Title

Nuclear Segmentation Using Convolutional Neural Networks with Limited Training Data

DGIST Authors

Hwang, Jaeyoun ; Kwon, Mungi ; Park, Sanghyun

Advisor

박상현

Co-Advisor(s)

Jaeyoun Hwang

Issued Date

2020

Awarded Date

2020-02

Citation

Mungi Kwon. (2020). Nuclear Segmentation Using Convolutional Neural Networks with Limited Training Data. doi: 10.22677/Theses.200000285883

Type

Thesis

Description

Nuclear, Instance segmentation, Deep convolutional neural network, Watershed algorithm, Siamese network, Using unlabeled data, Label smoothing

Table Of Contents

Ⅰ. INTRODUCTION 1
1.1 Introduction 1
1.2 Related works 2
ⅠⅠ. Method 4
2.1 Nuclei segmentation using deep learning based watershed algorithm 4
A. Deep learning based watershed algorithm 4
B. Implementation details 5
2.2 Separation of adjacent nuclei using a Siamese neural network 5
A. Overall procedure 6
B. Proposed network 7
C. Implementation details 8
2.3 Segmentation using unlabeled data 9
A. Label smoothing 9
B. Implementation details 10
ⅠⅠⅠ. Experimental results 10
3.1 Data set 10
3.2 Evaluation metrics 11
3.3 Comparison methods 11
3.4 Segmentation accuracy 13
3.5 Classification accuracy 13
3.6 Accuracy with labeled and unlabeled data 15
VI. Conclusion 17
References 18

URI

http://dgist.dcollection.net/common/orgView/200000285883
http://hdl.handle.net/20.500.11750/11958

DOI

10.22677/Theses.200000285883

Degree

Master

Department

Robotics Engineering

Publisher

DGIST

Show Full Item Record