TY - GEN
T1 - Wisdom of the crowd for early detection in barrett's esophagus
AU - Law, Justin
AU - Paulson, Thomas G.
AU - Sanchez, Carissa A.
AU - Galipeau, Patricia C.
AU - Jansen, Marnix
AU - Stachler, Matthew D.
AU - Maley, Carlo C.
AU - Yuan, Yinyin
N1 - Funding Information:
Y.Y. acknowledges funding from Cancer Research UK Career Establishment Award (CRUKC45982/A21808), CRUK Early Detection Program Award (C9203/A28770), CRUK Sarcoma Accelerator (C56167/A29363), CRUK Brain Tumour Award (C25858/A28592), Rosetrees Trust (A2714), Children’s Cancer and Leukaemia Group (CCLGA201906), NIH U54 CA217376,NIH R01 CA185138, CDMRP Breast Cancer Research Program Award BC132057, European Commission ITN (H2020-MSCA-ITN-2019), and The Royal Marsden/ICR National Institute of Health Research Biomedical Research Centre.
Funding Information:
T.G.P, P.C.G and C.A.S were supported by NIH P01 CA91955 and P30 CA015704.
Publisher Copyright:
© 2021 IEEE.
PY - 2021/4/13
Y1 - 2021/4/13
N2 - Cell detection is an essential task for characterizing and studying tumor microenvironments (TME). Automatic cell detection in histopathology is challenging due to the diversity of cell shape, size, morphology, as well as stain variations between laboratories. Though deep learning has become the choice method to tackle this task, typically training requires a large number of annotations, which can be laborious and time consuming. While recent developments to tackle this annotation problem have seen success, typically these pipelines add complexity to training and may not be easy to implement. In this paper we demonstrate that using several public datasets we can train a competitive cell detection network for Barrett's Esophagus (BE) premalignant tissue samples using conventional supervised training methods. To adapt the network to clinical BE tissue sections, pseudolabels were generated to retrain the network. The results indicate that public cell detection datasets can be used to train networks that generalize well to pre-cancer tissue samples without requiring any manual annotations, which can accelerate digital pathology research for early detection.
AB - Cell detection is an essential task for characterizing and studying tumor microenvironments (TME). Automatic cell detection in histopathology is challenging due to the diversity of cell shape, size, morphology, as well as stain variations between laboratories. Though deep learning has become the choice method to tackle this task, typically training requires a large number of annotations, which can be laborious and time consuming. While recent developments to tackle this annotation problem have seen success, typically these pipelines add complexity to training and may not be easy to implement. In this paper we demonstrate that using several public datasets we can train a competitive cell detection network for Barrett's Esophagus (BE) premalignant tissue samples using conventional supervised training methods. To adapt the network to clinical BE tissue sections, pseudolabels were generated to retrain the network. The results indicate that public cell detection datasets can be used to train networks that generalize well to pre-cancer tissue samples without requiring any manual annotations, which can accelerate digital pathology research for early detection.
KW - Barrett's esophagus
KW - Cell detection
KW - Convolutional neural network
KW - Deep learning
KW - Histopathology
KW - Pre-cancer
UR - http://www.scopus.com/inward/record.url?scp=85107192415&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85107192415&partnerID=8YFLogxK
U2 - 10.1109/ISBI48211.2021.9433763
DO - 10.1109/ISBI48211.2021.9433763
M3 - Conference contribution
AN - SCOPUS:85107192415
T3 - Proceedings - International Symposium on Biomedical Imaging
SP - 531
EP - 535
BT - 2021 IEEE 18th International Symposium on Biomedical Imaging, ISBI 2021
PB - IEEE Computer Society
T2 - 18th IEEE International Symposium on Biomedical Imaging, ISBI 2021
Y2 - 13 April 2021 through 16 April 2021
ER -