Clinical application of digital pathology in small intestinal neuroendocrine tumours: proposal of a new G0 category

Abstract

Introduction

Small intestinal neuroendocrine tumours (siNETs) are usually indolent neoplasms, most of which are classified as grade 1 with very low proliferative activity. In practice, Ki-67 in G1 siNETs is often reported simply as ‘<3%’, limiting prognostic discrimination. Digital pathology now enables reliable automated quantification of very low Ki-67 values. We investigated whether precise Ki-67 quantification could improve prognostic stratification in G1 siNETs.

Methods

We retrospectively included 63 consecutive patients who underwent surgery for a G1 siNET at Beaujon Hospital between 2001 and 2017. Ki-67 was quantified on primary tumours (PTs) and, when available, lymph node and liver metastases, using QuPath automated image analysis. Ki-67 was assessed in the highest-proliferation hotspot (0.4 mm²; >2,000 tumour cells; Ki-67_max) and as the mean of five hotspots (Ki-67_mean). The primary endpoint was recurrence-free survival (RFS).

Results

Among 63 patients, 33 underwent curative-intent surgery (median age: 56 years; 69% male). Median tumour size was 20 mm. At surgery, 87% were node-positive and 67% had synchronous metastases. Median Ki-67_max/Ki-67_mean in PTs was 0.99%/0.71%. In localised disease, Ki-67_max predicted RFS (AUROC = 0.84). Median follow-up was 60.2 months. Median RFS was 112 months in patients with Ki-67_max ≥1%, whereas no recurrences occurred in those with Ki-67_max <1% (P = 0.04). Ki-67_mean provided no additional prognostic value.

Conclusion

Precise digital Ki-67 quantification refines prognostic stratification in G1 siNETs. Tumours with Ki-67 <1% may represent a biologically distinct ‘G0’ subgroup associated with extremely low recurrence risk after curative surgery.

Introduction

Although the incidence of small intestinal, ileal neuroendocrine tumours (siNETs) is low at 0.29/100,000 inhabitants and ranks third among gastroenteropancreatic NETs, it remains the second leading cause of small intestine tumours, with a rapidly increasing incidence (1). They derive from enterochromaffin cells, which produce serotonin. Their biology differs from other digestive neuroendocrine tumours. These tumours show very low cell proliferation. Although the numbers vary across studies, most siNETs fall into grade 1 (about 75.5%), with fewer classified as grade 2 (18%) or grade 3 (6.5%) (2). Despite this low-proliferative nature, siNETs are most often discovered at an advanced stage, either locally advanced primary tumours (67.4% of cases) or liver metastases (28.3% of cases) (2). Collectively, ileal NETs have a greater metastatic potential than other well-differentiated gastrointestinal NETs, with 76.3–89% of patients presenting with distant or nodes metastases at diagnosis or during follow-up, compared with 58.8% for all NETs (3, 4). Predicting the behaviour of siNETs is, thus, clinically important for stratifying patient treatment and follow-up. It is well known that Ki-67 is currently the best marker of progression-free survival (5, 6, 7).

In current practice, in our experience and as also observed in routine pathology reports we have reviewed, Ki-67 is often reported as <3% for grade 1 siNETs without precise quantification, although this practice is not formally addressed in current pathology reporting recommendations. Very low Ki-67 rates, below 1 or 2%, are not calculated accurately in the vast majority of cases, as this requires counting a very large number of negative cells for each positive cell, and therefore resorting to automatic quantification. However, sub-stratification within grade 1 by precise quantitative Ki-67 has been suggested as a possible refinement of prognostic assessment (7, 8).

Recent evidence has demonstrated the prognostic value of sub-stratifying tumours within established WHO grades. Notably, splitting grade 2 tumours by Ki-67 index has been shown to identify biologically and prognostically distinct subgroups with significantly different patient survival (9), suggesting that refined stratification within low-grade categories may yield clinically meaningful prognostic discrimination. This principle raises the question of whether a similar Ki-67 sub-stratification within grade 1 siNETs would identify a biologically distinct, more indolent subgroup with better outcomes. The stakes of this question are underscored by a striking numerical asymmetry: within grade 1 alone, proliferation indices can span more than a hundredfold range, far exceeding the mere 7-fold range across the entire grade 2 category. Calculating the proliferation index can be a long and tedious process, but new tools are available that enable the proliferation index to be measured quantitatively in a consistent and reproducible manner. Freely available software such as QuPath has already been used in several studies to quantify Ki-67, particularly in neuroendocrine tumours (10, 11, 12, 13). Rare studies have attempted to better classify patients with a very low grade index, but without fully following the current classification and, thus, mixing grade 1 and 2 patients (14).

In this study, we explored whether precise Ki-67 assessment with digital pathology could help better predict the prognosis of patients with grade 1 siNETs, and especially verify if a very low proliferation index could indicate a better prognosis that could change patients’ care.

Materials and methods

Study population

We retrospectively included all consecutive patients who underwent surgery for a G1 (i.e. Ki-67 <3%) ileal NET at Beaujon Hospital, Clichy, France, between 2001 and 2017. Information such as patient age, sex, the presence and location of metastases, recurrence/progression, and death was gathered from clinical, surgical, pathological, and imaging reports. Inclusion criteria included patients being over 18 years old, having given consent to participate in the study, and having undergone surgery. Patients who did not give consent were not eligible for the study. The study was approved by the Institutional Review Board of Bichat-Claude Bernard Hospital (ID RCB No. 2018-A02716-49).

Staining and immunohistochemistry

For each tumour, we performed a standard haematoxylin and eosin (H&E) staining and a Ki-67 immunohistochemical staining using a standard streptavidin-biotin-peroxidase protocol on 3 μm thick slices of formalin-fixed, paraffin-embedded (FFPE) tissue, using an anti-Mib1 antibody (clone Mib1, IgG1, 1:100). Due to the long induction period, the Ki-67 staining was redone to standardise the quality of the staining and ensure reproducible results.

Histological evaluation

The diagnosis, based on H&E, was confirmed by two pathologists (AC and NG). The recorded pathological features included tumour size, differentiation, vascular or perineural invasion, necrosis, and resection quality (defined as microscopically negative margins or positive margins). The Ki-67 index was determined in the primary tumour and, when metastatic, in the lymph node and/or distant metastasis by computer-assisted image analysis using the QuPath system (version QuPath 5.0).

For each case, the Ki-67 index was assessed in hotspots previously identified by the pathologist. Quantification of Ki-67 was performed in the five highest-proliferation hotspots, each measuring 0.4 mm², within every sample. The highest value among these five measurements was defined as the classical hotspot Ki-67_max index. To evaluate intratumoural homogeneity, we additionally calculated Ki-67_mean, defined as the arithmetic mean of the Ki-67 indices across the five hotspots,

$$Ki{67}_{mean}=\frac{{\sum }_{hotspot1}^{hotspot5}Positive\hspace{0.17em}tumour\hspace{0.17em}cells}{{\sum }_{hotspot1}^{hotspot5}Tumour\hspace{0.17em}cells}$$

Cell detection was performed using QuPath. An object classifier was then trained to distinguish tumour cells from stromal cells. Quality control was carried out at each step (both on the cell detection and the machine learning classification outputs). Finally, all Ki-67-positive cells were highlighted and counted exclusively within the tumour cell population.

Statistical analysis

Continuous variables are presented as median (range), and categorical variables are shown as numbers with percentages. Recurrence-free survival (RFS) was estimated using the Kaplan–Meier method. All statistical tests were two-sided, and P ≤ 0.050 was considered statistically significant in all analyses. Analyses were performed using R (version 4.5.2).

Results

Population

The study population had a median age of 57 years (range: 27–81), with a balanced sex distribution, including 60% men. Tumour size at pathology ranged widely, with 40% of lesions measuring ≥20 mm. Multifocality was frequent (38%). All cases were well-differentiated grade 1 tumours, with 87% presenting with lymph node metastases (Fig. 1). Vascular and neural invasion were also frequently identified, in 65 and 54% of cases, respectively. Regarding staging, the majority of patients presented with locally advanced primary tumours, with 83% classified as T3–T4. Distant metastases at diagnosis were present in 67% of the cohort. Among the 63 patients, 33 (52.4%) underwent surgery with curative intent. Concerning follow-up and treatment, the mean follow-up duration was 60.2 months (range: 19.7–94.3 months) (Table 1).

Figure 1.

Small intestinal neuroendocrine tumours (siNETs) and digital pathology Ki-67 quantification. (A) Low-magnification view of an siNET. (B) High-magnification view showing stromal infiltration by cords of homogeneous tumour cells. (C) Ki-67 immunohistochemical staining demonstrating very few Ki-67-positive tumour cells (brown nuclear staining). (D) Example of digital pathology-based Ki-67 quantification. After detection of all cells, classification is performed in two steps: first, positive versus negative cells, and second, tumour versus non-tumour cells. Red cells represent Ki-67-positive tumour cells, blue cells represent Ki-67-negative tumour cells, and green cells represent Ki-67-negative non-tumour cells. (E and F) Second example of a grade 0 siNET with similar morphology, showing a Ki-67 index <1%.

Table 1.

Description of the population by group (R1/R0 M+/R0 M−).

	All (n = 63)	R1 (n = 30)	R0 M+ (n = 14)	R0 M− (n = 19)	P-value (3 groups)	P (R0 M+ vs R0 M−)
Male gender	38 (60.3%)	15 (50.0%)	9 (64.3%)	14 (73.7%)	0.241	0.707
Age at surgery (years), median (min–max)	57.1 (27–82)	61.0 (35–82)	55.5 (49–76)	55.8 (27–75)	0.129	1.000
Symptoms at diagnosis
Carcinoid syndrome	23 (36.5%)	17 (56.7%)	6 (42.9%)	0 (0.0%)	<0.001	0.003
Haemorrhage	4 (6.3%)	0 (0.0%)	0 (0.0%)	4 (21.1%)	0.007	0.119
Obstruction	6 (9.5%)	3 (10.0%)	0 (0.0%)	3 (15.8%)	0.309	0.244
Abdominal pain	29 (46.0%)	15 (50.0%)	7 (50.0%)	7 (36.8%)	0.630	0.497
Incidental	21 (33.3%)	8 (26.7%)	5 (35.7%)	8 (42.1%)	0.524	1.000
Urgent surgery	4 (6.3%)	2 (6.7%)	0 (0.0%)	2 (10.5%)	0.470	0.496
Work-up
Urine 5-HIAA (>ULN)	26 (41.3%)	17 (56.7%)	6 (42.9%)	3 (15.8%)	0.018	0.122
CT-scan	63 (100.0%)	30 (100.0%)	14 (100.0%)	19 (100.0%)	1.000	1.000
MRI	30 (47.6%)	21 (70.0%)	6 (42.9%)	3 (15.8%)	<0.001	0.122
SSTR imaging*	53 (84.1%)	25 (83.3%)	13 (92.9%)	15 (78.9%)	0.550	0.366
FDOPA-PET	20 (31.7%)	9 (30.0%)	4 (28.6%)	7 (36.8%)	0.846	0.719
Pathology
Tumour size >20 mm	25 (39.7%)	16 (53.3%)	4 (28.6%)	5 (26.3%)	0.107	1.000
Multifocal tumours	24 (38.1%)	12 (40.0%)	6 (42.9%)	6 (31.6%)	0.770	0.716
T stage					<0.001	0.066
T1	2 (3.2%)	0 (0.0%)	0 (0.0%)	2 (10.5%)
T2	8 (12.7%)	0 (0.0%)	6 (42.9%)	2 (10.5%)
T3	25 (39.7%)	12 (40.0%)	3 (21.4%)	10 (52.6%)
T4	27 (42.9%)	17 (56.7%)	5 (35.7%)	5 (26.3%)
Vascular invasion	41 (65.1%)	20 (66.7%)	11 (78.6%)	10 (52.6%)	0.294	0.160
Perineural invasion	34 (54.0%)	17 (56.7%)	5 (35.7%)	12 (63.2%)	0.271	0.166
Node metastasis	55 (87.3%)	26 (86.7%)	14 (100.0%)	15 (78.9%)	0.198	0.119
Metastasis
No metastases	21 (33.3%)	2 (6.7%)	0 (0.0%)	19 (100.0%)	<0.0001	<0.0001
Metastases	42 (66.7%)	28 (93.3%)	14 (100.0%)	0 (0.0%)	<0.0001	<0.0001
Liver	38 (60.3%)	25 (83.3%)	13 (92.9%)	0 (0.0%)	<0.0001	<0.0001
Peritoneum	12 (19.0%)	11 (36.7%)	1 (7.1%)	0 (0.0%)	0.003	0.424
Distant lymph nodes	8 (12.7%)	7 (23.3%)	1 (7.1%)	0 (0.0%)	0.045	0.424
M classification					<0.0001	<0.0001
M0	21 (33.3%)	2 (6.7%)	0 (0.0%)	19 (100.0%)
M1a	23 (36.5%)	11 (36.7%)	12 (85.7%)	0 (0.0%)
M1b	3 (4.8%)	2 (6.7%)	1 (7.1%)	0 (0.0%)
M1c	15 (23.8%)	14 (46.7%)	1 (7.1%)	0 (0.0%)

^*SSTR imaging = Octreoscan or DOTATOC-PET.

Ki-67 quantification

The proliferation index was measured in all specimens using the same standardised technique (Fig. 1D). The median number of tumour cells counted within the 0.4-mm² hotspot was 2,769 (range: 800–5,153). By specimen type, the median tumour cell counts were 2,553 (1,044–5,083) in primary tumours, 2,971 (1,245–5,153) in lymph node metastases, and 3,198 (1,662–4,934) in liver metastases.

Regarding the analysis of multiple samples, 50 patients had at least two different types of specimens among the three sites assessed (primary tumour, lymph node metastasis, and liver metastasis). The median Ki-67_max index in the hotspot was 0.99% in primary tumours, 0.88% in lymph node metastases, and 0.94% in liver metastases. Across all cases, 26 patients showed a higher Ki-67_max index in the primary tumour, whereas 24 had a higher Ki-67_max index in either lymph node or liver metastases. Among the 29 patients with a primary tumour Ki-67_max <1% and at least one additional specimen assessed, 23 (79%) had a Ki-67_max that remained <1% at all available metastatic sites, confirming that the <1% threshold is largely concordant across specimen types.

Ki-67 quantification was performed within a 0.4-mm² hotspot and in four additional hotspots; the average of these five measurements was used to evaluate intratumoural Ki-67 heterogeneity. Ki-67_mean was 0.71% (0.01–4.95) in primary tumours, 0.51% (0.01–4.93) in lymph node metastases, and 0.64% (0.02–1.66) in liver metastases (Table 2).

Table 2.

Computer-assisted image analysis of Ki-67 in primary tumours and node and liver metastases of siNETs by group.

	All (n = 63)	R1 (n = 30)	R0 M+ (n = 14)	R0 M− (n = 19)	P-value (3 groups)	P (R0 M+ vs R0 M−)
Ki-67 on primary tumour (n = 63/30/14/19)
Ki-67 max	0.99 (0.02–9.42)	0.85 (0.02–9.42)	0.92 (0.05–1.89)	1.22 (0.47–2.94)	0.056	0.017
Ki-67 mean	0.71 (0.01–4.95)	0.62 (0.01–4.95)	0.48 (0.01–1.05)	1.23 (0.19–2.70)	0.007	0.001
Ki-67 on lymph node metastasis (n = 43/23/8/12)
Ki-67 max	0.88 (0.04–6.51)	0.98 (0.04–6.51)	0.21 (0.07–0.70)	0.96 (0.05–2.75)	0.025	0.004
Ki-67 mean	0.51 (0.01–4.93)	0.73 (0.01–4.93)	0.17 (0.04–0.41)	0.65 (0.03–2.25)	0.030	0.007
Ki-67 on liver metastasis (n = 14/3/11/0)
Ki-67 max	0.94 (0.03–3.17)	2.14 (0.67–3.17)	0.93 (0.03–2.04)	-	0.186	n/a
Ki-67 mean	0.64 (0.02–1.66)	1.64 (0.42–1.66)	0.58 (0.02–1.64)	-	0.243	n/a

Values = median (min–max) of Ki-67 labelling index (%) by computer-assisted image analysis.

Survival analysis

Among the 63 consecutive patients retrospectively included in this study, 33 (55%) underwent surgery with curative intent. In practice, 19 patients presenting with only a localised tumour were resected. In addition, when the disease was discovered at a metastatic stage with synchronous liver metastases, combined resection of the primary tumour and the liver metastases was performed in 14 cases (Table 1). These 14 patients are precisely those with synchronous metastases discovered at surgery, in whom complete macroscopic (R0) resection of both the primary tumour and all liver lesions was achieved.

Among these 33 patients, 69% were male, the mean age was 56 years (range: 27–81), and carcinoid syndrome was present in 6 patients (17%). The mean tumour size was 21 mm (range: 7–60). At the time of surgery, lymph node involvement (N+) was observed in 88% of patients and distant metastases (M+) were observed in 56%. The mean follow-up duration was 49 months (range: 19.7–94.3 months). All tumours were readily identified as well-differentiated neuroendocrine tumours on H&E-stained sections. Recurrence-free survival was assessed in patients treated with curative intent. Among these 33 patients, 18 had a Ki-67_max <1% and 15 had a Ki-67_max ≥ 1% in the primary tumour. No significant difference in recurrence-free survival was observed between these two groups according to the log-rank test (P = 0.1826).

However, in the subgroup of patients with localised tumours without metastases at diagnosis (n = 19), 10 with Ki-67_max ≥1% and 9 with Ki-67_max <1%, the difference in recurrence-free survival were statistically significant (P = 0.0441). Median RFS was 112 months in patients with Ki-67_max ≥1%, whereas no recurrences occurred in those with Ki-67_max <1% (Fig. 2).

Figure 2.

Recurrence-free survival of patients with siNETs according to a Ki-67 cut-off of 1%. Kaplan–Meier recurrence-free survival analyses based on a Ki-67 index cut-off of 1%. (A and B) Analyses restricted to all 33 patients treated with curative-intent surgery (including 14 with synchronous liver metastases who underwent combined resection), shown (A) across all tumour localisations or (B) in primary tumours only. (C and D) Analyses restricted to the 19 patients with strictly localised disease and no metastasis at diagnosis, shown (C) across all tumour localisations or (D) in primary tumours only. Time 0 corresponds to the date of curative-intent surgery.

In contrast, among the 14 patients who presented with metastatic disease (9 with a primary Ki-67_max <1% and 5 with ≥1%), recurrence-free survival did not differ significantly between Ki-67 subgroups (P = 0.4981), a result likely limited by the small sample size of this subgroup. Moreover, the same results were found when counting Ki-67 in the primary or different localisations.

Discussion

In our cohort of surgically resected siNETs, we demonstrate that in patients with grade 1 and no evidence of metastatic disease, precise and accurate grading of the Ki-67 index remains crucial to ensure appropriate classification. Our results also demonstrate that the Ki-67_max index can be extremely low, reaching 0.01%, and that a cut-off of 1% identifies prognostically distinct subgroups within grade 1 tumours.

The WHO and AJCC classification systems for digestive NETs are widely used in routine practice and classify well-differentiated NETs into three grades according to the Ki-67 index. Although this system predicts prognosis across digestive NETs, its discriminative value appears more limited in siNETs, where most tumours have a Ki-67 index below 3%, estimated at 86% in a recent study (5). This raises questions regarding the ability of current grading thresholds to capture prognostic heterogeneity of siNETs.

Our findings show marked heterogeneity in Ki-67 values even within grade 1 tumours. In primary tumours, Ki-67_max ranged from 0.01 to 2.7%, representing a 270-fold variation among tumours classified as G1. This broad range suggests that biologically and clinically relevant differences exist within this category.

This concept is consistent with previous work showing that further Ki-67-based stratification within the WHO grades may refine prognostic assessment. In G2 siNETs, a 10% threshold has already been shown to distinguish subgroups with different outcomes (9). Only a few studies, however, have examined the significance of very low Ki-67 values in siNETs (15, 16). Cunningham et al. similarly identified a 1% cut-off as prognostically relevant. However, their cohort included only metastatic tumours, compared patients with Ki-67 ≤ 1% to a heterogeneous group including G2 neoplasms, and did not report curative-intent surgery or postoperative disease status. Our study was restricted to strictly confirmed G1 tumours, used a standardised hotspot selection method, and focused on non-metastatic disease, allowing assessment of disease-free survival.

Despite these design differences, our results converge with those of Cunningham et al. (16) and support the biological relevance of the 1% threshold. A Ki-67 index below 1% appears to identify a subgroup of particularly indolent tumours, whereas values above this threshold remain associated with a higher risk of recurrence even within G1. This suggests that a refined stratification of G1 siNETs may be clinically meaningful and supports the idea of an emerging ‘G0’ subgroup for tumours with extremely low proliferative activity.

These findings may also have practical implications for pathology reporting. If a 1% cut-off has prognostic value in non-metastatic disease, pathology reports should not only assign the WHO grade but also provide precise numerical Ki-67 values and clearly indicate whether the tumour falls below or above 1%. Standardised digital pathology workflows and automated image analysis tools, such as QuPath, may help ensure that these measurements are reproducible and clinically useful. In our cohort, we evaluated both the hotspot Ki-67 (Ki-67_max) and the mean Ki-67 (Ki-67_mean) across five 0.4-mm² regions; the prognostic signal observed at the 1% cut-off was driven by Ki-67_max, while Ki-67_mean did not reach statistical significance for recurrence-free survival. This is consistent with prior work on Ki-67 intratumoural heterogeneity showing that hotspot count differs significantly from random-area count and supports the continued use of hotspot-based assessment for grading purposes (8). However, given the small sample size, this comparison should be regarded as exploratory and warrants confirmation in larger series before firm methodological recommendations can be drawn.

Several limitations must nevertheless be acknowledged. This was a retrospective single-centre study with a limited number of patients, particularly in the localised-disease subgroup, which comprised only 19 patients. Potential confounders, including tumour stage, size, multifocality, vascular and perineural invasion, and patient age, could not be adjusted for in a multivariable model because of sample size constraints. It therefore cannot be excluded that Ki-67 captures the effect of one or more correlated variables rather than acting as an independent prognostic factor. In addition, although staining and quantification were standardised within our institution, inter-laboratory variability in Ki-67 assessment remains an important unresolved issue. Reassuringly, the <1% threshold showed good within-patient reproducibility, remaining concordant across specimen types in 79% of patients in whom more than one specimen was assessed.

Future studies should validate these findings in larger multicentre cohorts using harmonised staining and analysis protocols. It will also be important to determine whether similarly low Ki-67 thresholds have prognostic value in other gastrointestinal NETs, including gastric and pancreatic NETs, to further refine surveillance strategies and better tailor follow-up intensity to tumour biology.

Declaration of interest

The authors declare that there is no conflict of interest that could be perceived as prejudicing the impartiality of the work reported.

Funding

This work did not receive any specific grant from any funding agency in the public, commercial, or not-for-profit sector.