Leveraging weak complementary labels enhances semantic segmentation of hepatocellular carcinoma and intrahepatic cholangiocarcinoma
Abstract In this paper we present a deep learning segmentation approach to classify and quantify the two most prevalent primary liver cancers – hepatocellular carcinoma and intrahepatic cholangiocarcinoma – from hematoxylin and eosin (H&E) stained whole slide images. While semantic segmentation...
Saved in:
| Main Authors: | , , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Nature Portfolio
2024-10-01
|
| Series: | Scientific Reports |
| Online Access: | https://doi.org/10.1038/s41598-024-75256-w |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1850203760981180416 |
|---|---|
| author | Miriam Hägele Johannes Eschrich Lukas Ruff Maximilian Alber Simon Schallenberg Adrien Guillot Christoph Roderburg Frank Tacke Frederick Klauschen |
| author_facet | Miriam Hägele Johannes Eschrich Lukas Ruff Maximilian Alber Simon Schallenberg Adrien Guillot Christoph Roderburg Frank Tacke Frederick Klauschen |
| author_sort | Miriam Hägele |
| collection | DOAJ |
| description | Abstract In this paper we present a deep learning segmentation approach to classify and quantify the two most prevalent primary liver cancers – hepatocellular carcinoma and intrahepatic cholangiocarcinoma – from hematoxylin and eosin (H&E) stained whole slide images. While semantic segmentation of medical images typically requires costly pixel-level annotations by domain experts, there often exists additional information which is routinely obtained in clinical diagnostics but rarely utilized for model training. We propose to leverage such weak information from patient diagnoses by deriving complementary labels that indicate to which class a sample cannot belong to. To integrate these labels, we formulate a complementary loss for segmentation. Motivated by the medical application, we demonstrate for general segmentation tasks that including additional patches with solely weak complementary labels during model training can significantly improve the predictive performance and robustness of a model. On the task of diagnostic differentiation between hepatocellular carcinoma and intrahepatic cholangiocarcinoma, we achieve a balanced accuracy of 0.91 (CI 95%: 0.86-0.95) at case level for 165 hold-out patients. Furthermore, we also show that leveraging complementary labels improves the robustness of segmentation and increases performance at case level. |
| format | Article |
| id | doaj-art-d08e9148e45f44629cc40f3db86efa21 |
| institution | OA Journals |
| issn | 2045-2322 |
| language | English |
| publishDate | 2024-10-01 |
| publisher | Nature Portfolio |
| record_format | Article |
| series | Scientific Reports |
| spelling | doaj-art-d08e9148e45f44629cc40f3db86efa212025-08-20T02:11:25ZengNature PortfolioScientific Reports2045-23222024-10-0114111310.1038/s41598-024-75256-wLeveraging weak complementary labels enhances semantic segmentation of hepatocellular carcinoma and intrahepatic cholangiocarcinomaMiriam Hägele0Johannes Eschrich1Lukas Ruff2Maximilian Alber3Simon Schallenberg4Adrien Guillot5Christoph Roderburg6Frank Tacke7Frederick Klauschen8Machine learning group, Technische Universität BerlinDepartment of Hepatology and Gastroenterology, Charité Universitätsmedizin BerlinAignostics GmbHAignostics GmbHInstitute of Pathology, Charité Universitätsmedizin BerlinDepartment of Hepatology and Gastroenterology, Charité Universitätsmedizin BerlinClinic for Gastroenterology, Hepatology and Infectious Diseases, University Hospital Düsseldorf, Medical Faculty of Heinrich Heine University DüsseldorfDepartment of Hepatology and Gastroenterology, Charité Universitätsmedizin BerlinBIFOLD - Berlin Institute for the Foundations of Learning and DataAbstract In this paper we present a deep learning segmentation approach to classify and quantify the two most prevalent primary liver cancers – hepatocellular carcinoma and intrahepatic cholangiocarcinoma – from hematoxylin and eosin (H&E) stained whole slide images. While semantic segmentation of medical images typically requires costly pixel-level annotations by domain experts, there often exists additional information which is routinely obtained in clinical diagnostics but rarely utilized for model training. We propose to leverage such weak information from patient diagnoses by deriving complementary labels that indicate to which class a sample cannot belong to. To integrate these labels, we formulate a complementary loss for segmentation. Motivated by the medical application, we demonstrate for general segmentation tasks that including additional patches with solely weak complementary labels during model training can significantly improve the predictive performance and robustness of a model. On the task of diagnostic differentiation between hepatocellular carcinoma and intrahepatic cholangiocarcinoma, we achieve a balanced accuracy of 0.91 (CI 95%: 0.86-0.95) at case level for 165 hold-out patients. Furthermore, we also show that leveraging complementary labels improves the robustness of segmentation and increases performance at case level.https://doi.org/10.1038/s41598-024-75256-w |
| spellingShingle | Miriam Hägele Johannes Eschrich Lukas Ruff Maximilian Alber Simon Schallenberg Adrien Guillot Christoph Roderburg Frank Tacke Frederick Klauschen Leveraging weak complementary labels enhances semantic segmentation of hepatocellular carcinoma and intrahepatic cholangiocarcinoma Scientific Reports |
| title | Leveraging weak complementary labels enhances semantic segmentation of hepatocellular carcinoma and intrahepatic cholangiocarcinoma |
| title_full | Leveraging weak complementary labels enhances semantic segmentation of hepatocellular carcinoma and intrahepatic cholangiocarcinoma |
| title_fullStr | Leveraging weak complementary labels enhances semantic segmentation of hepatocellular carcinoma and intrahepatic cholangiocarcinoma |
| title_full_unstemmed | Leveraging weak complementary labels enhances semantic segmentation of hepatocellular carcinoma and intrahepatic cholangiocarcinoma |
| title_short | Leveraging weak complementary labels enhances semantic segmentation of hepatocellular carcinoma and intrahepatic cholangiocarcinoma |
| title_sort | leveraging weak complementary labels enhances semantic segmentation of hepatocellular carcinoma and intrahepatic cholangiocarcinoma |
| url | https://doi.org/10.1038/s41598-024-75256-w |
| work_keys_str_mv | AT miriamhagele leveragingweakcomplementarylabelsenhancessemanticsegmentationofhepatocellularcarcinomaandintrahepaticcholangiocarcinoma AT johanneseschrich leveragingweakcomplementarylabelsenhancessemanticsegmentationofhepatocellularcarcinomaandintrahepaticcholangiocarcinoma AT lukasruff leveragingweakcomplementarylabelsenhancessemanticsegmentationofhepatocellularcarcinomaandintrahepaticcholangiocarcinoma AT maximilianalber leveragingweakcomplementarylabelsenhancessemanticsegmentationofhepatocellularcarcinomaandintrahepaticcholangiocarcinoma AT simonschallenberg leveragingweakcomplementarylabelsenhancessemanticsegmentationofhepatocellularcarcinomaandintrahepaticcholangiocarcinoma AT adrienguillot leveragingweakcomplementarylabelsenhancessemanticsegmentationofhepatocellularcarcinomaandintrahepaticcholangiocarcinoma AT christophroderburg leveragingweakcomplementarylabelsenhancessemanticsegmentationofhepatocellularcarcinomaandintrahepaticcholangiocarcinoma AT franktacke leveragingweakcomplementarylabelsenhancessemanticsegmentationofhepatocellularcarcinomaandintrahepaticcholangiocarcinoma AT frederickklauschen leveragingweakcomplementarylabelsenhancessemanticsegmentationofhepatocellularcarcinomaandintrahepaticcholangiocarcinoma |