Dataset Information

Development and validation of deep learning classifiers to detect Epstein-Barr virus and microsatellite instability status in gastric cancer: a retrospective multicentre cohort study.

ABSTRACT:

Background

Response to immunotherapy in gastric cancer is associated with microsatellite instability (or mismatch repair deficiency) and Epstein-Barr virus (EBV) positivity. We therefore aimed to develop and validate deep learning-based classifiers to detect microsatellite instability and EBV status from routine histology slides.

Methods

In this retrospective, multicentre study, we collected tissue samples from ten cohorts of patients with gastric cancer from seven countries (South Korea, Switzerland, Japan, Italy, Germany, the UK and the USA). We trained a deep learning-based classifier to detect microsatellite instability and EBV positivity from digitised, haematoxylin and eosin stained resection slides without annotating tumour containing regions. The performance of the classifier was assessed by within-cohort cross-validation in all ten cohorts and by external validation, for which we split the cohorts into a five-cohort training dataset and a five-cohort test dataset. We measured the area under the receiver operating curve (AUROC) for detection of microsatellite instability and EBV status. Microsatellite instability and EBV status were determined to be detectable if the lower bound of the 95% CI for the AUROC was above 0·5.

Findings

Across the ten cohorts, our analysis included 2823 patients with known microsatellite instability status and 2685 patients with known EBV status. In the within-cohort cross-validation, the deep learning-based classifier could detect microsatellite instability status in nine of ten cohorts, with AUROCs ranging from 0·597 (95% CI 0·522-0·737) to 0·836 (0·795-0·880) and EBV status in five of eight cohorts, with AUROCs ranging from 0·819 (0·752-0·841) to 0·897 (0·513-0·966). Training a classifier on the pooled training dataset and testing it on the five remaining cohorts resulted in high classification performance with AUROCs ranging from 0·723 (95% CI 0·676-0·794) to 0·863 (0·747-0·969) for detection of microsatellite instability and from 0·672 (0·403-0·989) to 0·859 (0·823-0·919) for detection of EBV status.

Interpretation

Classifiers became increasingly robust when trained on pooled cohorts. After prospective validation, this deep learning-based tissue classification system could be used as an inexpensive predictive biomarker for immunotherapy in gastric cancer.

Funding

German Cancer Aid and German Federal Ministry of Health.

SUBMITTER: Muti HS

PROVIDER: S-EPMC8460994 | biostudies-literature | 2021 Oct

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Development and validation of deep learning classifiers to detect Epstein-Barr virus and microsatellite instability status in gastric cancer: a retrospective multicentre cohort study.

Muti Hannah Sophie HS Heij Lara Rosaline LR Keller Gisela G Kohlruss Meike M Langer Rupert R Dislich Bastian B Cheong Jae-Ho JH Kim Young-Woo YW Kim Hyunki H Kook Myeong-Cherl MC Cunningham David D Allum William H WH Langley Ruth E RE Nankivell Matthew G MG Quirke Philip P Hayden Jeremy D JD West Nicholas P NP Irvine Andrew J AJ Yoshikawa Takaki T Oshima Takashi T Huss Ralf R Grosser Bianca B Roviello Franco F d'Ignazio Alessia A Quaas Alexander A Alakus Hakan H Tan Xiuxiang X Pearson Alexander T AT Luedde Tom T Ebert Matthias P MP Jäger Dirk D Trautwein Christian C Gaisa Nadine Therese NT Grabsch Heike I HI Kather Jakob Nikolas JN

The Lancet. Digital health 20210817 10

<h4>Background</h4>Response to immunotherapy in gastric cancer is associated with microsatellite instability (or mismatch repair deficiency) and Epstein-Barr virus (EBV) positivity. We therefore aimed to develop and validate deep learning-based classifiers to detect microsatellite instability and EBV status from routine histology slides.<h4>Methods</h4>In this retrospective, multicentre study, we collected tissue samples from ten cohorts of patients with gastric cancer from seven countries (Sout ...[more]

PMID: 34417147

Similar Datasets

Project description:BackgroundThe prognostic potential of PD-L1 is currently unclear in gastric carcinomas, although the immune checkpoint PD-1/PD-L1 inhibitors have produced promising results in clinical trials.MethodsWe explored the prognostic implications of programmed death ligand 1 (PD-L1) in 514 consecutive surgically-resected gastric carcinomas. Overall survival and recurrence-free survival were evaluated. Immunohistochemistry for PD-L1, CD8, FOXP3, and PD-1, and molecular grouping by in situ hybridization for Epstein-Barr virus (EBV)-encoded small RNAs and multiplex PCR for microsatellite instability (MSI) markers were performed. Additionally, to explore the function inherent to PD-L1, PD-L1-specific siRNA transfection, cell proliferation, invasion, migration and apoptosis assays were conducted in five gastric carcinoma cell lines.ResultsPD-L1(+) tumor and immune cells were observed in 101 (20%) and 244 patients (47%), respectively. "Tumoral PD-L1(+)/immune cell PD-L1(-)/CD8+/low tumor-infiltrating lymphocytes (TILs)," and more advanced-stage tumors were associated with unfavorable clinical outcomes in the entire cohort through multivariate analysis. Furthermore, tumoral PD-L1(+)/FOXP3+/low TILs were associated with worse clinical outcomes in EBV-positive and MSI-high carcinomas. Tumoral PD-L1(+) alone was an adverse prognostic factor in EBV-positive carcinomas, but not in MSI-high carcinomas, whereas PD-L1(+) immune cells or FOXP3+/high TILs alone were correlated with a favorable prognosis. PD-L1 knockdown in gastric carcinoma cells suppressed cell proliferation, invasion and migration, and increased apoptosis, which were all statistically significant in two EBV(+) cell lines, but not all in three EBV(-) cell lines.ConclusionsThe prognostic impact of PD-L1 may depend on the tumor microenvironment, and statuses of EBV and MSI, although PD-L1 innately promotes cancer cell survival in cell-based assays. The combination of "tumoral PD-L1/immune cell PD-L1/CD8+ TILs" may serve as an independent prognostic factor. Tumoral PD-L1(+)/immune cell PD-L1(-)/CD8+/low TILs showing a worse prognosis may be beneficial for combinatorial therapies of anti-PD-L1/PD-1 and anti-cytotoxic T-lymphocyte associated antigen 4 (CTLA4) that would promote effector T cells, thus attack the tumor.

Project description:ImportanceEpstein-Barr virus (EBV)-associated gastric cancer (EBV-GC) is 1 of 4 molecular subtypes of GC and is confirmed by an expensive molecular test, EBV-encoded small RNA in situ hybridization. EBV-GC has 2 histologic characteristics, lymphoid stroma and lace-like tumor pattern, but projecting EBV-GC at biopsy is difficult even for experienced pathologists.ObjectiveTo develop and validate a deep learning algorithm to predict EBV status from pathology images of GC biopsy.Design, setting, and participantsThis diagnostic study developed a deep learning classifier to predict EBV-GC using image patches of tissue microarray (TMA) and whole slide images (WSIs) of GC and applied it to GC biopsy specimens from GCs diagnosed at Kangbuk Samsung Hospital between 2011 and 2020. For a quantitative evaluation and EBV-GC prediction on biopsy specimens, the area of each class and the fraction in total tissue or tumor area were calculated. Data were analyzed from March 5, 2021, to February 10, 2022.Main outcomes and measuresEvaluation metrics of predictive model performance were assessed on accuracy, recall, precision, F1 score, area under the receiver operating characteristic curve (AUC), and κ coefficient.ResultsThis study included 137 184 image patches from 16 TMAs (708 tissue cores), 24 WSIs, and 286 biopsy images of GC. The classifier was able to classify EBV-GC image patches from TMAs and WSIs with 94.70% accuracy, 0.936 recall, 0.938 precision, 0.937 F1 score, and 0.909 κ coefficient. The classifier was used for predicting and measuring the area and fraction of EBV-GC on biopsy tissue specimens. A 10% cutoff value for the predicted fraction of EBV-GC to tissue (EBV-GC/tissue area) produced the best prediction results in EBV-GC biopsy specimens and showed the highest AUC value (0.8723; 95% CI, 0.7560-0.9501). That cutoff also obtained high sensitivity (0.895) and moderate specificity (0.745) compared with experienced pathologist sensitivity (0.842) and specificity (0.854) when using the presence of lymphoid stroma and a lace-like pattern as diagnostic criteria. On prediction maps, EBV-GCs with lace-like pattern and lymphoid stroma showed the same prediction results as EBV-GC, but cases lacking these histologic features revealed heterogeneous prediction results of EBV-GC and non-EBV-GC areas.Conclusions and relevanceThis study showed the feasibility of EBV-GC prediction using a deep learning algorithm, even in biopsy samples. Use of such an image-based classifier before a confirmatory molecular test will reduce costs and tissue waste.

Project description:Gastric cancer is a heterogenous disease with different phenotypes, genotypes, and clinical outcomes, including sensitivity to treatments and prognoses. Recent medical advances have enabled the classification of this heterogenous disease into several groups and the consequent analysis of their clinicopathological characteristics. Gastric cancer associated with Epstein-Barr virus (EBV) and microsatellite-unstable tumors are considered to be the two major subtypes as they are clearly defined by well-established methodologies, such as in situ hybridization and polymerase chain reaction-based analyses, respectively. However, discrepancies in the histological diagnosis of gastric neoplasms remain problematic, and international harmonization should be performed to improve our understanding of gastric carcinogenesis. We re-evaluated Japanese cases of early gastric cancer according to the current World Health Organization (WHO) criteria and classified them into genomic subtypes based on microsatellite instability (MSI) and EBV positivity to determine the initial genetic events in gastric carcinogenesis. A total of 113 Japanese early gastric cancers (including low- and high-grade dysplasias) treated with endoscopic resection over 5 years were archived in our hospital. A histological re-evaluation according to the WHO criteria revealed 54 adenocarcinomas, which were divided into 6 EBV-positive (11.1%), 7 MSI-high (MSI-H, 13.0%), and 41 microsatellite stable cases (75.9%). MSI-H adenocarcinoma was confirmed by an immunohistochemistry assay of mismatch repair proteins. Programmed death-ligand 1 immunostaining with two antibodies (E1L3N and SP263) was positive in tumor cells of one MSI-H adenocarcinoma case (1/7, 14.3%). The proportion of stained cells was higher with clone SP263 than with E1L3N. Histologically, EBV-positive carcinomas were poorly differentiated (83.8%), and MSI-H cancers were frequent in well to moderately differentiated adenocarcinoma (85.7%), indicating that the EBV-positive subtype presented with high-grade morphology even when an early lesion. Our study indicates that the WHO criteria are useful for subdividing Japanese early gastric cancers, and this subdivision may be useful for comparative analysis of precursor lesions and early carcinoma.

Project description:Epstein-Barr virus positivity (EBV(+)) and high-microsatellite instability (MSI-H) have been identified as molecular subgroups in gastric carcinoma. The aim of our study was to determine the prognostic and predictive relevance of these subgroups in the context of platinum/5-fluorouracil (5-FU) based preoperative chemotherapy (CTx). Additionally, we investigated the clinical relevance of the low-MSI (MSI-L) phenotype. We analysed 760 adenocarcinomas of the stomach or the gastro-oesophageal junction encompassing 143 biopsies before CTx and 617 resected tumours (291 without and 326 after CTx). EBV was determined by PCR and in situ hybridisation for selected cases. MSI was analysed by PCR using five microsatellite markers and classified as MSI-H and MSI-L. Frequencies of EBV(+), MSI-H and MSI-L in the biopsies before CTx were 4.2, 10.5 and 4.9% respectively. EBV(+) or MSI-H did not correlate with response, but MSI-L was associated with better response (p = 0.011). In the resected tumours, frequencies of EBV(+), MSI-H and MSI-L were 3.9, 9.6 and 4.5% respectively. Overall survival (OS) was significantly different in the non-CTx group (p = 0.014). Patients with EBV(+) tumours showed the best OS, followed by MSI-H. MSI-L was significantly associated with worse OS (hazard ratio [HR], 2.21; 95% confidence interval [CI], 1.21-4.04, p = 0.01). In the resected tumours after CTx, MSI-H was also associated with increased OS (HR, 0.54; 95% CI, 0.26-1.09, p = 0.085). In multivariable analysis, molecular classification was an independent prognostic factor in the completely resected (R0) non-CTx group (p = 0.035). In conclusion, MSI-H and EBV(+) are not predictive of response to neoadjuvant platinum/5-FU based CTx, but they are indicative of a good prognosis. In particular, MSI-H indicates a favourable prognosis irrespective of treatment with CTx. MSI-L predicts good response to CTx and its negative prognostic effect for patients treated with surgery alone suggests that MSI-L might help to identify patients with potentially high-benefit from preoperative CTx.

Project description:BackgroundMicrosatellite instability (MSI) and Epstein-Barr virus (EBV)-positive molecular subtypes exhibit complex immune responses in gastric cancer (GC), and PD-L1 has emerged as a prognostic biomarker associated with the cancer immune microenvironment. This study aimed to determine the prognostic value of molecular subtypes and whether the addition of PD-L1 would accurately predict the prognosis and guide postoperative chemotherapy for GC patients.MethodsWe performed molecular subtyping of tissue microarray slides from 226 GC patients who were treated with radical gastrectomy. The MSI status and PD-L1 expression were evaluated through immunohistochemistry (IHC) and EBV status through situ hybridization. Multiplex polymerase chain reaction (PCR) was also performed on 50 cases to validate the accuracy of IHC in defining MSI status. Differences in overall survival (OS) were assessed using the Kaplan-Meier method, log-rank test and Cox proportional hazards regression model.ResultsAmong the 226 GC patients, 52 (23.2%) patients were classified as the MSI subtype, 11 (4.9%) were EBV+ subtype, and 161 (71.9%) were MSS (Microsatellite stable) /EBV subtype according to TCGA analysis. Two patients were both positive for MSI and EBV infection. EBV+ cases showed higher PD-L1 positivity than MSI cases and MSS/EBV cases (81.8% vs. 50.0% vs. 35.4%, P = 0.003). Compared with the non-MSS/EBV (MSI or EBV+ cases) subgroup, GC patients with MSS/EBV were associated with the worst outcomes (HR = 1.610, 95% CI [1.0462.479], P = 0.031). MSS/EBV GCs alone could benefit from postoperative chemotherapy (HR = 0.452, 95% CI [0.2990.682], P<0.001), and PD-L1-positive expression could also predict a better prognosis (HR = 0.612, 95% CI [0.3890.962], P = 0.033) in this subgroup. Considering both chemotherapy efficacy and PD-L1 expression in the MSS/EBV subgroup, chemotherapy could improve the prognosis for PD-L1-negative MSS/EBV GCs (HR = 0.357, 95% CI [0.2170.587], P <0.001) but not PD-L1-positive MSS/EBV GCs.ConclusionsMolecular subtyping combined with PD-L1 expression could serve as a potential strategy to better predict prognosis and guide postoperative chemotherapy of GC patients.

Project description:Approximately 15% of patients with colorectal cancer (CRC) exhibit a distinct molecular phenotype known as microsatellite instability (MSI). Accurate and non-invasive prediction of MSI status is crucial for cost savings and guiding clinical treatment strategies. The retrospective study enrolled 307 CRC patients between January 2020 and October 2022. Preoperative images of computed tomography and postoperative status of MSI information were available for analysis. The stratified fivefold cross-validation was used to avoid sample bias in grouping. Feature extraction and model construction were performed as follows: first, inter-/intra-correlation coefficients and the least absolute shrinkage and selection operator algorithm were used to identify the most predictive feature subset. Subsequently, multiple discriminant models were constructed to explore and optimize the combination of six feature preprocessors (Box-Cox, Yeo-Johnson, Max-Abs, Min-Max, Z-score, and Quantile) and three classifiers (logistic regression, support vector machine, and random forest). Selecting the one with the highest average value of the area under the curve (AUC) in the test set as the radiomics model, and the clinical screening model and combined model were also established using the same processing steps as the radiomics model. Finally, the performances of the three models were evaluated and analyzed using decision and correction curves.We observed that the logistic regression model based on the quantile preprocessor had the highest average AUC value in the discriminant models. Additionally, tumor location, the clinical of N stage, and hypertension were identified as independent clinical predictors of MSI status. In the test set, the clinical screening model demonstrated good predictive performance, with the average AUC of 0.762 (95% confidence interval, 0.635-0.890). Furthermore, the combined model showed excellent predictive performance (AUC, 0.958; accuracy, 0.899; sensitivity, 0.929) and favorable clinical applicability and correction effects. The logistic regression model based on the quantile preprocessor exhibited excellent performance and repeatability, which may further reduce the variability of input data and improve the model performance for predicting MSI status in CRC.

Dataset Information

Development and validation of deep learning classifiers to detect Epstein-Barr virus and microsatellite instability status in gastric cancer: a retrospective multicentre cohort study.

Background

Methods

Findings

Interpretation

Funding

Publications

Development and validation of deep learning classifiers to detect Epstein-Barr virus and microsatellite instability status in gastric cancer: a retrospective multicentre cohort study.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets