Hypothesis‐free deep survival learning applied to the tumour microenvironment in gastric cancer
Abstract The biological complexity reflected in histology images requires advanced approaches for unbiased prognostication. Machine learning and particularly deep learning methods are increasingly applied in the field of digital pathology. In this study, we propose new ways to predict risk for cance...
| Main Authors: | , , , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Wiley
2020-10-01
|
| Series: | The Journal of Pathology: Clinical Research |
| Subjects: | |
| Online Access: | https://doi.org/10.1002/cjp2.170 |
| _version_ | 1819206925779206144 |
|---|---|
| author | Armin Meier Katharina Nekolla Lindsay C Hewitt Sophie Earle Takaki Yoshikawa Takashi Oshima Yohei Miyagi Ralf Huss Günter Schmidt Heike I Grabsch |
| author_facet | Armin Meier Katharina Nekolla Lindsay C Hewitt Sophie Earle Takaki Yoshikawa Takashi Oshima Yohei Miyagi Ralf Huss Günter Schmidt Heike I Grabsch |
| author_sort | Armin Meier |
| collection | DOAJ |
| description | Abstract The biological complexity reflected in histology images requires advanced approaches for unbiased prognostication. Machine learning and particularly deep learning methods are increasingly applied in the field of digital pathology. In this study, we propose new ways to predict risk for cancer‐specific death from digital images of immunohistochemically (IHC) stained tissue microarrays (TMAs). Specifically, we evaluated a cohort of 248 gastric cancer patients using convolutional neural networks (CNNs) in an end‐to‐end weakly supervised scheme independent of subjective pathologist input. To account for the time‐to‐event characteristic of the outcome data, we developed new survival models to guide the network training. In addition to the standard H&E staining, we investigated the prognostic value of a panel of immune cell markers (CD8, CD20, CD68) and a proliferation marker (Ki67). Our CNN‐derived risk scores provided additional prognostic value when compared to the gold standard prognostic tool TNM stage. The CNN‐derived risk scores were also shown to be superior when systematically compared to cell density measurements or a CNN score derived from binary 5‐year survival classification, which ignores time‐to‐event. To better understand the underlying biological mechanisms, we qualitatively investigated risk heat maps for each marker which visualised the network output. We identified patterns of biological interest that were related to low risk of cancer‐specific death such as the presence of B‐cell predominated clusters and Ki67 positive sub‐regions and showed that the corresponding risk scores had prognostic value in multivariate Cox regression analyses (Ki67&CD20 risks: hazard ratio (HR) = 1.47, 95% confidence interval (CI) = 1.15–1.89, p = 0.002; CD20&CD68 risks: HR = 1.33, 95% CI = 1.07–1.67, p = 0.009). Our study demonstrates the potential additional value that deep learning in combination with a panel of IHC markers can bring to the field of precision oncology. |
| first_indexed | 2024-12-23T05:15:21Z |
| format | Article |
| id | doaj.art-214b0075da184485a1149f62c1e0253f |
| institution | Directory Open Access Journal |
| issn | 2056-4538 |
| language | English |
| last_indexed | 2024-12-23T05:15:21Z |
| publishDate | 2020-10-01 |
| publisher | Wiley |
| record_format | Article |
| series | The Journal of Pathology: Clinical Research |
| spelling | doaj.art-214b0075da184485a1149f62c1e0253f2022-12-21T17:58:51ZengWileyThe Journal of Pathology: Clinical Research2056-45382020-10-016427328210.1002/cjp2.170Hypothesis‐free deep survival learning applied to the tumour microenvironment in gastric cancerArmin Meier0Katharina Nekolla1Lindsay C Hewitt2Sophie Earle3Takaki Yoshikawa4Takashi Oshima5Yohei Miyagi6Ralf Huss7Günter Schmidt8Heike I Grabsch9Image Data Sciences Definiens GmbH Munich GermanyImage Data Sciences Definiens GmbH Munich GermanyDepartment of Pathology GROW School for Oncology and Developmental Biology, Maastricht University Medical Center+ Maastricht The NetherlandsDivision of Pathology and Data Analytics, Leeds Institute of Medical Research at St. James's University of Leeds Leeds UKDepartment of Gastric Surgery National Cancer Center Hospital Tokyo JapanDepartment of Gastrointestinal Surgery Kanagawa Cancer Center Hospital Yokohama JapanMolecular Pathology and Genetics Division Kanagawa Cancer Center Research Institute Yokohama JapanInstitute of Pathology and Molecular Diagnostic University Hospital Augsburg Augsburg GermanyImage Data Sciences Definiens GmbH Munich GermanyDepartment of Pathology GROW School for Oncology and Developmental Biology, Maastricht University Medical Center+ Maastricht The NetherlandsAbstract The biological complexity reflected in histology images requires advanced approaches for unbiased prognostication. Machine learning and particularly deep learning methods are increasingly applied in the field of digital pathology. In this study, we propose new ways to predict risk for cancer‐specific death from digital images of immunohistochemically (IHC) stained tissue microarrays (TMAs). Specifically, we evaluated a cohort of 248 gastric cancer patients using convolutional neural networks (CNNs) in an end‐to‐end weakly supervised scheme independent of subjective pathologist input. To account for the time‐to‐event characteristic of the outcome data, we developed new survival models to guide the network training. In addition to the standard H&E staining, we investigated the prognostic value of a panel of immune cell markers (CD8, CD20, CD68) and a proliferation marker (Ki67). Our CNN‐derived risk scores provided additional prognostic value when compared to the gold standard prognostic tool TNM stage. The CNN‐derived risk scores were also shown to be superior when systematically compared to cell density measurements or a CNN score derived from binary 5‐year survival classification, which ignores time‐to‐event. To better understand the underlying biological mechanisms, we qualitatively investigated risk heat maps for each marker which visualised the network output. We identified patterns of biological interest that were related to low risk of cancer‐specific death such as the presence of B‐cell predominated clusters and Ki67 positive sub‐regions and showed that the corresponding risk scores had prognostic value in multivariate Cox regression analyses (Ki67&CD20 risks: hazard ratio (HR) = 1.47, 95% confidence interval (CI) = 1.15–1.89, p = 0.002; CD20&CD68 risks: HR = 1.33, 95% CI = 1.07–1.67, p = 0.009). Our study demonstrates the potential additional value that deep learning in combination with a panel of IHC markers can bring to the field of precision oncology.https://doi.org/10.1002/cjp2.170gastric cancerdeep learningsurvival analysiscomputational pathologytumour infiltrating immune cellsKi67 |
| spellingShingle | Armin Meier Katharina Nekolla Lindsay C Hewitt Sophie Earle Takaki Yoshikawa Takashi Oshima Yohei Miyagi Ralf Huss Günter Schmidt Heike I Grabsch Hypothesis‐free deep survival learning applied to the tumour microenvironment in gastric cancer The Journal of Pathology: Clinical Research gastric cancer deep learning survival analysis computational pathology tumour infiltrating immune cells Ki67 |
| title | Hypothesis‐free deep survival learning applied to the tumour microenvironment in gastric cancer |
| title_full | Hypothesis‐free deep survival learning applied to the tumour microenvironment in gastric cancer |
| title_fullStr | Hypothesis‐free deep survival learning applied to the tumour microenvironment in gastric cancer |
| title_full_unstemmed | Hypothesis‐free deep survival learning applied to the tumour microenvironment in gastric cancer |
| title_short | Hypothesis‐free deep survival learning applied to the tumour microenvironment in gastric cancer |
| title_sort | hypothesis free deep survival learning applied to the tumour microenvironment in gastric cancer |
| topic | gastric cancer deep learning survival analysis computational pathology tumour infiltrating immune cells Ki67 |
| url | https://doi.org/10.1002/cjp2.170 |
| work_keys_str_mv | AT arminmeier hypothesisfreedeepsurvivallearningappliedtothetumourmicroenvironmentingastriccancer AT katharinanekolla hypothesisfreedeepsurvivallearningappliedtothetumourmicroenvironmentingastriccancer AT lindsaychewitt hypothesisfreedeepsurvivallearningappliedtothetumourmicroenvironmentingastriccancer AT sophieearle hypothesisfreedeepsurvivallearningappliedtothetumourmicroenvironmentingastriccancer AT takakiyoshikawa hypothesisfreedeepsurvivallearningappliedtothetumourmicroenvironmentingastriccancer AT takashioshima hypothesisfreedeepsurvivallearningappliedtothetumourmicroenvironmentingastriccancer AT yoheimiyagi hypothesisfreedeepsurvivallearningappliedtothetumourmicroenvironmentingastriccancer AT ralfhuss hypothesisfreedeepsurvivallearningappliedtothetumourmicroenvironmentingastriccancer AT gunterschmidt hypothesisfreedeepsurvivallearningappliedtothetumourmicroenvironmentingastriccancer AT heikeigrabsch hypothesisfreedeepsurvivallearningappliedtothetumourmicroenvironmentingastriccancer |