Header

UZH-Logo

Maintenance Infos

Can algorithms of artificial intelligence replace radiologists? : a study of machine learning applications to breast cancer imaging


Hejduk, Patryk. Can algorithms of artificial intelligence replace radiologists? : a study of machine learning applications to breast cancer imaging. 2023, University of Zurich, Faculty of Science.

Abstract

Table of Contents
1 Abstract...................................................................................................................................................6
2 Preface....................................................................................................................................................8
3 Contributions.........................................................................................................................................9
4 Introduction......................................................................................................................................... 11
4.1 Mammographic examination qualityassessment.................................................................... 13
4.2 Mammographic breast density................................................................................................14
4.3 Follow-up examinations........................................................................................................... 15
4.4 Breast density in Breast CT.......................................................................................................15
4.5 Deep Convolutional Neural Networks..................................................................................... 16
4.6 Aim and outline of the thesis.................................................................................................. 16
5 Fully automatic classification of automated breast ultrasound (ABUS) imaging according to BI-RADS
using a deep convolutional neural network...................................................................................................19
5.1 Abstract....................................................................................................................................20
5.2 Introduction.............................................................................................................................21
5.3 Materials and methods............................................................................................................23
5.3.1 Patient data.................................................................................................................. 23
5.3.2 ABUS examination........................................................................................................23
5.3.3 Image selection and data preparation..........................................................................23
5.3.4 dCNN architecture........................................................................................................24
5.3.5 Image analysis and lesion detection............................................................................ 25
5.3.6 Comparison with human reading................................................................................ 26
5.3.7 Statistical analysis........................................................................................................ 27
5.4 Results.....................................................................................................................................28
5.4.1 Model training..............................................................................................................28
5.4.2 Validation on single images presenting lesions............................................................28
5.4.3 Validation on full 3D ABUS datasets............................................................................. 31
5.5 Discussion.................................................................................................................................33
6 Automatic and standardized quality control of digital mammography and tomosynthesis with deep
convolutional neural networks......................................................................................................................35
6.1 Abstract
6.2 Introduction...........................................................................................................................36
6.3 Materials and methods ............................................................................................................39
6.3.1 Patient data....................................................................................................................39
6.3.2 Data preparation...........................................................................................................39
6.3.3 Image selection and data preparation.......................................................................... 39
6.3.4 Model selection.............................................................................................................41
6.3.5 dCNN architecture.........................................................................................................44
6.3.6 Image analysis...............................................................................................................44
6.3.7 Statistical analysis......................................................................................................... 45
6.4 Results.....................................................................................................................................46
6.4.1 Training, validation and test results..............................................................................46
6.4.2 Parenchyma classification............................................................................................ 46
6.4.3 Inframammary fold classification.................................................................................46
6.4.4 Pectoralis muscle classification..................................................................................... 47
6.4.5 Nipple depiction........................................................................................................... 47
6.4.6 Regression models........................................................................................................ 47
6.4.7 Quality features calculations........................................................................................ 47
6.5 Discussion................................................................................................................................48
7 Classification of Mammographic Breast Microcalcifications Using a Deep Convolutional Neural
Network: A BI-RADS-Based Approach............................................................................................................50
7.1 Abstract....................................................................................................................................51
7.2 Introduction.............................................................................................................................52
7.3 Materials and methods............................................................................................................54
7.3.1 Database Search............................................................................................................54
7.3.2 Data Preparation...........................................................................................................55
7.3.3 Training of dCNN Models.............................................................................................. 56
7.3.4 Validation Using "Real-World" Data.............................................................................57
7.3.5 Computation of Probability Maps.................................................................................57
7.4 Results......................................................................................................................................58
7.4.1 Preparation of Data and Training of dCNN Models......................................................58
7.4.2 Validation on "Real-World” Data..................................................................................60
7.4.3 Probability Maps........................................................................................................... 61
7.5 Discussion.................................................................................................................................62
8 Applied Machine Learning in Spiral Breast-CT: Can We Train a Deep Convolutional Neural Network
for Automatic, Standardized and Observer Independent Classification of Breast Density?..........................66
8.1 Introduction........................................................................................................................... 67
8.2 Materials and Methods.......................................................................................................... 69
8.2.1 Patient Selection............................................................................................................69
8.2.2 BCT Examinations......................................................................................................... 69
8.2.3 Breast Density............................................................................................................... 69
8.2.4 Data Preparation.......................................................................................................... 70
8.2.5 dCNN Architecture and Training...................................................................................71
8.2.6 Human Readout "Real-World" Subsets........................................................................72
8.2.7 Statistical Analyses....................................................................................................... 72
8.3 Results.......................................................................................................................................73
8.3.1 Patient Selection and Image Processing.......................................................................73
8.3.2 Accuracies in Training, Validation and "Real-World" Test Datasets............................. 73
8.3.3 Human Readout............................................................................................................75
8.4 Discussion 79
9 Density measurement for tomosynthesis in comparison: deep convolutional neural networks versus
human BI-RADS density scores.....................................................................................................................82
9.1 Abstract...................................................................................................................................83
9.2 Introduction............................................................................................................................84
9.3 Methods................................................................................................................................. 84
9.3.1 Study population.........................................................................................................84
9.3.2 Tomosynthesis AcquisitionParameters.........................................................................85
9.3.3 Data preparation..........................................................................................................85
9.3.4 Reference standard/ Ground Truth.............................................................................. 85
9.3.5 Model........................................................................................................................... 85
9.3.6 Statistical analysis......................................................................................................... 86
9.3.7 Breast Density Assessment............................................................................................86
9.4 Results......................................................................................................................................87
9.4.1 Study population...........................................................................................................87
9.4.2 Inter-reader Agreement................................................................................................88
9.4.3 Model performance...... ................ 88
9.5 Discussion................................ ....;...... 91
9.5.1 Study limitations............... :......... 92
9.5.2 Conclusion..................................................................................................................... 93
10 BI-RADS-based classification of mammographic soft-tissue opacities using a deep convolutional
neural network...............................................................................................................................................94
10.1 Abstract......................................................................................................................................... 95
10.2 Introduction.............................................................................................................................95
10.3 Materials and Methods.............................................................................................................97
10.3.1 Database search............................................................................................................97
10.3.2 Data preparation...........................................................................................................97
10.3.3 Training of dCNN models.............................................................................................. 98
10.3.4 Validation using "real world" data................................................................................ 99
10.3.5 Computation of probability maps................................................................................. 99
10.3.6 Statistical Analysis.......................................................................................................100
10.4 Results.................................................................................................................................... 101
10.4.1 Data preparation and training of dCNN models......................................................... 101
10.4.2 Validation on "real world" data.................................................................................. 102
10.4.3 Probability maps..........................................................................................................104
10.4.4 Discussion....................................................................................................................105
11 Conclusions.................................................................................................................................107
11.1.1 Quality assurance........................................................................................................107
11.1.2 Density assessment.....................................................................................................107
11.1.3 Breast cancer diagnostics............................................................................................108
11.1.4 Supplementary techniques..........................................................................................108
11.1.5 Summary.....................................................................................................................108
12 References.......................................................................................................

Abstract

Table of Contents
1 Abstract...................................................................................................................................................6
2 Preface....................................................................................................................................................8
3 Contributions.........................................................................................................................................9
4 Introduction......................................................................................................................................... 11
4.1 Mammographic examination qualityassessment.................................................................... 13
4.2 Mammographic breast density................................................................................................14
4.3 Follow-up examinations........................................................................................................... 15
4.4 Breast density in Breast CT.......................................................................................................15
4.5 Deep Convolutional Neural Networks..................................................................................... 16
4.6 Aim and outline of the thesis.................................................................................................. 16
5 Fully automatic classification of automated breast ultrasound (ABUS) imaging according to BI-RADS
using a deep convolutional neural network...................................................................................................19
5.1 Abstract....................................................................................................................................20
5.2 Introduction.............................................................................................................................21
5.3 Materials and methods............................................................................................................23
5.3.1 Patient data.................................................................................................................. 23
5.3.2 ABUS examination........................................................................................................23
5.3.3 Image selection and data preparation..........................................................................23
5.3.4 dCNN architecture........................................................................................................24
5.3.5 Image analysis and lesion detection............................................................................ 25
5.3.6 Comparison with human reading................................................................................ 26
5.3.7 Statistical analysis........................................................................................................ 27
5.4 Results.....................................................................................................................................28
5.4.1 Model training..............................................................................................................28
5.4.2 Validation on single images presenting lesions............................................................28
5.4.3 Validation on full 3D ABUS datasets............................................................................. 31
5.5 Discussion.................................................................................................................................33
6 Automatic and standardized quality control of digital mammography and tomosynthesis with deep
convolutional neural networks......................................................................................................................35
6.1 Abstract
6.2 Introduction...........................................................................................................................36
6.3 Materials and methods ............................................................................................................39
6.3.1 Patient data....................................................................................................................39
6.3.2 Data preparation...........................................................................................................39
6.3.3 Image selection and data preparation.......................................................................... 39
6.3.4 Model selection.............................................................................................................41
6.3.5 dCNN architecture.........................................................................................................44
6.3.6 Image analysis...............................................................................................................44
6.3.7 Statistical analysis......................................................................................................... 45
6.4 Results.....................................................................................................................................46
6.4.1 Training, validation and test results..............................................................................46
6.4.2 Parenchyma classification............................................................................................ 46
6.4.3 Inframammary fold classification.................................................................................46
6.4.4 Pectoralis muscle classification..................................................................................... 47
6.4.5 Nipple depiction........................................................................................................... 47
6.4.6 Regression models........................................................................................................ 47
6.4.7 Quality features calculations........................................................................................ 47
6.5 Discussion................................................................................................................................48
7 Classification of Mammographic Breast Microcalcifications Using a Deep Convolutional Neural
Network: A BI-RADS-Based Approach............................................................................................................50
7.1 Abstract....................................................................................................................................51
7.2 Introduction.............................................................................................................................52
7.3 Materials and methods............................................................................................................54
7.3.1 Database Search............................................................................................................54
7.3.2 Data Preparation...........................................................................................................55
7.3.3 Training of dCNN Models.............................................................................................. 56
7.3.4 Validation Using "Real-World" Data.............................................................................57
7.3.5 Computation of Probability Maps.................................................................................57
7.4 Results......................................................................................................................................58
7.4.1 Preparation of Data and Training of dCNN Models......................................................58
7.4.2 Validation on "Real-World” Data..................................................................................60
7.4.3 Probability Maps........................................................................................................... 61
7.5 Discussion.................................................................................................................................62
8 Applied Machine Learning in Spiral Breast-CT: Can We Train a Deep Convolutional Neural Network
for Automatic, Standardized and Observer Independent Classification of Breast Density?..........................66
8.1 Introduction........................................................................................................................... 67
8.2 Materials and Methods.......................................................................................................... 69
8.2.1 Patient Selection............................................................................................................69
8.2.2 BCT Examinations......................................................................................................... 69
8.2.3 Breast Density............................................................................................................... 69
8.2.4 Data Preparation.......................................................................................................... 70
8.2.5 dCNN Architecture and Training...................................................................................71
8.2.6 Human Readout "Real-World" Subsets........................................................................72
8.2.7 Statistical Analyses....................................................................................................... 72
8.3 Results.......................................................................................................................................73
8.3.1 Patient Selection and Image Processing.......................................................................73
8.3.2 Accuracies in Training, Validation and "Real-World" Test Datasets............................. 73
8.3.3 Human Readout............................................................................................................75
8.4 Discussion 79
9 Density measurement for tomosynthesis in comparison: deep convolutional neural networks versus
human BI-RADS density scores.....................................................................................................................82
9.1 Abstract...................................................................................................................................83
9.2 Introduction............................................................................................................................84
9.3 Methods................................................................................................................................. 84
9.3.1 Study population.........................................................................................................84
9.3.2 Tomosynthesis AcquisitionParameters.........................................................................85
9.3.3 Data preparation..........................................................................................................85
9.3.4 Reference standard/ Ground Truth.............................................................................. 85
9.3.5 Model........................................................................................................................... 85
9.3.6 Statistical analysis......................................................................................................... 86
9.3.7 Breast Density Assessment............................................................................................86
9.4 Results......................................................................................................................................87
9.4.1 Study population...........................................................................................................87
9.4.2 Inter-reader Agreement................................................................................................88
9.4.3 Model performance...... ................ 88
9.5 Discussion................................ ....;...... 91
9.5.1 Study limitations............... :......... 92
9.5.2 Conclusion..................................................................................................................... 93
10 BI-RADS-based classification of mammographic soft-tissue opacities using a deep convolutional
neural network...............................................................................................................................................94
10.1 Abstract......................................................................................................................................... 95
10.2 Introduction.............................................................................................................................95
10.3 Materials and Methods.............................................................................................................97
10.3.1 Database search............................................................................................................97
10.3.2 Data preparation...........................................................................................................97
10.3.3 Training of dCNN models.............................................................................................. 98
10.3.4 Validation using "real world" data................................................................................ 99
10.3.5 Computation of probability maps................................................................................. 99
10.3.6 Statistical Analysis.......................................................................................................100
10.4 Results.................................................................................................................................... 101
10.4.1 Data preparation and training of dCNN models......................................................... 101
10.4.2 Validation on "real world" data.................................................................................. 102
10.4.3 Probability maps..........................................................................................................104
10.4.4 Discussion....................................................................................................................105
11 Conclusions.................................................................................................................................107
11.1.1 Quality assurance........................................................................................................107
11.1.2 Density assessment.....................................................................................................107
11.1.3 Breast cancer diagnostics............................................................................................108
11.1.4 Supplementary techniques..........................................................................................108
11.1.5 Summary.....................................................................................................................108
12 References.......................................................................................................

Statistics

Additional indexing

Item Type:Dissertation (monographical)
Referees:Unkelbach Jan, Boss Andreas, Neupert T
Communities & Collections:07 Faculty of Science > Physics Institute
UZH Dissertations
Dewey Decimal Classification:530 Physics
Language:English
Date:2023
Deposited On:24 Aug 2023 10:05
Last Modified:24 Aug 2023 10:05
Number of Pages:120
OA Status:Closed
Full text not available from this repository.