Development of breast tumours models database Kristina Bliznakova - - PowerPoint PPT Presentation

▶
development of
SMART_READER_LITE
LIVE PREVIEW

Development of breast tumours models database Kristina Bliznakova - - PowerPoint PPT Presentation

Nov 16, 2022 535 likes •1.68k views

H2020-TWINN-2015, Grant Agreement 692097 Development of breast tumours models database Kristina Bliznakova Katholieke University of Leuven Technical University of Varna University of Naples Federico II Belgium Bulgaria Italy Laboratory

slide-1
SLIDE 1

Technical University of Varna

Bulgaria

University of Naples – Federico II

Italy

Katholieke University of Leuven

Belgium

H2020-TWINN-2015, Grant Agreement 692097

Development of breast tumours models database

Kristina Bliznakova

slide-2
SLIDE 2

Laboratory of Computer Simulations in Medicine

slide-3
SLIDE 3

Maxima team

slide-4
SLIDE 4

location 

slide-5
SLIDE 5

The place

slide-6
SLIDE 6

Outline

  • Why realistic models of abnormalities?
slide-7
SLIDE 7

Projection images Detector

X-ray tube

Mathematical phantom

Simulation

slide-8
SLIDE 8
  • simulation
  • experiment

Planar image Tomosynthesis slice Planar image Tomosynthesis slice

Earlier validation studies

Bliznakova, Russo, et al In-line phase-contrast breast tomosynthesis: a phantom feasibility study at a synchrotron radiation facility, PMB, 61(16):6243-63

slide-9
SLIDE 9

Earlier validations

slide-10
SLIDE 10

? A Anthropomorphic breast phantoms ? ?

slide-11
SLIDE 11

Creation of a Breast Model

 Tool for creation of breast models.

BreastSimulator

 Shape  Size  Glandular tissues  Adipose tissue  Skin  Abnormalities

slide-12
SLIDE 12

Anthropomorphic phantom

Modeled breast phantom Compressed breast phantom

slide-13
SLIDE 13

Development of physical breast phantoms for phase contrast imaging

Tissue to be printed Breast models Processing of the model 3D printing STL format Additional processing

slide-14
SLIDE 14

Development of physical breast phantoms for phase contrast imaging

Tissue to be printed Breast models Processing of the model 3D printing STL format Additional processing

slide-15
SLIDE 15

Development of physical breast phantoms for phase contrast imaging

Tissue to be printed Breast models Processing of the model 3D printing STL format Additional processing

slide-16
SLIDE 16

Development of physical breast phantoms for phase contrast imaging

Tissue to be printed Breast models Processing of the model 3D printing STL format Additional processing

slide-17
SLIDE 17

Development of physical breast phantoms for phase contrast imaging

Tissue to be printed Breast models Processing of the model 3D printing STL format Additional processing

slide-18
SLIDE 18

Development of physical breast phantoms for phase contrast imaging

Tissue to be printed Breast models Processing of the model 3D printing STL format Additional processing

slide-19
SLIDE 19
slide-20
SLIDE 20

Anthropomorphic phantoms

  • Breast shape – Clear resin;
  • Glandular tree – Clear resin;
  • Animal fat;
  • Thickness – 31 mm;
  • Wall thickness – 1.7 mm.
slide-21
SLIDE 21

Physical phantoms

Courtesy Danail Ivanov et al, FOCHOS, 2016

Development of physical breast phantoms for phase contrast imaging

slide-22
SLIDE 22

? Realistic in shape,

size and absorption properties tumors

slide-23
SLIDE 23

Outline

  • Our approach and advances
slide-24
SLIDE 24

The Maxima objective

  • The main objective of the project is to increase the

research and innovation capacity of the host

  • rganisation in the field of computational modelling of

breast tumours and their use in studies of advanced x- ray breast imaging techniques such as breast tomosynthesis and phase contrast imaging.

slide-25
SLIDE 25

sample scanning 3D image Histology of the sample & Photo Database Database 3D image of the tumor (segmented) Computational tumor model Computational breast model

Simulation experiment of phase contrast

Physical model of the tumor Physical breast model

Experimental testing of phase-contrast

slide-26
SLIDE 26

sample scanning

MAXIMA

3D image Database Computational tumor model Computational breast model

slide-27
SLIDE 27

sample scanning

MAXIMA

3D image Database Physical model of the tumor Physical breast model Computational tumor model Computational breast model

slide-28
SLIDE 28

Aim

To develop the MaXIMA Breast Tumours Models’ Database, which is intended to provide researchers with computer- based breast tumours models, both:

  • Segmented from Patient data
  • Mathematically modelled realistic in shape
slide-29
SLIDE 29

Methods

  • Sources
  • Algorithms
  • Evaluation
  • Storage
  • Mathematical algorithms
  • Evaluation
  • Storage

Segmented lesions Modelled lesions

slide-30
SLIDE 30

Methods

  • Sources
  • Algorithms
  • Evaluation
  • Storage
  • Mathematical algorithms
  • Evaluation
  • Storage

Segmented lesions Modelled lesions

slide-31
SLIDE 31

Initial idea ……

Micro CT scanner

Segmented lesions

Biopsy specimen Tube containing tissue in CH20 Reconstructed images from micro CT Segmented images 3D model of microcalcification Courtesy of Prof Hilde Bosmans, KU Leuven

slide-32
SLIDE 32

Input Data For each slice

Non contrast abdominal CT scans Defining the slide where the liver appears initially Region Growing within the slice Post-processing the segmented region Verify the segmented area Click in the region of the liver Liver volume

Idea

      

slide-33
SLIDE 33

example Outlining the tumor by the algorithm

slide-34
SLIDE 34

Outlining the tumor by a doctor

slide-35
SLIDE 35

Comparison

Bliznakova et al :Computer-aided pre-operative evaluation of the residual liver volume using Computed Tomography images J Digit Imaging. 2015 Apr; 28(2): 231–239.

slide-36
SLIDE 36

Segmentation of f le lesio ions fr from patient data

Tomograms from Breast tomosynthesis/CT Filtering the images to reduce the tomographic noise Select the lesion region Region Growing within the slice Post-processing the segmented lesion Lesion volume

slide-37
SLIDE 37

Segmentation of f le lesio ions fr from patient data

Tomograms from Breast tomosynthesis/CT Filtering the images to reduce the tomographic noise Select the lesion region Region Growing within the slice Post-processing the segmented lesion Lesion volume

slide-38
SLIDE 38

Sources of data

CT patient data from radiotherapy department

  • Image size 512 х 512 pixels, 1.17mm х 1.17mm
  • Slice thickness – 5 mm
  • 188 slices of the whole body
  • Philips GEMINI TF TOF 16

Courtesy: Nikolay Dukov, Fochos 2016

slide-39
SLIDE 39

 Tomosynthesis – Alexandrovska Hospital, Sofia University Hospital, Leuven

Sources of data

CT - mastectomy University Hospital of Varna

slide-40
SLIDE 40

Segmentation of f le lesio ions fr from patient data

Tomograms from Breast tomosynthesis/CT Filtering the images to reduce the tomographic noise Select the lesion region Region Growing within the slice Post-processing the segmented lesion Lesion volume

slide-41
SLIDE 41

Segmenting the tumour area

Choosing ROI Obtaining binary image Artefact reduction Tumor segmentation

Dukov et al , RAD, 2017

slide-42
SLIDE 42

Segmentation of f le lesio ions fr from patient data

Tomograms from Breast tomosynthesis/CT Filtering the images to reduce the tomographic noise Select the lesion region Region Growing within the slice Post-processing the segmented lesion Lesion volume

slide-43
SLIDE 43

Manual post-processing

Dukov et al , RAD, 2017

slide-44
SLIDE 44

Segmentation of f le lesio ions fr from patient data

Tomograms from Breast tomosynthesis/CT Filtering the images to reduce the tomographic noise Select the lesion region Region Growing within the slice Post-processing the segmented lesion Lesion volume

slide-45
SLIDE 45

Segmented images Combined Images Unprocessed Images Example

slide-46
SLIDE 46

Segmented tumors

 More than sixty different 3D models of lesions with irregular shapes were created.  Еvaluated by experienced radiologists.

slide-47
SLIDE 47

Segmentations

Dukov et al Models of breast abnormalities based on three-dimensional medical x-ray breast images, under revision in Physica Medica

slide-48
SLIDE 48

Comparison of segmented slices obtained by using the algorithm and outlined by three radiologists and the corresponding volumes

Dukov et al Models of breast abnormalities based on three-dimensional medical x-ray breast images, under revision in Physica Medica

slide-49
SLIDE 49

A comparison of segmented (on tomographic slices) shapes

  • btained with the algorithm (in

red) and the three radiologists (green) who participated in the study

Dukov et al Models of breast abnormalities based on three-dimensional medical x-ray breast images, under revision in Physica Medica

slide-50
SLIDE 50

Future work

  • Correct the shape in z direction
  • Classification and full characterization of segmented abnormalities
  • Generation of more spiculated in shape abnormalities
  • Radiomics with the use of deep learning combined with machine

learning

  • Generation of new abnormality types from the existing models

Paolo

slide-51
SLIDE 51

Methods

  • Sources
  • Algorithms
  • Evaluation
  • Storage
  • Mathematical algorithms
  • Evaluation
  • Storage

Segmented lesions Modelled lesions

slide-52
SLIDE 52

M - tumor size Nbrownian_runs Nrun_length 3D averaging 3D dilation 3D erosion 3D random walk M

Mathematically modelled lesions - irregular masses

slide-53
SLIDE 53

Modelling is based on random walk is a random process consisting of a sequence of discrete steps of fixed length. At each step, the walk goes randomly a unit distance along one of the neighbor pixels.

M M - tumor size Nbrownian_runs Nrun_length 3D averaging 3D dilation 3D erosion 3D random walk

Mathematically modelled lesions - irregular masses

slide-54
SLIDE 54

M M - tumor size Nbrownian_runs Nrun_length 3D averaging 3D dilation 3D erosion 3D random walk

Modelling is based on random walk is a random process consisting of a sequence of discrete steps of fixed length. At each step, the walk goes randomly a unit distance along one of the neighbor pixels.

Mathematically modelled lesions - irregular masses

slide-55
SLIDE 55

M M - tumor size Nbrownian_runs Nrun_length 3D averaging 3D dilation 3D erosion 3D random walk

Modelling is based on random walk is a random process consisting of a sequence of discrete steps of fixed length. At each step, the walk goes randomly a unit distance along one of the neighbor pixels.

Mathematically modelled lesions - irregular masses

slide-56
SLIDE 56

M M - tumor size Nbrownian_runs Nrun_length 3D averaging 3D dilation 3D erosion 3D random walk

Modelling is based on random walk is a random process consisting of a sequence of discrete steps of fixed length. At each step, the walk goes randomly a unit distance along one of the neighbor pixels.

Mathematically modelled lesions - irregular masses

slide-57
SLIDE 57

M - tumor size Nbrownian_runs Nrun_length 3D averaging 3D dilation 3D erosion 3D random walk

Modelling is based on random walk is a random process consisting of a sequence of discrete steps of fixed length. At each step, the walk goes randomly a unit distance along one of the neighbor pixels.

M

Mathematically modelled lesions - irregular masses

slide-58
SLIDE 58

M - tumor size Nbrownian_runs Nrun_length 3D averaging 3D dilation 3D erosion 3D random walk

Modelling is based on random walk is a random process consisting of a sequence of discrete steps of fixed length. At each step, the walk goes randomly a unit distance along one of the neighbor pixels.

M

Mathematically modelled lesions - irregular masses

slide-59
SLIDE 59

M - tumor size Nbrownian_runs Nrun_length 3D averaging 3D dilation 3D erosion 3D random walk

Modelling is based on random walk is a random process consisting of a sequence of discrete steps of fixed length. At each step, the walk goes randomly a unit distance along one of the neighbor pixels.

M

Mathematically modelled lesions - irregular masses

slide-60
SLIDE 60

M - tumor size Nbrownian_runs Nrun_length 3D averaging 3D dilation 3D erosion 3D random walk

Modelling is based on random walk is a random process consisting of a sequence of discrete steps of fixed length. At each step, the walk goes randomly a unit distance along one of the neighbor pixels.

M

Mathematically modelled lesions - irregular masses

slide-61
SLIDE 61

M - tumor size Nbrownian_runs Nrun_length 3D averaging 3D dilation 3D erosion 3D random walk

Modelling is based on random walk is a random process consisting of a sequence of discrete steps of fixed length. At each step, the walk goes randomly a unit distance along one of the neighbor pixels.

M

Mathematically modelled lesions - irregular masses

slide-62
SLIDE 62

M - tumor size Nbrownian_runs Nrun_length 3D averaging 3D dilation 3D erosion 3D random walk

Modelling is based on random walk is a random process consisting of a sequence of discrete steps of fixed length. At each step, the walk goes randomly a unit distance along one of the neighbor pixels.

M

Mathematically modelled lesions - irregular masses

slide-63
SLIDE 63

M - tumor size Nbrownian_runs Nrun_length 3D averaging 3D dilation 3D erosion 3D random walk

Mathematically modelled lesions - irregular masses

slide-64
SLIDE 64

Models

M = 200 Nbrownian_runs = 300 Nrun_length =200

3D erosion 3D dilation 3D averaging

slide-65
SLIDE 65

Models

slide-66
SLIDE 66

Evaluation

slide-67
SLIDE 67

NN_1_200_2000_2000_4

NN_1_200_2000_3000_4

nn_200_2000_2000_4_256x256 0o nn_200_2000_2000_4_256x256 90o nn_200_2000_3000_4_256x256 0o nn_200_2000_3000_4_256x256 90o

slide-68
SLIDE 68

NN_1_200_2000_4000_4 NN_1_200_2000_5000_4

nn_200_2000_4000_4_256x256 0o nn_200_2000_5000_4_256x256 90o nn_200_2000_5000_4_256x256 0o nn_200_2000_4000_4_256x256 9 0o

slide-69
SLIDE 69

NN_1_500_500_1000_4 NN_1_500_500_2000_4 nn_500_500_1000_4_256x256 0o nn_500_500_1000_4_256x256 90o nn_500_500_2000_4_256x256 90o nn_500_500_2000_4_256x256 0o

slide-70
SLIDE 70

NN_1_500_500_3000_4 NN_1_500_500_4000_4 nn_500_500_3000_4_256x256 90o nn_500_500_3000_4_256x256 0o nn_500_500_4000_4_256x256 90o nn_500_500_4000_4_256x256 0o

slide-71
SLIDE 71

NN_1_500_500_5000_4 NN_1_500_1000_1000_4 NN_1_500_1000_1000_4_300x 0o NN_1_500_1000_1000_4_300x 90o nn_500_500_5000_4_256x256 0o nn_500_500_5000_4_256x256 90o

slide-72
SLIDE 72

NN_1_500_1000_2000_4 NN_1_500_1000_3000_4 nn_500_1000_2000_4_300x 300 0o nn_500_1000_2000_4_300x 300 90o nn_500_1000_3000_4_300x 300 0o nn_500_1000_3000_4_300x 300 90o

slide-73
SLIDE 73

NN_1_500_1000_5000_4 nn_500_1000_4000_4_300x 300 0o nn_500_1000_4000_4_300x 300 90o nn_500_1000_5000_4_300x 300 90o nn_500_1000_5000_4_300x 300 0o

slide-74
SLIDE 74

NN_1_500_2000_1000_4 NN_1_500_2000_2000_4 nn_500_2000_1000_4_ 450x450 0o nn_500_2000_1000_4_ 450x450 90o nn_500_2000_2000_4_ 450x450 0o nn_500_2000_2000_4_ 450x450 90o

slide-75
SLIDE 75

nn_500_2000_3000_4_ 450x450 0o nn_500_2000_3000_4_ 450x450 9 0o nn_500_2000_4000_4_ 450x450 0o nn_500_2000_4000_4_ 450x450 90o

slide-76
SLIDE 76

nn_500_2000_5000_4_ 450x450 0o

slide-77
SLIDE 77

NN_1_500_3000_1000_1 NN_1_500_3000_1000_4 nn_500_3000_1000_4_ 550x550 0o nn_500_3000_1000_4_ 550x550 90o nn_500_3000_1000_1_ 550x550 0o nn_500_3000_1000_1_ 550x550 90o

slide-78
SLIDE 78

NN1000_1000_1000_01_400x400 NN1000_1000_1000_01_300x300 nn1000_1000_1000_4_256x256 NN_1000_1000_1000 NN1000_1000_2000

Nn_1000_1000_2000_4_256x256 0o

Ray step 0.05 mm

Nn_1000_1000_2000_4_256x256 90o

slide-79
SLIDE 79

NN1000_1000_3000 NN1000_1000_4000 nn_1000_1000_3000_4_400x400 0o nn_1000_1000_5000_4_400x400 90o nn_1000_1000_4000_4_400x400 0o nn_1000_1000_4000_4_400x400 90o

slide-80
SLIDE 80

The created Images are stored in a database and will be used in the further assessment, research and educational activities.

MLO view CC view

Gospodinova, 2018, ACT2018, Ohrid

Subjective evaluation 4 AFC

Under development

Objective evaluation

slide-81
SLIDE 81

Database organisation

slide-82
SLIDE 82

Database web-based interface

slide-83
SLIDE 83

Database organisation

slide-84
SLIDE 84

Database organisations

slide-85
SLIDE 85

Database organisation

slide-86
SLIDE 86

Irregular masses - description

slide-87
SLIDE 87

Outline

  • Exploitation of results in practice
slide-88
SLIDE 88

Exploitation of results

  • Research
  • Education
  • Fun
slide-89
SLIDE 89

Exploitation of results

  • Research
slide-90
SLIDE 90

To study novel breast imaging techniques

Research

slide-91
SLIDE 91

Simulation of the compression procedure

Creation of breast model

Dukov et al, IUPESM, 2018

slide-92
SLIDE 92

Compressed breast model

3D view

slice

Dukov et al, IUPESM, 2018

slide-93
SLIDE 93

Example

X-ray tube detector software breast phantom source-detector distance source-object distance

  • 250

+ 250

Software X-ray Imaging Simulators

slide-94
SLIDE 94

Images obtained with the software phantom

(a) (b) (c) Dukov et al, IUPESM, 2018

slide-95
SLIDE 95

Phantoms for the study

  • Breast shape – Clear resin;
  • Glandular tree – Clear resin;
  • Animal fat;
  • Thickness – 31 mm;
  • Wall thickness – 1.7 mm.
  • Container – Clear resin;
  • Glandular tree – Clear resin;
  • Water;
  • Thickness – 49 mm;
  • Wall thickness – 2.4 mm.
slide-96
SLIDE 96

Phantoms for the study

  • 27 spheres from Gray resin;
  • radiuses [6 - 13] mm;
  • white resin container;
  • ~ 33000 PMMA spheres;
  • radiuses [0.79 - 7.94] mm;
  • animal fat;
  • wall thickness 3 mm.
  • PMMA container;
  • water.
slide-97
SLIDE 97

Setup

60keV Planar images Tomosynthesis - 25 projection images

slide-98
SLIDE 98

Exploitation of results

  • Education
slide-99
SLIDE 99

Eutempe-net training course, Varna 22-26/05/2017

  • How to model x-ray

imaging chain;

  • How to model breast;
  • How to model breast

cancer;

  • How to work in a

team;

  • How to write

abstracts;

  • How to present a

scientific work.

slide-100
SLIDE 100
slide-101
SLIDE 101

Dimension of the phantom: Width : 641 pixels Length : 357 pixels Height : 175 pixels Voxel size: 0.27 mm [each side]

Phantom

Kristina Tri Wigati, Hannah Manssens, Anthropomorphic phantoms, Eutempe-net training course, Varna 22-26/05/2017

slide-102
SLIDE 102

Breast abnormalities

  • The voxelized of irregular mass matrix was placed about the

center of the normal breast phantom;

  • Resolution: 0.27 mm/pixel

Irregular mass Kristina Tri Wigati, Hannah Manssens, Anthropomorphic phantoms, Eutempe-net training course, Varna 22-26/05/2017

slide-103
SLIDE 103

Virtual clinical study

Kristina Tri Wigati, Hannah Manssens, JMPB, 5(1), p. 124-138, 2018

slide-104
SLIDE 104

Addition of noise

Kristina Tri Wigati, Hannah Manssens, Anthropomorphic phantoms, Eutempe-net training course, Varna 22-26/05/2017

slide-105
SLIDE 105

Qualitative comparison

Mammography Tomosynthesis with 26 projection images Kristina Tri Wigati, Hannah Manssens, Anthropomorphic phantoms, Eutempe-net training course, Varna 22-26/05/2017

slide-106
SLIDE 106

Exploitation of results

  • Fun
slide-107
SLIDE 107

Exploitation

Dukov et al, FOCHOS 2017

slide-108
SLIDE 108

CT Scanning

Dukov et al, FOCHOS 2017

slide-109
SLIDE 109

Segmenting a health part

binarization segmenting the health part mirror image morphological corrections Dukov et al, FOCHOS 2017

slide-110
SLIDE 110

3D printing

Dukov et al, FOCHOS 2017

slide-111
SLIDE 111

Exploitation

Dukov et al, FOCHOS 2017

slide-112
SLIDE 112

Conclusions

  • The developed database will serve as an imaging data

source for researchers, working on breast imaging and early breast cancer detection with the help of existing

  • r newly developed imaging modalities.
slide-113
SLIDE 113

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 692097.

Acknowledgments