tumor cells at the invasive margin with higher cell density and smaller interstitium. ROC curves for ADC and bThreshold showed large AUCs (> 0.7), indicating that both values may be used to distinguish patients in the Bd 3 group from patients in the Bd 1+2 group.

We performed DCA to assess the performance of ADC and bThreshold values in distinguishing the Bd 1+2 group from the Bd 3 group. The net benefit of bThreshold maps was better than those of ADC and had threshold probabilities of 0.15–1.0. We found that bThreshold values can distinguish between Bd 1 patients from Bd 3 patients, and Bd 2 patients from Bd 3 patients. However, ADC values can only distinguish Bd 1 patients from Bd 3 patients, indicating that bThreshold maps are better for preoperative prognosis of TB for LARC patients. The underlying mechanism of bThreshold values outperform ADC may be explained as follows: Firstly, bThreshold maps offer a positive contrast in dense tissues which more conform to the doctor’s viewing habits, while the ADC maps show a negative contrast in dense lesions. Secondly, the signal intensities of bThreshold maps represent b-values at which the diffusion signal drops under a given threshold [29]. In the present study, 50 a.u. was optimized and used for evaluating rectal lesions. Compared with ADC, it has large dynamical range on bThreshold maps among different Bd grades. Thirdly, the values of bThreshold maps among budding grades 1, 2, and 3 were 1.508 (1.151–1.761), 1.547 (1.361–1.634), and 1.982 (1.844–2.057) ×103 s/mm2, respectively. Those maps have some similarity to DWI images acquired with high b-values of 1500–2000 s/mm2; however, ADC map was derived from relatively low b-values of 0 and 1000 s/mm2. Therefore, better lesion-to-normal tissue contrast was obtained using bThreshold maps, as well as improved diagnostic performance in differentiating TB grades (Bd 1+2 vs Bd 3).

We found that bThreshold maps provided higher CNRs and improved visualization and detection of lesions compared with DWI images. This is consistent with observations that bThreshold maps provide better lesion visualizations for rectal tumors [23, 27, 28]. Rectal tumors show hyperintensity in bThreshold maps in comparison with normal tissues, which are more familiar to physicians [23]. We found that bThreshold maps significantly improve the signal contrast between lesions and normal tissue and provided significantly higher CNR than DWI images with a b = 1000 s/mm2. RC lesions often are irregularly shaped and cannot be easily distinguished from the surrounding adipose tissues due to inflammation and blood vessel invasion. The signal contrast of bThreshold maps helps to detect such lesions [40]. Improved CNR would also allow for more accurate ROIs for quantitative measurements. Apart from the definition and a standardized scoring system, the application of TB has been hindered by the lack of reproducibility. Puppa G found that the reproducibility assessment of TB is higher in early colorectal cancer and experienced gastrointestinal pathologists [41]. Immunohistochemistry (IHC) highlights tumor budding cells and improves the visualization of TBs [42,43,44,45] and the reproducibility. Our study demonstrated that bThreshold maps could be used to evaluate the TB grades preoperatively in patients with LARC, showing excellent reproducibility among bThreshold values (ICC 0.958; CV 7.399%). Narrow intervals observed in Bland-Altman plots indicated that interobserver variability would be low in clinical use. No complex post-processing technique is needed to assess TB using bThreshold maps with a given optimized b value. By quantifying the optimal cut-off threshold, less time would be needed to perform the complex diagnosis (e.g., high TB grades have bThreshold values > 1.773).

There were several limitations in this study. First, a small number of patients were enrolled in our retrospective study, making the study prone to selection bias. The small sample and retrospective nature of the study may explain the lack of predictive value of TB grades for lymph node positivity in this cohort. Second, it was a single-center study with only one MRI system. Finally, our final budding count in this work was based on H&E assessments as per the ITBCC group recommendations. Despite IHC’s potential usefulness in effectively confirming bud count in challenging cases, the H&E method is more cost-effective and can reduce the economic burden of patients. Large multicenter randomized controlled trials assessing bThreshold maps that also compare the effectiveness of H&E staining with IHC are necessary to validate our results and provide additional information.

In conclusion, bThreshold maps may serve as a preoperative non-invasive alternative for evaluating TB in patients with LARC. bThreshold values could distinguish among different TB grades in LARC patients due to their higher CNR. TB grades based on bThreshold values could be considered along with adverse clinicopathological parameters in LARC patients when assessing individualized therapeutic strategies.