ABSTRACT
Background
The objective of this study was to establish a new model for predicting the prognosis of endometrial carcinoma (EC) using tumor‐infiltrating lymphocytes (TILs) based on artificial intelligence (AI).
Methods
Patients with EC who were treated between 1989 and 2022 were included in this study. For each patient, one hematoxylin and eosin‐stained slide containing the most invasive frontline of the tumor was selected and digitized. The area within a 500 μm width span, extending 250 μm toward the stroma and tumor from the manually annotated invasive frontline, was automatically annotated. The average number of lymphocytes per area (μm2) in the annotated area was calculated using AI. Patients were classified into the High‐TIL and Low‐TIL groups, and survival analysis was conducted. Four mismatch repair (MMR)‐related proteins were evaluated using immunohistochemical staining.
Results
A total of 659 patients were included: 346 (52.5%) in the High‐TIL group and 313 (47.5%) in the Low‐TIL group. MMR deficiency was observed more frequently in the High‐TIL group than in the Low‐TIL group (p < 0.01). Progression‐free survival (PFS) and overall survival (OS) were better in the High‐TIL group than in the Low‐TIL group (both p < 0.01). Multivariate analysis revealed that TIL status was a prognostic factor for PFS (hazard ratio [HR] (95% confidence interval [CI]) 0.61 (0.43–0.87); p < 0.01) and OS (HR (95% CI) 0.54 (0.33–0.86); p = 0.01).
Conclusion
TILs evaluated using AI could accurately and significantly predict the prognosis of EC. Further studies are needed to establish new methods for evaluating TILs in ECs.
Keywords: artificial intelligence, endometrial carcinoma, mismatch protein repair, prognosis, tumor‐infiltrating lymphocytes
1. Introduction
Endometrial carcinoma (EC) is the most common gynecological carcinoma and accounts for 4.3% of carcinomas in women worldwide [1]. Total hysterectomy with bilateral salpingo‐oophorectomy is the primary surgical intervention for EC, and pelvic and para‐aortic lymphadenectomy can be selected based on the disease status [2]. Thereafter, adjuvant therapies such as radiation or chemotherapy are administered according to the degree to factors associated with recurrence, including age, lymphovascular invasion, tumor size, lower uterine segment or surface glandular involvement, myometrial invasion, and histological subtype [3, 4, 5, 6].
The presence of tumor‐infiltrating lymphocytes (TILs) has been identified as a prognostic factor for several types of carcinomas, including esophageal, breast, prostate, colon, and ovarian carcinoma [7, 8, 9, 10, 11, 12]. Furthermore, TILs were found to correlate with the clinical outcome of EC [7, 13, 14, 15, 16]. Conversely, TIL status was shown to be associated with mismatch repair (MMR) status [17]. According to the guidelines for TIL assessment based on the International Immuno‐Oncology Biomarker Working Group, the area of evaluation for TILs of ECs was defined as only within the tumor area, from the invasive frontline or within 1 mm from the tumor cell nest in the stromal area [7]. However, guidelines regarding the area for TIL evaluation or methods for lymphocyte enumeration are lacking, necessitating the development of standard models to evaluate TILs in EC.
Recent research has extensively focused on the effectiveness of artificial intelligence (AI) in pathological diagnosis [18, 19, 20, 21, 22]. AI can be applied in pathological diagnosis at various points, such as to enhance work efficiency, reduce discrepancies in diagnosis among observers, and quantify diagnosis and evaluation. For example, the efficacy of automated deep learning algorithms for detecting lymph node metastasis has been reported in the field of breast carcinoma [19]. In esophageal and colorectal carcinoma, the effectiveness of deep learning algorithms has been demonstrated for the detection of desmoplastic reactions [21, 22]. In the field of gynecological carcinoma, an AI‐based comprehensive evaluation of intraepithelial and stromal TILs of ovarian high‐grade serous carcinoma has been reported [23]. However, reports on the use of AI in the field of EC are lacking.
Therefore, in this study, we aimed to establish a new AI‐based model for TILs in the area around the invasive frontline of EC.
2. Methods
Patients with EC who underwent primary surgery between 1989 and 2022 at the National Defense Medical College Hospital were included in this study. Patients without clinical information or surgical tissue samples, as well as those with a prior history of chemotherapy, were excluded. Clinical information was collected from the medical records. We conducted pathological reviews using the 2020 World Health Organization criteria and classified these histological subtypes based on the 2023 International Federation of Gynecology and Obstetrics (FIGO) criteria [24, 25]. Non‐aggressive types were defined as grade 1 and grade 2 endometrioid carcinoma, and aggressive types were defined as grade 3 endometrioid carcinoma, serous carcinoma, clear cell carcinoma, carcinosarcoma, mucinous carcinoma, undifferentiated carcinoma, dedifferentiated carcinoma, and mixed carcinoma based on the 2023 FIGO criteria [25]. Additionally, staging was re‐evaluated or performed using the 2023 FIGO criteria [25].
For all patients, hematoxylin and eosin (HE)‐stained slides were used for AI‐based pathological diagnosis. A single slide per case containing the invasive frontline of the tumor was selected and digitized using NanoZoomer SQ (Hamamatsu Photonics K.K., Shizuoka, Japan). Digital whole‐slide images were uploaded to the HALO software (Indica Labs, Corrales, NM, USA). On each uploaded slide, the invasive front line was manually annotated (Figure 1a). An area within a width span of 500 μm, extending 250 μm toward the stroma and tumor from the invasive frontline, was automatically annotated by AI (Figure 1b), and the area size (μm2) was automatically calculated.
FIGURE 1.
Representative images of the invasive frontline and the area for evaluating tumor‐infiltrating lymphocytes. (a) The yellow line indicates the manually annotated invasive frontline. (b) The area between two red lines automatically annotated by artificial intelligence indicates that within a width span of 500 μm, 250 μm toward both stroma and tumor from the invasive front line.
To evaluate lymphocytes, classifier training for lymphocytes was conducted using Object Phenotyper, which has the function of automatically detecting spherical structures, such as lymphocytes or histiocytes, in the designated area. From among the detected structures, only lymphocytes were manually selected, and these selected lymphocytes were then trained using AI. A total of 2235 lymphocytes from four cases were trained for 5025 iterations. The accuracy of the classifier was evaluated using the F1 scores for two cases that were not used for training. The F1 score explains the extent of agreement between automatic AI detection and manual annotation of lymphocytes. After training, the classifier was applied to all cases, and the number of lymphocytes in the AI‐annotated area was automatically counted. Representative images of the AI‐detected structures are shown in Figure 2, where red indicates lymphocytes and green indicates other spherical structures such as histiocytes. The average number of lymphocytes per area size (/μm2) was subsequently calculated for all cases, and the optimum cut‐off value was determined using the receiver operator characteristic (ROC) curve for recurrence or progression. Cases with lymphocyte counts per area equal to or greater than the cut‐off value were defined as the High‐TIL group, whereas those with counts less than the cut‐off value were defined as the Low‐TIL group.
FIGURE 2.
Representative images of lymphocytes (a) and structures detected by artificial intelligence (b). Red indicates lymphocytes, and green indicates other spherical structures, such as nuclei or histiocytes.
In this study, the area for evaluating TILs was determined based on previous literature that reported the relationship between prognosis and TILs around the invasive frontline in ECs, and on exploratory analyses comparing multiple candidate areas [21]. Exploratory analyses included the following three candidate areas: (1) A range of 500 μm toward the stromal side from the invasive frontline, (2) A range of 250 μm toward the stromal side from the invasive frontline, (3) A total range of 500 μm, extending 250 μm toward the stroma and tumor from the invasive frontline. Details of these exploratory analyses are provided in Supporting Information (Figures S1 and S2 and Tables S1 and S2). Furthermore, to determine the cut‐off value for the High‐TIL and Low‐TIL groups, sensitivity analyses were performed using alternative cut‐off values, including the first quartile, median, and third quartile. The cut‐off value that showed the most appropriate prognostic association was selected for further analysis.
Immunohistochemistry (IHC) was performed using formalin‐fixed, paraffin‐embedded tissues. Tissue microarray blocks were prepared as previously described [26]. For IHC staining, we used the following antibodies: mouse monoclonal antibody for mutL homolog 1 (MLH1) (clone ES05; dilution 1:50; Dako, Glostrup, Denmark); rabbit monoclonal antibody for postmeiotic segregation increased 2 (PMS2) (clone EP51; dilution 1:40; Dako); mouse monoclonal antibody for mutS homolog 2 (MSH2) (clone FE11; dilution 1:50; Dako); and rabbit monoclonal antibody for mutS homolog 6 (MSH6) (clone EP49; dilution 1:50; Dako). All slides were deparaffinized and rehydrated using a graded ethanol series. Endogenous peroxidase activity was blocked by adding methanol to 0.3% hydrogen peroxide. Antigen retrieval was performed using Tris/EDTA buffer (pH 9.0). Additionally, all the slides were boiled in Tris/EDTA buffer at 95°C for 20 min. Subsequently, the slides were incubated at room temperature for 20 min with primary antibodies against MLH1, MSH2, and MSH6, and for 30 min with primary antibodies against PMS2. In addition, the slides were incubated with the Dako EnVision+ system‐horseradish peroxidase‐labeled polymer as a secondary antibody for 30 min at room temperature. Specific antigen–antibody reactions were visualized with 0.2% diaminobenzidine tetrahydrochloride and hydrogen peroxide and counterstained with Mayer's hematoxylin. For each antibody, negative control experiments were performed without primary antibodies. Representative images of positive cells from each IHC staining are shown in Figure 3.
FIGURE 3.
Representative images of positive immunohistochemistry (IHC) staining for mismatch repair proteins (×200) and complete absence of IHC staining. (a) MutL homolog 1 (MLH1). (b) Postmeiotic segregation increased 2 (PMS2). (c) MutS homolog 2 (MSH2). (d) MutS homolog 6 (MSH6). (e) Complete negative for staining.
To interpret the IHC intensities of MLH1, PMS2, MSH2, and MSH6, the percentage of positive cells was scored as follows: (1) less than 1% of the nuclei stained; (2) 1%–33% of nuclei stained; (3) 34%–67% of the nuclei stained; and (4) > 67% of the nuclei stained. Staining intensity was scored as follows: 0, absent; 1, weak; 2, moderate; and 3, strong. The sum of these parameters was defined as the IHC score. Cases with less than 1% positive cells, regardless of intensity, were defined as negative (Figure 3e); cases with negative staining for one or more of the four proteins were defined as MMR‐deficient (dMMR); and others were defined as MMR‐proficient.
Statistical analyses were performed using JMP Pro 17 software (SAS Institute Inc., Cary, NC, USA). The chi‐square test, Fisher's exact test, and Mann–Whitney U test were used to evaluate the clinical significance of clinicopathological factors. Progression‐free survival (PFS) was defined as the period from the day of primary surgery to the day of death, disease recurrence, or progression. Overall survival (OS) was defined as the period from the day of the primary surgery to the day of death or the last follow‐up. PFS and OS curves were generated using the Kaplan–Meier method. The survival distribution was compared using a log‐rank test. Univariate and multivariate analyses were performed using Cox proportional hazard regression models. To evaluate whether the prognostic effect of TILs differed according to MMR status, TILs and MMR interaction terms were included in the Cox regression model. Variables with statistical significance in the univariate analysis were included in the multivariate analysis using backward stepwise selection. The level of statistical significance was set at p < 0.05.
In addition, subgroup analyses stratified by histological subtypes and MMR status were performed to evaluate the prognostic impact of TILs status.
3. Results
A total of 659 patients with EC were included in this study. The F1 score was 0.81. The median number of lymphocytes counted by AI was 49 677 (21–242 372), and the median size of area for TIL evaluation was 10 893 771 (1 016 704–35 917 193) μm2. The median number of average lymphocytes per area size was 0.00493 (0.00001–0.01672)/μm2. The ROC curves for recurrence and progression are shown in Figure 4. The area under the curve was 0.570, and the cut‐off value was 0.00478. Sensitivity analyses using alternative cut‐off values, including the first quartile, median, and third quartile of TIL density, were performed. The cut‐off value defined by ROC analysis showed similar prognostic trends to the cut‐off value determined by median number, and the comparison of model fit using akaike information criteria and bayesian information criteria indicated similar results. Detailed results are provided in the Supporting Information (Figures S3 and S4 and Tables S3–S6).
FIGURE 4.
Receiver operating characteristics curve analysis of the average number of lymphocytes per area (/μm2) based on artificial intelligence for recurrence or progression. The area under the curve is 0.573.
Of the 659 patients, 346 (52.5%) were assigned to the High‐TIL group and 313 (47.5%) were assigned to the Low‐TIL group, and all patients were included in the analysis (Table 1). The median observational period was 63 (1–394) months. Patients in the High‐TIL group were diagnosed at an earlier FIGO stage (p = 0.049), had a higher frequency of less than 50% of myometrial invasion (p = 0.02), and a more frequent dMMR status (p < 0.01) than those in the Low‐TIL group. Other factors did not differ significantly between the two groups. The High‐TIL group had better PFS (p < 0.01) and OS (p < 0.01) than the Low‐TIL group (Figure 5).
TABLE 1.
Characteristics and results of immunohistochemical staining of 346 cases in the high‐tumor‐infiltrating lymphocyte group and 313 cases in the low‐tumor‐infiltrating lymphocyte group.
| Variables | High‐TIL group | Low‐TIL group | p | ||
|---|---|---|---|---|---|
| n = 346 | n = 313 | ||||
| Age (years) (%) | 59.58 ± 10.47 | 60.86 ± 12.16 | 0.06 | ||
| Average ± SD | 0.06 | ||||
| ≥ 60 | 157 | (45.4) | 165 | (52.7) | |
| < 60 | 189 | (54.6) | 148 | (47.3) | |
| Performance status (%) | 0.40 | ||||
| 0, 1 | 341 | (98.6) | 305 | (97.4) | |
| ≥ 2 | 5 | (1.4) | 8 | (2.6) | |
| FIGO stage (%) | 0.049 | ||||
| I | 248 | (71.7) | 195 | (62.3) | |
| II | 27 | (7.8) | 40 | (12.8) | |
| III | 52 | (15.0) | 60 | (19.2) | |
| IV | 19 | (5.5) | 18 | (5.7) | |
| Histology (%) | 0.27 | ||||
| Grade 1 endometrioid carcinoma | 166 | (48.0) | 165 | (52.7) | |
| Grade 2 endometrioid carcinoma | 76 | (22.0) | 51 | (16.3) | |
| Grade 3 endometrioid carcinoma | 36 | (10.4) | 36 | (11.5) | |
| Serous carcinoma | 16 | (4.6) | 15 | (4.8) | |
| Clear cell carcinoma | 9 | (2.6) | 9 | (2.9) | |
| Carcinosarcoma | 15 | (4.3) | 23 | (7.4) | |
| Mucinous carcinoma | 3 | (0.9) | 3 | (0.9) | |
| Undifferentiated carcinoma | 1 | (0.3) | 1 | (0.3) | |
| Dedifferentiated carcinoma | 1 | (0.3) | 1 | (0.3) | |
| Mixed carcinoma | 23 | (6.6) | 9 | (2.9) | |
| Classified histological type (%) | 0.80 | ||||
| Aggressive type | 104 | (30.1) | 97 | (31.0) | |
| Non‐aggressive type | 242 | (69.9) | 216 | (69.0) | |
| Myometrial invasion (%) | 0.02 | ||||
| ≥ 1/2 | 104 | (30.1) | 121 | (38.7) | |
| < 1/2 | 242 | (69.9) | 192 | (61.3) | |
| Cervical involvement (%) | 0.06 | ||||
| Positive | 40 | (11.6) | 53 | (16.9) | |
| Negative | 306 | (88.4) | 260 | (83.1) | |
| Ovarian metastasis (%) | 0.22 | ||||
| Positive | 24 | (6.9) | 30 | (9.6) | |
| Negative | 322 | (93.1) | 283 | (90.4) | |
| Lymph node metastasis (%) | 0.91 | ||||
| Positive | 43 | (12.4) | 40 | (12.8) | |
| Negative | 303 | (87.6) | 273 | (87.2) | |
| Distant metastasis (%) | 0.99 | ||||
| Positive | 19 | (5.5) | 18 | (5.8) | |
| Negative | 327 | (94.5) | 295 | (94.2) | |
| Lymphovascular invasion (%) | 0.63 | ||||
| Positive | 130 | (37.6) | 124 | (39.6) | |
| Negative | 216 | (62.4) | 189 | (60.4) | |
| Ascites or lavage cytology (%) | 0.21 | ||||
| Positive | 41 | (11.9) | 48 | (15.3) | |
| Negative | 305 | (88.1) | 265 | (84.7) | |
| Adjuvant therapy (%) | 0.39 | ||||
| Yes | 172 | (49.7) | 167 | (53.4) | |
| No | 174 | (51.3) | 146 | (46.6) | |
| MMR status (%) | < 0.01 | ||||
| Deficient | 147 | (42.5) | 101 | (32.3) | |
| Proficient | 199 | (57.5) | 212 | (67.7) | |
Abbreviations: FIGO, International Federation of Obstetrics and Gynecology; MMR, mismatch repair; SD, standard deviation; TIL, tumor‐infiltrating lymphocyte.
FIGURE 5.
Survival analysis of the High‐TIL and Low‐TIL groups. (a) Progression‐free survival (PFS) curves of all cases in the High‐TIL and Low‐TIL groups. The High‐TIL group has a better prognosis than the Low‐TIL group (p < 0.01). (b) Overall survival (OS) curves of all cases in the High‐TIL and Low‐TIL groups. The High‐TIL group has a better prognosis than the Low‐TIL group (p < 0.01).
Multivariate analysis for PFS and OS revealed that TIL status was a prognostic factor for PFS (hazard ratio [HR] (95% confidence interval [CI]) 0.61 (0.43–0.87); p < 0.01) and OS (HR (95% CI) 0.54 (0.33–0.86); p = 0.01) (Table 2). We also performed subgroup analyses according to histological subtype and MMR status. In aggressive type, TIL status was a significant prognostic factor for PFS and OS, whereas in non‐aggressive types, multivariate analysis did not show the prognostic significance of TILs status (Figures S5 and S6 and Tables S7 and S8). In addition, in cases with dMMR, TIL status was a significant prognostic factor for both PFS and OS, whereas in cases with pMMR, TIL status was significantly associated with PFS but not with OS (Figures S7 and S8 and Tables S9 and S10). Detailed results, including ROC curves, Kaplan–Meier analyses, and Cox regression models for each subgroup, are shown in the Supporting Information.
TABLE 2.
Univariate and multivariate analyses of progression‐free and overall survival rates for all cases.
| Variables | Progression‐free survival | Overall survival | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Univariate analysis | Multivariate analysis | Univariate analysis | Multivariate analysis | |||||||||
| HR (95% CI) | p | HR (95% CI) | p | HR (95% CI) | p | HR (95% CI) | p | |||||
| Age (years) | ||||||||||||
| ≥ 60 vs. < 60 | 2.60 | (1.81–3.75) | < 0.01 | 2.33 | (1.58–3.44) | < 0.01 | 2.12 | (1.35–3.34) | < 0.01 | 1.82 | (1.12–2.96) | 0.02 |
| Performance status | ||||||||||||
| 0, 1 vs. ≥ 2 | 0.29 | (0.11–0.54) | < 0.01 | 0.53 | (0.22–1.27) | 0.15 | 0.18 | (0.07–0.50) | < 0.01 | 0.73 | (0.24–2.22) | 0.57 |
| Histological subtypes | ||||||||||||
| Aggressive type vs. Non‐aggressive type | 4.24 | (3.01–5.98) | < 0.01 | 2.25 | (1.53–3.01) | < 0.01 | 4.32 | (2.76–6.77) | < 0.01 | 2.51 | (1.50–4.17) | < 0.01 |
| Myometrial invasion | ||||||||||||
| < 1/2 vs. ≥ 1/2 | 0.24 | (0.17–0.34) | < 0.01 | 0.61 | (0.39–0.94) | 0.03 | 0.25 | (0.16–0.40) | < 0.01 | 0.46 | (0.26–0.84) | 0.01 |
| Cervical involvement | ||||||||||||
| Positive vs. Negative | 2.17 | (1.47–3.24) | < 0.01 | 1.27 | (0.83–1.94) | 0.26 | 1.89 | (1.12–3.19) | 0.02 | 1.20 | (0.67–2.12) | 0.54 |
| Ovarian metastasis | ||||||||||||
| Positive vs. Negative | 5.75 | (3.87–8.53) | < 0.01 | 1.77 | (1.07–2.92) | 0.03 | 8.49 | (5.33–13.54) | < 0.01 | 2.82 | (1.54–5.17) | < 0.01 |
| Lymph node metastasis | ||||||||||||
| Positive vs. Negative | 5.30 | (3.73–7.53) | < 0.01 | 2.55 | (1.66–3.93) | < 0.01 | 3.65 | (2.29–5.83) | < 0.01 | 1.72 | (0.97–3.08) | 0.07 |
| Distant metastasis | ||||||||||||
| Positive vs. Negative | 13.12 | (8.69–19.81) | < 0.01 | 6.91 | (4.12–11.57) | < 0.01 | 13.56 | (8.35–22.02) | < 0.01 | 6.76 | (3.61–12.66) | < 0.01 |
| Lymphovascular invasion | ||||||||||||
| Positive vs. Negative | 4.15 | (2.88–5.96) | < 0.01 | 2.16 | (1.39–3.36) | < 0.01 | 3.28 | (2.07–5.21) | < 0.01 | 1.85 | (1.03–3.35) | 0.04 |
| Ascites or lavage cytology | ||||||||||||
| Positive vs. Negative | 4.53 | (3.17–6.45) | < 0.01 | 2.01 | (1.27–3.18) | < 0.01 | 5.78 | (3.72–8.97) | < 0.01 | 2.54 | (1.38–4.69) | < 0.01 |
| Adjuvant therapy | ||||||||||||
| Yes vs. No | 2.37 | (1.64–3.44) | < 0.01 | 0.46 | (0.28–0.74) | < 0.01 | 1.91 | (1.19–3.06) | < 0.01 | 0.26 | (0.14–0.49) | < 0.01 |
| TILs status | ||||||||||||
| High vs. Low | 0.56 | (0.40–0.79) | < 0.01 | 0.61 | (0.43–0.87) | < 0.01 | 0.55 | (0.35–0.85) | < 0.01 | 0.54 | (0.33–0.86) | 0.01 |
| MMR status | ||||||||||||
| Deficiency vs. proficiency | 1.21 | (0.86–1.70) | 0.26 | 1.42 | (0.92–2.20) | 0.11 | ||||||
| TILs and MMR interaction terms | ||||||||||||
| High‐TILs/MMR deficiency vs. others | 0.84 | (0.56–1.28) | 0.42 | 0.83 | (0.49–1.41) | 0.49 | ||||||
Abbreviations: CI, confidence interval; HR, hazard ratio; MMR, mismatch repair; TIL, tumor‐infiltrating lymphocyte.
4. Discussion
In the current study, we evaluated the AI‐based quantification of TILs in the area around the invasive frontline of EC and demonstrated that the High‐TIL group had better PFS and OS than the Low‐TIL group. Univariate and multivariate analyses revealed that TIL status was a prognostic factor for both PFS and OS.
Recently, the F1 score was used to assess the agreement between automatic detection using AI and manual annotation. The F1 score is the harmonic mean of precision (P), recall (R), and [2PR/(P + R)]. The value ranges between 0 and 1.0, and a value approaching 1.0 reflects high precision and recall [27]. In studies assessing TILs of malignant tumors, the F1 score has been used to determine the accuracy of the degree of agreement between lymphocytes detected automatically by AI and those manually annotated, yielding a score ranging between 0.78 and 0.97 [28, 29, 30]. In the current study, we also used the F1 score to evaluate the accuracy of the degree of agreement between automatic AI detection and manual annotation, resulting in an F1 score of 0.81. This value was within the range reported previously, indicating that its accuracy is retained.
In EC, TIL status was found to be associated with clinicopathological features, including FIGO stage, myometrial invasion, lymph node metastasis, distant metastasis, and lymphovascular invasion [15, 31, 32, 33, 34]. However, TILs were previously evaluated by human observers. In the current study, TILs were evaluated using AI, and patients in the High‐TIL group were diagnosed at an earlier FIGO stage and had a higher frequency of less than 50% of myometrial invasion than those in the Low‐TIL group. There was no discrepancy between the findings of previous studies and our study, underscoring the validity of AI‐based evaluation.
In addition, dMMR was identified in approximately 25%–30% of patients with EC and was associated with a high number of TILs [17, 35, 36]. In the current study, the rate of dMMR was higher than that reported previously, and dMMR was associated with TIL status. However, recent studies have shown that TIL status is associated with polymerase epsilon (POLE) status [37, 38]. Furthermore, several studies have reported the effect of POLE mutations on MMR function, as well as the possible co‐occurrence of dMMR and POLE mutations [39, 40, 41]. Additional studies are needed to investigate the relationship between TILs evaluated by AI, MMR status, and POLE status.
Nevertheless, the ideal prognostic model for evaluating any type of TIL in EC remains unclear. The clinical relevance of cluster of differentiation (CD) 4, CD8, and forkhead box P3 in EC has been reported [31, 32, 33]. Consistent with our study, previous studies have explored the clinical relevance of different TILs in EC [15, 31, 32, 33, 42]. Future studies should examine the clinical differences according to the type of AI‐detected TIL.
This study has some limitations. First, we evaluated TILs using HE slides, and orthogonal validation using other methods, such as deconvolution of bulk ribonucleic acid sequencing data or single‐cell and spatial transcriptomic analyses, was not performed. Second, this study was retrospective from a single institution, involving TIL evaluation. Third, the POLE status was not evaluated. Fourth, the deepest myometrial invasive frontline might not be the most appropriate site for evaluating TILs in cases with advanced‐stage, such as approaching the serosal surface or involving the cervix, lymph nodes, ovaries, and peritoneum. In addition, TILs at the metastatic sites or invasion frontline in the cervical invasive site were not assessed. Furthermore, the invasive frontline was manually annotated, and the inter‐observer reproducibility of this annotation process was not assessed. Those might represent a potential cause of inconsistency in the evaluation of TILs. Fifth, while this study demonstrated a high level of F1 score and the relationship between the TILs status and the MMR status, which indicated a high level of accuracy in the detection of lymphocytes, the dataset used to train the AI classifier was relatively small and did not consider the histological subtypes. Those might limit the model's consistency and ability. In addition, external validation using independent cohorts from other institutions was not performed. Furthermore, the analysis of features predicted by AI was limited to comparisons of clinicopathological characteristics, and other analyses considering tumor molecular profiles, key IHC markers, tumor mutation burden, or other data were not performed. Accordingly, additional studies are needed to compare HE and IHC staining for TIL evaluation, validate the observed findings through prospective studies from multiple institutions, evaluate the POLE status, conduct a more comprehensive evaluation of TILs across various stages, evaluate the consistency of invasive frontline annotation, include a large number of cases with considering the histological subtypes in training cohort, and perform analyses of features predicted by AI with molecular or genomic profiles. Furthermore, future studies should include external validation using independent cohorts and orthogonal validation using ribonucleic acid sequencing or single‐cell and spatial transcriptomic data to evaluate the performance of the AI‐based TILs assessment.
5. Conclusions
AI‐based evaluation of TILs in the area around the invasive frontline may be useful for predicting the prognosis of EC, and further studies are needed to establish methods for evaluating TILs in EC.
Author Contributions
Takahiro Einama: formal analysis, resources, software, writing – review and editing. Yuka Otsuka: data curation, writing – review and editing. Makiko Koga: formal analysis, software. Morikazu Miyamoto: conceptualization, methodology, writing – original draft, project administration, visualization. Hiroaki Soyama: writing – review and editing. Kohei Omatsu: writing – review and editing. Yoshinobu Hamada: writing – review and editing. Kento Kato: data curation, writing – review and editing. Kimiya Sato: resources, supervision, writing – review and editing. Masashi Takano: project administration, supervision, writing – review and editing. Taira Hada: conceptualization, methodology, validation, formal analysis, investigation, data curation, writing – original draft, visualization. Soichiro Kakimoto: writing – review and editing. Takanori Watanabe: data curation, writing – review and editing. Jin Suminokura: investigation, writing – review and editing. Tsubasa Ito: investigation, writing – review and editing. Naohisa Kishimoto: writing – review and editing. Risa Tanabe: writing – review and editing. Soko Nishimura: writing – review and editing.
Funding
The authors have nothing to report.
Ethics Statement
This study was approved by the Institutional Review Board of National Defense Medical College, Tokorozawa, Japan. The records and information of all female participants included in the study were anonymized and de‐identified prior to analysis.
Consent
The authors have nothing to report.
Conflicts of Interest
The authors declare no conflicts of interest.
Supporting information
Table S1: Univariate and Multivariate analysis of progression‐free survival and overall survival for all cases of endometrial carcinoma including tumor‐infiltrating lymphocytes evaluated using the area with the range of 500 μm toward the stromal side from the invasive frontline.
Table S2: Univariate and Multivariate analysis of progression‐free survival and overall survival for all cases of endometrial carcinoma including tumor‐infiltrating lymphocytes evaluated using the area with the range of 250 μm toward the stromal side from the invasive frontline.
Table S3: Univariate analysis of progression‐free survival and overall survival for all cases of endometrial carcinoma including the tumor‐infiltrative lymphocytes status classified based on the first quartile.
Table S4: Univariate and Multivariate analysis of progression‐free survival and overall survival for all cases of endometrial carcinoma including the tumor‐infiltrative lymphocytes status classified based on median number.
Table S5: Univariate and multivariate analysis of progression‐free survival and overall survival for all cases of endometrial carcinoma including the tumor‐infiltrative lymphocytes status classified based on the third quartile.
Table S6: Akaike and bayesian information criteria for multivariate analyses of progression‐free and overall survival according to tumor‐infiltrating lymphocyte status defined by different cut‐off values.
Table S7: Univariate and multivariate analysis of progression‐free survival and overall survival for cases of aggressive type endometrial carcinoma.
Table S8: Univariate and multivariate analysis of progression‐free survival and overall survival for cases of non‐aggressive type endometrial carcinoma.
Table S9: Univariate and multivariate analysis of progression‐free survival and overall survival for cases with mismatch repair deficiency.
Table S10: Univariate and multivariate analysis of progression‐free survival and overall survival for cases with mismatch repair proficiency.
Figure S1: Receiver operating characteristics (ROC) curve analysis of the average number of lymphocytes per area (/μm2) based on artificial intelligence for recurrence or progression. (a) ROC curve analysis of the average number of lymphocytes per area (/μm2) within the range of 500 μm toward the stromal side from the invasive frontline for recurrence or progression. The area under the curve is 0.570 and the cut‐off value was 0.00513. (b) ROC curve analysis of the average number of lymphocytes per area (/μm2) within the range of 250 μm toward the stromal side from the invasive frontline for recurrence or progression. The area under the curve is 0.573 and the cut‐off value was 0.00639.
Figure S2: Survival analysis of the High‐TIL and Low‐TIL groups. (a) Progression‐free survival (PFS) curves of all cases in the High‐TIL and Low‐TIL groups classified based on the area with the range of 500 μm toward the stromal side from the invasive frontline. The High‐TIL group has a better prognosis than the Low‐TIL group (p < 0.01). (b) Overall survival (OS) curves of all cases in the High‐TIL and Low‐TIL groups classified based on the area with the range of 500 μm toward the stromal side from the invasive frontline. The High‐TIL group has a better prognosis than the Low‐TIL group (p < 0.01). (c) PFS curves of all cases in the High‐TIL and Low‐TIL groups classified based on the area with the range of 250 μm toward the stromal side from the invasive frontline. The High‐TIL group has a better prognosis than the Low‐TIL group (p < 0.01). (d) OS curves of all cases in the High‐TIL and Low‐TIL groups classified based on the area with the range of 250 μm toward the stromal side from the invasive frontline. The High‐TIL group has a better prognosis than the Low‐TIL group (p < 0.01).
Figure S3: The lymphocytes density histogram. The first quartile, median, and third quartile numbers are 0.00301, 0.00493, and 0.00673, respectively.
Figure S4: Survival analysis of the High‐TIL and Low‐TIL groups. (a) Progression‐free survival (PFS) curves of all cases in the High‐TIL and Low‐TIL groups classified based on the first quartile number of the histogram of lymphocyte density. There is no prognostic significance between two groups (p = 0.48). (b) Overall survival (OS) curves of all cases in the High‐TIL and Low‐TIL groups classified based on the first quartile number of the histogram of lymphocyte density. There is no prognostic significance between two groups (p = 0.48). (c) PFS curves of all cases in the High‐TIL and Low‐TIL groups classified based on the median number of the histogram of lymphocyte density. The High‐TIL group has a better prognosis than the Low‐TIL group (p < 0.01). (d) OS curves of all cases in the High‐TIL and Low‐TIL groups classified based on the median number of the histogram of lymphocyte density. The High‐TIL group has a better prognosis than the Low‐TIL group (p < 0.01). (e) PFS curves of all cases in the High‐TIL and Low‐TIL groups classified based on the third quartile number of the histogram of lymphocyte density. The High‐TIL group has a better prognosis than the Low‐TIL group (p < 0.01). (f) OS curves of all cases in the High‐TIL and Low‐TIL groups classified based on the third quartile number of the histogram of lymphocyte density. The High‐TIL group has a better prognosis than the Low‐TIL group (p < 0.01).
Figure S5: Receiver operating characteristics (ROC) curve analysis of the average number of lymphocytes per area (/μm2) based on artificial intelligence for recurrence or progression. (a) ROC curve analysis of the average number of lymphocytes per area (/μm2) for recurrence or progression in cases with aggressive type endometrial carcinoma. The area under the curve is 0.614 and the cut‐off value was 0.00569. (b) ROC curve analysis of the average number of lymphocytes per area (/μm2) for recurrence or progression in cases with non‐aggressive type endometrial carcinoma. The area under the curve is 0.555 and the cut‐off value was 0.00478.
Figure S6: Survival analysis of the High‐TIL and Low‐TIL groups. (a) Progression‐free survival (PFS) curves of all cases in the High‐TIL and Low‐TIL groups in cases with aggressive type endometrial carcinoma. The High‐TIL group has a better prognosis than the Low‐TIL group (p < 0.01). (b) Overall survival (OS) curves of all cases in the High‐TIL and Low‐TIL groups in cases with aggressive type endometrial carcinoma. The High‐TIL group has a better prognosis than the Low‐TIL group (p < 0.01). (c) PFS curves of all cases in the High‐TIL and Low‐TIL groups in cases with non‐aggressive type endometrial carcinoma. The High‐TIL group has a better prognosis than the Low‐TIL group (p = 0.02). (d) OS curves of all cases in the High‐TIL and Low‐TIL groups in cases with non‐aggressive type endometrial carcinoma. There is no prognostic significance between two groups (p = 0.27).
Figure S7: Receiver operating characteristics (ROC) curve analysis of the average number of lymphocytes per area (/μm2) based on artificial intelligence for recurrence or progression. (a) ROC curve analysis of the average number of lymphocytes per area (/μm2) for recurrence or progression in cases with mismatch repair deficiency. The area under the curve is 0.597 and the cut‐off value was 0.00533. (b) ROC curve analysis of the average number of lymphocytes per area (/μm2) for recurrence or progression in cases with mismatch repair proficiency. The area under the curve is 0.558 and the cut‐off value was 0.00478.
Figure S8: Survival analysis of the High‐TIL and Low‐TIL groups. (a) Progression‐free survival (PFS) curves of all cases in the High‐TIL and Low‐TIL groups in cases with mismatch repair deficiency. The High‐TIL group has a better prognosis than the Low‐TIL group (p < 0.01). (b) Overall survival (OS) curves of all cases in the High‐TIL and Low‐TIL groups in cases with mismatch repair deficiency. The High‐TIL group has a better prognosis than the Low‐TIL group (p < 0.01). (c) PFS curves of all cases in the High‐TIL and Low‐TIL groups in cases with mismatch repair proficiency. The High‐TIL group has a better prognosis than the Low‐TIL group (p < 0.01). (d) OS curves of all cases in the High‐TIL and Low‐TIL groups in cases with mismatch repair proficiency. There is no prognostic significance between two groups (p = 0.06).
Acknowledgments
The authors would like to thank Editage (www.editage.jp) for the English language editing. Also, the authors appreciate Defense Medicine Basic Research Program B and Indica Labs Inc.
Data Availability Statement
The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.
References
- 1. Bray F., Laversanne M., Sung H., et al., “Global Cancer Statistics 2022: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries,” CA: A Cancer Journal for Clinicians 74, no. 3 (2024): 229–263, 10.3322/caac.21834. [DOI] [PubMed] [Google Scholar]
- 2. Abu‐Rustum N., Yashar C., Arend R., et al., “Uterine Neoplasms, Version 1.2023, NCCN Clinical Practice Guidelines in Oncology,” Journal of the National Comprehensive Cancer Network 21, no. 2 (2023): 181–209, 10.6004/jnccn.2023.0006. [DOI] [PubMed] [Google Scholar]
- 3. Creutzberg C. L., van Stiphout R. G. P. M., Nout R. A., et al., “Nomograms for Prediction of Outcome With or Without Adjuvant Radiation Therapy for Patients With Endometrial Cancer: A Pooled Analysis of PORTEC‐1 and PORTEC‐2 Trials,” International Journal of Radiation Oncology, Biology, Physics 91, no. 3 (2015): 530–539, 10.1016/j.ijrobp.2014.11.022. [DOI] [PubMed] [Google Scholar]
- 4. SGO Clinical Practice Endometrial Cancer Working Group , Burke W. M., Orr J., et al., “Endometrial Cancer: A Review and Current Management Strategies: Part II,” Gynecologic Oncology 134, no. 2 (2014): 393–402, 10.1016/j.ygyno.2014.06.003. [DOI] [PubMed] [Google Scholar]
- 5. Creutzberg C. L., van Putten W. L., Koper P. C., et al., “Surgery and Postoperative Radiotherapy Versus Surgery Alone for Patients With Stage‐1 Endometrial Carcinoma: Multicentre Randomised Trial. PORTEC Study Group. Post Operative Radiation Therapy in Endometrial Carcinoma,” Lancet 355, no. 9213 (2000): 1404–1411, 10.1016/s0140-6736(00)02139-5. [DOI] [PubMed] [Google Scholar]
- 6. Creutzberg C. L., Nout R. A., Lybeert M. L. M., et al., “Fifteen‐Year Radiotherapy Outcomes of the Randomized PORTEC‐1 Trial for Endometrial Carcinoma,” International Journal of Radiation Oncology, Biology, Physics 81, no. 4 (2011): e631–e638, 10.1016/j.ijrobp.2011.04.013. [DOI] [PubMed] [Google Scholar]
- 7. Hendry S., Salgado R., Gevaert T., et al., “Assessing Tumor‐Infiltrating Lymphocytes in Solid Tumors: A Practical Review for Pathologists and Proposal for a Standardized Method From the International Immuno‐Oncology Biomarkers Working Group: Part 2: TILs in Melanoma, Gastrointestinal Tract Carcinomas, Non‐Small Cell Lung Carcinoma and Mesothelioma, Endometrial and Ovarian Carcinomas, Squamous Cell Carcinoma of the Head and Neck, Genitourinary Carcinomas, and Primary Brain Tumors,” Advances in Anatomic Pathology 24, no. 6 (2017): 311–335, 10.1097/PAP.0000000000000161. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Schumacher K., Haensch W., Röefzaad C., and Schlag P. M., “Prognostic Significance of Activated CD8(+) T Cell Infiltrations Within Esophageal Carcinomas,” Cancer Research 61, no. 10 (2001): 3932–3936. [PubMed] [Google Scholar]
- 9. Marrogi A. J., Munshi A., Merogi A. J., et al., “Study of Tumor Infiltrating Lymphocytes and Transforming Growth Factor‐Beta as Prognostic Factors in Breast Carcinoma,” International Journal of Cancer 74, no. 5 (1997): 492–501, 10.1002/(sici)1097-0215(19971021)74:5<492::aid-ijc3>3.0.co;2-z. [DOI] [PubMed] [Google Scholar]
- 10. Vesalainen S., Lipponen P., Talja M., and Syrjänen K., “Histological Grade, Perineural Infiltration, Tumour‐Infiltrating Lymphocytes and Apoptosis as Determinants of Long‐Term Prognosis in Prostatic Adenocarcinoma,” European Journal of Cancer 30A, no. 12 (1994): 1797–1803, 10.1016/0959-8049(94)e0159-2. [DOI] [PubMed] [Google Scholar]
- 11. Naito Y., Saito K., Shiiba K., et al., “CD8+ T Cells Infiltrated Within Cancer Cell Nests as a Prognostic Factor in Human Colorectal Cancer,” Cancer Research 58, no. 16 (1998): 3491–3494. [PubMed] [Google Scholar]
- 12. Zhang L., Conejo‐Garcia J. R., Katsaros D., et al., “Intratumoral T Cells, Recurrence, and Survival in Epithelial Ovarian Cancer,” New England Journal of Medicine 348, no. 3 (2003): 203–213, 10.1056/NEJMoa020177. [DOI] [PubMed] [Google Scholar]
- 13. Guo F., Dong Y., Tan Q., Kong J., and Yu B., “Tissue Infiltrating Immune Cells as Prognostic Biomarkers in Endometrial Cancer: A Meta‐Analysis,” Disease Markers 2020 (2020): 1805764, 10.1155/2020/1805764. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Miyamoto M., Takano M., Tsuda H., et al., “The Haphazard Pattern in Grade‐3 Endometrioid Carcinoma Is Associated With Poor Prognosis and Tumor Lymphocyte Infiltration,” Pathology Oncology Research 26, no. 2 (2020): 783–790, 10.1007/s12253-019-00624-1. [DOI] [PubMed] [Google Scholar]
- 15. Miyamoto M., Hada T., Ishibashi H., et al., “A New Model to Improve the Prediction of Prognosis of Endometrial Carcinoma by Combining Traditional Classification With the Presence of Tumor‐Infiltrating Lymphocytes,” Anticancer Research 41, no. 2 (2021): 1047–1053, 10.21873/anticanres.14861. [DOI] [PubMed] [Google Scholar]
- 16. Kono‐Sato T., Miyai K., Yamagishi Y., et al., “Intraepithelial Lymphocytes Are Indicators of Better Prognosis in Surgically Resected Endometrioid‐Type Endometrial Carcinomas at Early and Advanced Stages,” BMC Cancer 22, no. 1 (2022): 361, 10.1186/s12885-022-09363-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Bounous V. E., Ferrero A., Campisi P., et al., “Immunohistochemical Markers and TILs Evaluation for Endometrial Carcinoma,” Journal of Clinical Medicine 11, no. 19 (2022): 5678, 10.3390/jcm11195678. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Wang S., Yang D. M., Rong R., Zhan X., and Xiao G., “Pathology Image Analysis Using Segmentation Deep Learning Algorithms,” American Journal of Pathology 189, no. 9 (2019): 1686–1698, 10.1016/j.ajpath.2019.05.007. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Ehteshami Bejnordi B., Veta M., Johannes van Diest P., et al., “Diagnostic Assessment of Deep Learning Algorithms for Detection of Lymph Node Metastases in Women With Breast Cancer,” Journal of the American Medical Association 318, no. 22 (2017): 2199–2210, 10.1001/jama.2017.14585. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Kumar N., Verma R., Sharma S., Bhargava S., Vahadane A., and Sethi A., “A Dataset and a Technique for Generalized Nuclear Segmentation for Computational Pathology,” IEEE Transactions on Medical Imaging 36, no. 7 (2017): 1550–1560, 10.1109/TMI.2017.2677499. [DOI] [PubMed] [Google Scholar]
- 21. Kouzu K., Nearchou I. P., Kajiwara Y., et al., “Deep‐Learning‐Based Classification of Desmoplastic Reaction on H&E Predicts Poor Prognosis in Oesophageal Squamous Cell Carcinoma,” Histopathology 81, no. 2 (2022): 255–263, 10.1111/his.14708. [DOI] [PubMed] [Google Scholar]
- 22. Nearchou I. P., Ueno H., Kajiwara Y., et al., “Automated Detection and Classification of Desmoplastic Reaction at the Colorectal Tumour Front Using Deep Learning,” Cancers (Basel) 13, no. 7 (2021): 1615, 10.3390/cancers13071615. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Hamada K., Murakami R., Ueda A., et al., “A Deep Learning‐Based Assessment Pipeline for Intraepithelial and Stromal Tumor‐Infiltrating Lymphocytes in High‐Grade Serous Ovarian Carcinoma,” American Journal of Pathology 194, no. 7 (2024): 1272–1284, 10.1016/j.ajpath.2024.02.016. [DOI] [PubMed] [Google Scholar]
- 24. Lokuhetty D., White V. A., and Cree I. A., WHO Classification of Tumours of Female Reproductive Organs, 5th ed. (IARC, 2020). [Google Scholar]
- 25. Berek J. S., Matias‐Guiu X., Creutzberg C., et al., “FIGO Staging of Endometrial Cancer: 2023,” International Journal of Gynaecology and Obstetrics 162, no. 2 (2023): 383–394, 10.1002/ijgo.14923. [DOI] [PubMed] [Google Scholar]
- 26. Kakimoto S., Miyamoto M., Einama T., et al., “Co‐Expression of Mesothelin and CA125 Is Associated With the Poor Prognosis of Endometrial Serous Carcinoma and Mixed Carcinomas Including Serous Carcinoma,” Pathology Oncology Research 26, no. 4 (2020): 2299–2306, 10.1007/s12253-020-00823-1. [DOI] [PubMed] [Google Scholar]
- 27. Ghosh A. K., Thammasudjarit R., Jongkhajornpong P., Attia J., and Thakkinstian A., “Deep Learning for Discrimination Between Fungal Keratitis and Bacterial Keratitis: DeepKeratitis,” Cornea 41, no. 5 (2022): 616–622, 10.1097/ICO.0000000000002830. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Saha M., Guo X., and Sharma A., “ TilGAN: GAN for Facilitating Tumor‐Infiltrating Lymphocyte Pathology Image Synthesis With Improved Image Classification,” IEEE Access 9 (2021): 79829–79840, 10.1109/access.2021.3084597. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Barua B., Chyrmang G., Bora K., Ahmed G. N., Kakoti L., and Saikia M. J., “Classifying Tumour Infiltrating Lymphocytes in Oral Squamous Cell Carcinoma Histopathology Using Joint Learning Framework,” Scientific Reports 15, no. 1 (2025): 2938, 10.1038/s41598-025-86527-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. Swiderska‐Chadaj Z., Pinckaers H., van Rijthoven M., et al., “Learning to Detect Lymphocytes in Immunohistochemistry With Deep Learning,” Medical Image Analysis 58 (2019): 101547, 10.1016/j.media.2019.101547. [DOI] [PubMed] [Google Scholar]
- 31. de Jong R. A., Leffers N., Boezen H. M., et al., “Presence of Tumor‐Infiltrating Lymphocytes Is an Independent Prognostic Factor in Type I and II Endometrial Cancer,” Gynecologic Oncology 114, no. 1 (2009): 105–110, 10.1016/j.ygyno.2009.03.022. [DOI] [PubMed] [Google Scholar]
- 32. Suto A., Minaguchi T., Qi N., et al., “CD25+FOXP3+CD45RA−regulatory T‐Cell Infiltration as a Prognostic Biomarker for Endometrial Carcinoma,” BMC Cancer 24, no. 1 (2024): 1100, 10.1186/s12885-024-12851-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33. Jung I. K., Kim S. S., Suh D. S., Kim K. H., Lee C. H., and Yoon M. S., “Tumor‐Infiltration of T‐Lymphocytes Is Inversely Correlated With Clinicopathologic Factors in Endometrial Adenocarcinoma,” Obstetrics & Gynecology Science 57, no. 4 (2014): 266–273, 10.5468/ogs.2014.57.4.266. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34. Ozturk C., Askan G., Ozturk S. D., Okcu O., Sen B., and Bedir R., “High Tumor Infiltrating Lymphocytes Are Associated With Overall Survival and Good Prognostic Parameters in Endometrial Endometrioid Carcinoma Patients,” Turk Patoloji Dergisi 39, no. 1 (2023): 75–82, 10.5146/tjpath.2022.01596. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35. Addante F., d'Amati A., Santoro A., et al., “Mismatch Repair Deficiency as a Predictive and Prognostic Biomarker in Endometrial Cancer: A Review on Immunohistochemistry Staining Patterns and Clinical Implications,” International Journal of Molecular Sciences 25, no. 2 (2024): 1056, 10.3390/ijms25021056. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36. MacDonald N. D., Salvesen H. B., Ryan A., Iversen O. E., Akslen L. A., and Jacobs I. J., “Frequency and Prognostic Impact of Microsatellite Instability in a Large Population‐Based Study of Endometrial Carcinomas,” Cancer Research 60, no. 6 (2000): 1750–1752. [PubMed] [Google Scholar]
- 37. Howitt B. E., Shukla S. A., Sholl L. M., et al., “Association of Polymerase e‐Mutated and Microsatellite‐Instable Endometrial Cancers With Neoantigen Load, Number of Tumor‐Infiltrating Lymphocytes, and Expression of PD‐1 and PD‐L1,” JAMA Oncology 1, no. 9 (2015): 1319–1323, 10.1001/jamaoncol.2015.2151. [DOI] [PubMed] [Google Scholar]
- 38. van Gool I. C., Eggink F. A., Freeman‐Mills L., et al., “POLE Proofreading Mutations Elicit an Antitumor Immune Response in Endometrial Cancer,” Clinical Cancer Research 21, no. 14 (2015): 3347–3355, 10.1158/1078-0432.CCR-15-0057. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39. Hodel K. P., Sun M. J. S., Ungerleider N., et al., “POLE Mutation Spectra Are Shaped by the Mutant Allele Identity, Its Abundance, and Mismatch Repair Status,” Molecular Cell 78, no. 6 (2020): 1166–1177.e6, 10.1016/j.molcel.2020.05.012. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40. Castellsagué E., Li R., Aligue R., et al., “Novel POLE Pathogenic Germline Variant in a Family With Multiple Primary Tumors Results in Distinct Mutational Signatures,” Human Mutation 40, no. 1 (2019): 36–41, 10.1002/humu.23676. [DOI] [PubMed] [Google Scholar]
- 41. Shinbrot E., Henninger E. E., Weinhold N., et al., “Exonuclease Mutations in DNA Polymerase Epsilon Reveal Replication Strand Specific Mutation Patterns and Human Origins of Replication,” Genome Research 24, no. 11 (2014): 1740–1750, 10.1101/gr.174789.114. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 42. López‐Janeiro Á., Villalba‐Esparza M., Brizzi M. E., et al., “The Association Between the Tumor Immune Microenvironments and Clinical Outcome in Low‐Grade, Early‐Stage Endometrial Cancer Patients,” Journal of Pathology 258, no. 4 (2022): 426–436, 10.1002/path.6012. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Table S1: Univariate and Multivariate analysis of progression‐free survival and overall survival for all cases of endometrial carcinoma including tumor‐infiltrating lymphocytes evaluated using the area with the range of 500 μm toward the stromal side from the invasive frontline.
Table S2: Univariate and Multivariate analysis of progression‐free survival and overall survival for all cases of endometrial carcinoma including tumor‐infiltrating lymphocytes evaluated using the area with the range of 250 μm toward the stromal side from the invasive frontline.
Table S3: Univariate analysis of progression‐free survival and overall survival for all cases of endometrial carcinoma including the tumor‐infiltrative lymphocytes status classified based on the first quartile.
Table S4: Univariate and Multivariate analysis of progression‐free survival and overall survival for all cases of endometrial carcinoma including the tumor‐infiltrative lymphocytes status classified based on median number.
Table S5: Univariate and multivariate analysis of progression‐free survival and overall survival for all cases of endometrial carcinoma including the tumor‐infiltrative lymphocytes status classified based on the third quartile.
Table S6: Akaike and bayesian information criteria for multivariate analyses of progression‐free and overall survival according to tumor‐infiltrating lymphocyte status defined by different cut‐off values.
Table S7: Univariate and multivariate analysis of progression‐free survival and overall survival for cases of aggressive type endometrial carcinoma.
Table S8: Univariate and multivariate analysis of progression‐free survival and overall survival for cases of non‐aggressive type endometrial carcinoma.
Table S9: Univariate and multivariate analysis of progression‐free survival and overall survival for cases with mismatch repair deficiency.
Table S10: Univariate and multivariate analysis of progression‐free survival and overall survival for cases with mismatch repair proficiency.
Figure S1: Receiver operating characteristics (ROC) curve analysis of the average number of lymphocytes per area (/μm2) based on artificial intelligence for recurrence or progression. (a) ROC curve analysis of the average number of lymphocytes per area (/μm2) within the range of 500 μm toward the stromal side from the invasive frontline for recurrence or progression. The area under the curve is 0.570 and the cut‐off value was 0.00513. (b) ROC curve analysis of the average number of lymphocytes per area (/μm2) within the range of 250 μm toward the stromal side from the invasive frontline for recurrence or progression. The area under the curve is 0.573 and the cut‐off value was 0.00639.
Figure S2: Survival analysis of the High‐TIL and Low‐TIL groups. (a) Progression‐free survival (PFS) curves of all cases in the High‐TIL and Low‐TIL groups classified based on the area with the range of 500 μm toward the stromal side from the invasive frontline. The High‐TIL group has a better prognosis than the Low‐TIL group (p < 0.01). (b) Overall survival (OS) curves of all cases in the High‐TIL and Low‐TIL groups classified based on the area with the range of 500 μm toward the stromal side from the invasive frontline. The High‐TIL group has a better prognosis than the Low‐TIL group (p < 0.01). (c) PFS curves of all cases in the High‐TIL and Low‐TIL groups classified based on the area with the range of 250 μm toward the stromal side from the invasive frontline. The High‐TIL group has a better prognosis than the Low‐TIL group (p < 0.01). (d) OS curves of all cases in the High‐TIL and Low‐TIL groups classified based on the area with the range of 250 μm toward the stromal side from the invasive frontline. The High‐TIL group has a better prognosis than the Low‐TIL group (p < 0.01).
Figure S3: The lymphocytes density histogram. The first quartile, median, and third quartile numbers are 0.00301, 0.00493, and 0.00673, respectively.
Figure S4: Survival analysis of the High‐TIL and Low‐TIL groups. (a) Progression‐free survival (PFS) curves of all cases in the High‐TIL and Low‐TIL groups classified based on the first quartile number of the histogram of lymphocyte density. There is no prognostic significance between two groups (p = 0.48). (b) Overall survival (OS) curves of all cases in the High‐TIL and Low‐TIL groups classified based on the first quartile number of the histogram of lymphocyte density. There is no prognostic significance between two groups (p = 0.48). (c) PFS curves of all cases in the High‐TIL and Low‐TIL groups classified based on the median number of the histogram of lymphocyte density. The High‐TIL group has a better prognosis than the Low‐TIL group (p < 0.01). (d) OS curves of all cases in the High‐TIL and Low‐TIL groups classified based on the median number of the histogram of lymphocyte density. The High‐TIL group has a better prognosis than the Low‐TIL group (p < 0.01). (e) PFS curves of all cases in the High‐TIL and Low‐TIL groups classified based on the third quartile number of the histogram of lymphocyte density. The High‐TIL group has a better prognosis than the Low‐TIL group (p < 0.01). (f) OS curves of all cases in the High‐TIL and Low‐TIL groups classified based on the third quartile number of the histogram of lymphocyte density. The High‐TIL group has a better prognosis than the Low‐TIL group (p < 0.01).
Figure S5: Receiver operating characteristics (ROC) curve analysis of the average number of lymphocytes per area (/μm2) based on artificial intelligence for recurrence or progression. (a) ROC curve analysis of the average number of lymphocytes per area (/μm2) for recurrence or progression in cases with aggressive type endometrial carcinoma. The area under the curve is 0.614 and the cut‐off value was 0.00569. (b) ROC curve analysis of the average number of lymphocytes per area (/μm2) for recurrence or progression in cases with non‐aggressive type endometrial carcinoma. The area under the curve is 0.555 and the cut‐off value was 0.00478.
Figure S6: Survival analysis of the High‐TIL and Low‐TIL groups. (a) Progression‐free survival (PFS) curves of all cases in the High‐TIL and Low‐TIL groups in cases with aggressive type endometrial carcinoma. The High‐TIL group has a better prognosis than the Low‐TIL group (p < 0.01). (b) Overall survival (OS) curves of all cases in the High‐TIL and Low‐TIL groups in cases with aggressive type endometrial carcinoma. The High‐TIL group has a better prognosis than the Low‐TIL group (p < 0.01). (c) PFS curves of all cases in the High‐TIL and Low‐TIL groups in cases with non‐aggressive type endometrial carcinoma. The High‐TIL group has a better prognosis than the Low‐TIL group (p = 0.02). (d) OS curves of all cases in the High‐TIL and Low‐TIL groups in cases with non‐aggressive type endometrial carcinoma. There is no prognostic significance between two groups (p = 0.27).
Figure S7: Receiver operating characteristics (ROC) curve analysis of the average number of lymphocytes per area (/μm2) based on artificial intelligence for recurrence or progression. (a) ROC curve analysis of the average number of lymphocytes per area (/μm2) for recurrence or progression in cases with mismatch repair deficiency. The area under the curve is 0.597 and the cut‐off value was 0.00533. (b) ROC curve analysis of the average number of lymphocytes per area (/μm2) for recurrence or progression in cases with mismatch repair proficiency. The area under the curve is 0.558 and the cut‐off value was 0.00478.
Figure S8: Survival analysis of the High‐TIL and Low‐TIL groups. (a) Progression‐free survival (PFS) curves of all cases in the High‐TIL and Low‐TIL groups in cases with mismatch repair deficiency. The High‐TIL group has a better prognosis than the Low‐TIL group (p < 0.01). (b) Overall survival (OS) curves of all cases in the High‐TIL and Low‐TIL groups in cases with mismatch repair deficiency. The High‐TIL group has a better prognosis than the Low‐TIL group (p < 0.01). (c) PFS curves of all cases in the High‐TIL and Low‐TIL groups in cases with mismatch repair proficiency. The High‐TIL group has a better prognosis than the Low‐TIL group (p < 0.01). (d) OS curves of all cases in the High‐TIL and Low‐TIL groups in cases with mismatch repair proficiency. There is no prognostic significance between two groups (p = 0.06).
Data Availability Statement
The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.