SlideShare ist ein Scribd-Unternehmen logo
1 von 37
Learning-based vertebra segmentation,
identification and partitioning
DDH Open, 3 April 2019
Nikolas Lessmann
Extrapulmonary diseases in lung cancer screening
Cardiovascular disease
Osteoporosis
Lung cancer
Osteoporosis
Potential new biomarkers
• Height loss compared to
previous screening round?
• Density loss over time?
• …
CT-biomarkers for osteoporosis
Bone mineral density
(usually done with dual energy
X-ray absorptiometry or QCT)
Relative height loss
(compared with
neighboring vertebrae)
• Visible in many different types of scans (different FOV, context)
• Difficult to distinguish (anatomical label)
• Difficult to separate from each other in low-resolution scans
• How reliable are density measures in low-SNR scans?
Challenges for automatic vertebra analysis in CT
Part 1
Vertebra segmentation
Vertebrae often need to be separated not only
from the background, but also from each other
Generic algorithm should not make
assumptions about:
• How many vertebrae are visible
• Which vertebrae are visible
• Their location within the image
= instance segmentation
An instance segmentation problem
Dedicated loss functions, e.g., arXiv:1708.02551
Detection & segmentation pipelines
(e.g., Mask RCNN, based on region proposals)
Recurrent attention
(instance by instance segm.)
Instance segmentation with CNNs
arXiv:1703.06870
arXiv:1605.09410
• Only one class of objects (vertebra)
• Size and shape are relatively consistent
• Instances are located in a column, one next to the other
A restricted instance segmentation problem
Combined detection and segmentation
• Segment one vertebra
• Look for the next in the neighborhood
The network needs a memory, like
recurrent networks, to remember which
vertebrae have already been found
Iterative instance segmentation
ROI (field of view of CNN) moves
through the image volume
Patch-based iterative instance segmentation
Image Memory Segmentation
1. Iteration
Patch-based iterative instance segmentation
Image Memory Segmentation
2. Iteration
Patch-based iterative instance segmentation
Image Memory Segmentation
3. Iteration
• Use prior knowledge that vertebrae are located next to each other
• Search for spine, then follow the spine
Traversal along the spine
I = Image patch
M = Memory patch
S = Segmentation mask
L = Anatomical label
Prediction of label 1-24, global
probabilistic model to find most
plausible labeling
C = Completeness score
Vertebrae may be visible only
partially in the image, relevant for
following automatic steps
Network architecture
Label regression with probabilistic interpretation
Prediction: 14.8
Interpretation:
• 80% probability label 15
• 20% probability label 14
Maximum-likelihood approach
Anatomical labeling
Mean certainty per labeling sequence
0.24 0.56 0.89
• Some vertebrae are often only partially
contained in the scan
• For some applications, that is important to
know – biomarkers computed based on
incompletely visible vertebrae can be
unreliable
Completeness score
Sum of three error terms
• Segmentation: Minimize FP and FN voxels
• Labeling: L1 norm
• Completeness: Binary crossentropy
Loss function
Weight map to improve
segmentation accuracy at
the boundary between
vertebrae
• Adam optimizer with fixed learning rate + high momentum (0.999)
• Single patch per iteration (no mini-batches)
 each patch has ~2,000,000 voxels
• Also patches without vertebral bone are presented to the network
• Memory state derived from reference segmentation
• Data augmentation: elastic deformations, random Gaussian noise and smoothing, random
cropping
Training details
• Manual and semi-automatic annotation of NLST (low-dose chest CT) scans
• Several datasets from vertebra segmentation challenges
Evaluation
Lumbar spine CT Chest CT Lumbar spine
MR (VB only)
Evaluation
Part 2
Vertebra partitioning
• We can do vertebra segmentation in CT
• We can do vertebral body segmentation in MR
• We cannot do vertebral body segmentation in CT… (New training set? Transfer learning? GANs?)
Motivation
Alternative idea: Let a CNN fit a continuous surface that represents the boundary
Surface fitting
• Splines are defined by a set of control points
• Splines are smooth between control points
Think of a stick or a thin metal sheet
that is fixated at several points and
resists bending due to its rigidity
 CNN could predict location of control points
Fitting a thin-plate spline with a CNN
How to backpropagate through this?
Flowchart
Differentiable TPS surface?
Surface “height” across grid
Configuration of grid points w.r.t. control points
Configuration of control points w.r.t. control points
Control point “heights” (network predicition)
If we had vertebra and vertebral body segmentations of the same subjects (and maybe even the
same scans), we could train a simpler approach …
We have segmentation masks from
• Different subjects
• Different scans
• Different modalities even
How to train without matching segmentation pairs?
Segmentation masks do not match, but we can
learn something from this data – statistical
approach, capture plausible shapes
• Do not use image intensities at all
• Train convolutional autoencoder on vertebral
body segmentation masks and use as
out-of-distribution classifier
Shape encoding with autoencoders
Flowchart
Trained with vertebra segmentation masks from CT
scans (around 600 NLST scans, > 6000 vertebrae)
Previously trained with vertebral body segmentation
masks from MR scans (around 300 UMCU scans,
>1000 vertebral bodies)
Reconstruction error + several regularizing terms
• Volume penalty: Volume Vertebra – Volume Vertebral Body
• “Fights” the reconstruction loss by pushing the surface to the right (posterior)
• Empty segmentation masks are perfectly reconstructed by the AE…
• Curvature penalty: abs(det(Hessian))
• Prevents that surface is very wobbly (seems not really necessary)
Loss function
Loss function
Manual annotation of vertebrae and vertebral
bodies in 5 chest CT scans
Results
Measurements of the height of the vertebral
bodies and correlation with extracted volume
Average of 3 measurements in 389 scans
Results
We can…
• Find and segment the vertebrae
• Label them anatomically
• Determine if they are completely or partially visible
• Partition them into body and posterior elements
Next step: Application (fracture prediction, fracture classification, …)
Also possible: use partitioning results to train direct vertebral body segmentation
Conclusions and future work
Thank you!

Weitere ähnliche Inhalte

Was ist angesagt?

Macular Thickness Diagnosis in 3D Environment
Macular Thickness Diagnosis in 3D EnvironmentMacular Thickness Diagnosis in 3D Environment
Macular Thickness Diagnosis in 3D Environment
Kasun Abeysinghe
 
11.segmentation and feature extraction of tumors from digital mammograms
11.segmentation and feature extraction of tumors from digital mammograms11.segmentation and feature extraction of tumors from digital mammograms
11.segmentation and feature extraction of tumors from digital mammograms
Alexander Decker
 
Segmentation techniques for extraction and description of tumour region from ...
Segmentation techniques for extraction and description of tumour region from ...Segmentation techniques for extraction and description of tumour region from ...
Segmentation techniques for extraction and description of tumour region from ...
Swarada Kanap
 

Was ist angesagt? (20)

MNPS, System Presentation - 2018 v.2
MNPS, System Presentation - 2018  v.2MNPS, System Presentation - 2018  v.2
MNPS, System Presentation - 2018 v.2
 
IMAGE GUIDED SURGERY
IMAGE GUIDED SURGERYIMAGE GUIDED SURGERY
IMAGE GUIDED SURGERY
 
Brain Tumor Detection Using Image Processing
Brain Tumor Detection Using Image ProcessingBrain Tumor Detection Using Image Processing
Brain Tumor Detection Using Image Processing
 
A Re-Learning Based Post-Processing Step For Brain Tumor Segmentation From Mu...
A Re-Learning Based Post-Processing Step For Brain Tumor Segmentation From Mu...A Re-Learning Based Post-Processing Step For Brain Tumor Segmentation From Mu...
A Re-Learning Based Post-Processing Step For Brain Tumor Segmentation From Mu...
 
Brain tissue segmentation from MR images
Brain tissue segmentation from MR images Brain tissue segmentation from MR images
Brain tissue segmentation from MR images
 
Neural Network Based Brain Tumor Detection using MR Images
Neural Network Based Brain Tumor Detection using MR ImagesNeural Network Based Brain Tumor Detection using MR Images
Neural Network Based Brain Tumor Detection using MR Images
 
Macular Thickness Diagnosis in 3D Environment
Macular Thickness Diagnosis in 3D EnvironmentMacular Thickness Diagnosis in 3D Environment
Macular Thickness Diagnosis in 3D Environment
 
11.segmentation and feature extraction of tumors from digital mammograms
11.segmentation and feature extraction of tumors from digital mammograms11.segmentation and feature extraction of tumors from digital mammograms
11.segmentation and feature extraction of tumors from digital mammograms
 
CT computer aided diagnosis system
CT computer aided diagnosis systemCT computer aided diagnosis system
CT computer aided diagnosis system
 
Masters' whole work(big back-u_pslide)
Masters' whole work(big back-u_pslide)Masters' whole work(big back-u_pslide)
Masters' whole work(big back-u_pslide)
 
Tumour detection
Tumour detectionTumour detection
Tumour detection
 
IMAGE SEGMENTATION USING FCM ALGORITM | J4RV3I12021
IMAGE SEGMENTATION USING FCM ALGORITM | J4RV3I12021IMAGE SEGMENTATION USING FCM ALGORITM | J4RV3I12021
IMAGE SEGMENTATION USING FCM ALGORITM | J4RV3I12021
 
Segmentation techniques for extraction and description of tumour region from ...
Segmentation techniques for extraction and description of tumour region from ...Segmentation techniques for extraction and description of tumour region from ...
Segmentation techniques for extraction and description of tumour region from ...
 
LLTech microbiopsies may 2011
LLTech microbiopsies may 2011LLTech microbiopsies may 2011
LLTech microbiopsies may 2011
 
Image registration and data fusion techniques.pptx latest save
Image registration and data fusion techniques.pptx latest saveImage registration and data fusion techniques.pptx latest save
Image registration and data fusion techniques.pptx latest save
 
Lung Nodule detection System
Lung Nodule detection SystemLung Nodule detection System
Lung Nodule detection System
 
Neuro imaging and electro diagnostics
Neuro imaging and electro diagnosticsNeuro imaging and electro diagnostics
Neuro imaging and electro diagnostics
 
Modeling Cardiac Pacemakers With Timed Coloured Petri Nets And Related Tools
Modeling Cardiac Pacemakers With Timed Coloured Petri Nets And Related ToolsModeling Cardiac Pacemakers With Timed Coloured Petri Nets And Related Tools
Modeling Cardiac Pacemakers With Timed Coloured Petri Nets And Related Tools
 
IRJET - Classification of Cancer Images using Deep Learning
IRJET -  	  Classification of Cancer Images using Deep LearningIRJET -  	  Classification of Cancer Images using Deep Learning
IRJET - Classification of Cancer Images using Deep Learning
 
MNPS. Mevis Neurosurgery Planning System. Overview, Aug. 2018
MNPS.  Mevis Neurosurgery Planning System. Overview, Aug. 2018MNPS.  Mevis Neurosurgery Planning System. Overview, Aug. 2018
MNPS. Mevis Neurosurgery Planning System. Overview, Aug. 2018
 

Ähnlich wie Learning-based vertebra segmentation, identification and partioning

Image Quality And Artifacts in Computed Tomography.pptx
Image Quality And Artifacts in Computed Tomography.pptxImage Quality And Artifacts in Computed Tomography.pptx
Image Quality And Artifacts in Computed Tomography.pptx
ivanKeshari
 
120 mr imaging for the evaluation of carotid atherosclerosis
120 mr imaging for the evaluation of carotid atherosclerosis120 mr imaging for the evaluation of carotid atherosclerosis
120 mr imaging for the evaluation of carotid atherosclerosis
SHAPE Society
 
CBCT in dental practice
CBCT in dental practiceCBCT in dental practice
CBCT in dental practice
Zana Hussein
 

Ähnlich wie Learning-based vertebra segmentation, identification and partioning (20)

Physics of Multidetector CT Scan
Physics of Multidetector CT ScanPhysics of Multidetector CT Scan
Physics of Multidetector CT Scan
 
Comparison of ct and cbct
Comparison of ct and cbct Comparison of ct and cbct
Comparison of ct and cbct
 
EMBC conference poster
EMBC conference posterEMBC conference poster
EMBC conference poster
 
Design of Modified Bio-Inspired Algorithm for Identification and Segmentation...
Design of Modified Bio-Inspired Algorithm for Identification and Segmentation...Design of Modified Bio-Inspired Algorithm for Identification and Segmentation...
Design of Modified Bio-Inspired Algorithm for Identification and Segmentation...
 
Ct image quality artifacts and it remedy
Ct image quality artifacts and it remedyCt image quality artifacts and it remedy
Ct image quality artifacts and it remedy
 
Image Quality And Artifacts in Computed Tomography.pptx
Image Quality And Artifacts in Computed Tomography.pptxImage Quality And Artifacts in Computed Tomography.pptx
Image Quality And Artifacts in Computed Tomography.pptx
 
CBCT IN ORTHODONTICS
CBCT IN ORTHODONTICSCBCT IN ORTHODONTICS
CBCT IN ORTHODONTICS
 
IMAGE GUIDED SURGERY IN NOSE AND PNS.pptx
IMAGE GUIDED SURGERY IN NOSE AND PNS.pptxIMAGE GUIDED SURGERY IN NOSE AND PNS.pptx
IMAGE GUIDED SURGERY IN NOSE AND PNS.pptx
 
Olga Senyukova - Machine Learning Applications in Medicine
Olga Senyukova - Machine Learning Applications in MedicineOlga Senyukova - Machine Learning Applications in Medicine
Olga Senyukova - Machine Learning Applications in Medicine
 
Segmentation and Labelling of Human Spine MR Images Using Fuzzy Clustering
Segmentation and Labelling of Human Spine MR Images Using Fuzzy Clustering  Segmentation and Labelling of Human Spine MR Images Using Fuzzy Clustering
Segmentation and Labelling of Human Spine MR Images Using Fuzzy Clustering
 
SEGMENTATION AND LABELLING OF HUMAN SPINE MR IMAGES USING FUZZY CLUSTERING
SEGMENTATION AND LABELLING OF HUMAN SPINE MR IMAGES USING FUZZY CLUSTERINGSEGMENTATION AND LABELLING OF HUMAN SPINE MR IMAGES USING FUZZY CLUSTERING
SEGMENTATION AND LABELLING OF HUMAN SPINE MR IMAGES USING FUZZY CLUSTERING
 
Saloner presentation
Saloner presentationSaloner presentation
Saloner presentation
 
120 mr imaging for the evaluation of carotid atherosclerosis
120 mr imaging for the evaluation of carotid atherosclerosis120 mr imaging for the evaluation of carotid atherosclerosis
120 mr imaging for the evaluation of carotid atherosclerosis
 
120 mr imaging for the evaluation of carotid atherosclerosis
120 mr imaging for the evaluation of carotid atherosclerosis120 mr imaging for the evaluation of carotid atherosclerosis
120 mr imaging for the evaluation of carotid atherosclerosis
 
Mr based attenuation correction in pet
Mr based attenuation correction in petMr based attenuation correction in pet
Mr based attenuation correction in pet
 
Quantitative Cancer Image Analysis
Quantitative Cancer Image AnalysisQuantitative Cancer Image Analysis
Quantitative Cancer Image Analysis
 
ShortStoryPPT.pptx
ShortStoryPPT.pptxShortStoryPPT.pptx
ShortStoryPPT.pptx
 
CAT
CATCAT
CAT
 
Tissue Engineering introduction for physicists - Lecture three
Tissue Engineering introduction for physicists - Lecture threeTissue Engineering introduction for physicists - Lecture three
Tissue Engineering introduction for physicists - Lecture three
 
CBCT in dental practice
CBCT in dental practiceCBCT in dental practice
CBCT in dental practice
 

Kürzlich hochgeladen

Call Girls in Gagan Vihar (delhi) call me [🔝 9953056974 🔝] escort service 24X7
Call Girls in Gagan Vihar (delhi) call me [🔝  9953056974 🔝] escort service 24X7Call Girls in Gagan Vihar (delhi) call me [🔝  9953056974 🔝] escort service 24X7
Call Girls in Gagan Vihar (delhi) call me [🔝 9953056974 🔝] escort service 24X7
9953056974 Low Rate Call Girls In Saket, Delhi NCR
 
Call Girls in Lucknow Just Call 👉👉7877925207 Top Class Call Girl Service Avai...
Call Girls in Lucknow Just Call 👉👉7877925207 Top Class Call Girl Service Avai...Call Girls in Lucknow Just Call 👉👉7877925207 Top Class Call Girl Service Avai...
Call Girls in Lucknow Just Call 👉👉7877925207 Top Class Call Girl Service Avai...
adilkhan87451
 

Kürzlich hochgeladen (20)

Top Quality Call Girl Service Kalyanpur 6378878445 Available Call Girls Any Time
Top Quality Call Girl Service Kalyanpur 6378878445 Available Call Girls Any TimeTop Quality Call Girl Service Kalyanpur 6378878445 Available Call Girls Any Time
Top Quality Call Girl Service Kalyanpur 6378878445 Available Call Girls Any Time
 
Call Girls in Gagan Vihar (delhi) call me [🔝 9953056974 🔝] escort service 24X7
Call Girls in Gagan Vihar (delhi) call me [🔝  9953056974 🔝] escort service 24X7Call Girls in Gagan Vihar (delhi) call me [🔝  9953056974 🔝] escort service 24X7
Call Girls in Gagan Vihar (delhi) call me [🔝 9953056974 🔝] escort service 24X7
 
Call Girls in Lucknow Just Call 👉👉7877925207 Top Class Call Girl Service Avai...
Call Girls in Lucknow Just Call 👉👉7877925207 Top Class Call Girl Service Avai...Call Girls in Lucknow Just Call 👉👉7877925207 Top Class Call Girl Service Avai...
Call Girls in Lucknow Just Call 👉👉7877925207 Top Class Call Girl Service Avai...
 
Premium Call Girls In Jaipur {8445551418} ❤️VVIP SEEMA Call Girl in Jaipur Ra...
Premium Call Girls In Jaipur {8445551418} ❤️VVIP SEEMA Call Girl in Jaipur Ra...Premium Call Girls In Jaipur {8445551418} ❤️VVIP SEEMA Call Girl in Jaipur Ra...
Premium Call Girls In Jaipur {8445551418} ❤️VVIP SEEMA Call Girl in Jaipur Ra...
 
Low Rate Call Girls Bangalore {9179660964} ❤️VVIP NISHA Call Girls in Bangalo...
Low Rate Call Girls Bangalore {9179660964} ❤️VVIP NISHA Call Girls in Bangalo...Low Rate Call Girls Bangalore {9179660964} ❤️VVIP NISHA Call Girls in Bangalo...
Low Rate Call Girls Bangalore {9179660964} ❤️VVIP NISHA Call Girls in Bangalo...
 
Call Girls Mysore Just Call 8250077686 Top Class Call Girl Service Available
Call Girls Mysore Just Call 8250077686 Top Class Call Girl Service AvailableCall Girls Mysore Just Call 8250077686 Top Class Call Girl Service Available
Call Girls Mysore Just Call 8250077686 Top Class Call Girl Service Available
 
Call Girls Kolkata Kalikapur 💯Call Us 🔝 8005736733 🔝 💃 Top Class Call Girl Se...
Call Girls Kolkata Kalikapur 💯Call Us 🔝 8005736733 🔝 💃 Top Class Call Girl Se...Call Girls Kolkata Kalikapur 💯Call Us 🔝 8005736733 🔝 💃 Top Class Call Girl Se...
Call Girls Kolkata Kalikapur 💯Call Us 🔝 8005736733 🔝 💃 Top Class Call Girl Se...
 
Saket * Call Girls in Delhi - Phone 9711199012 Escorts Service at 6k to 50k a...
Saket * Call Girls in Delhi - Phone 9711199012 Escorts Service at 6k to 50k a...Saket * Call Girls in Delhi - Phone 9711199012 Escorts Service at 6k to 50k a...
Saket * Call Girls in Delhi - Phone 9711199012 Escorts Service at 6k to 50k a...
 
Call Girls Rishikesh Just Call 8250077686 Top Class Call Girl Service Available
Call Girls Rishikesh Just Call 8250077686 Top Class Call Girl Service AvailableCall Girls Rishikesh Just Call 8250077686 Top Class Call Girl Service Available
Call Girls Rishikesh Just Call 8250077686 Top Class Call Girl Service Available
 
Call Girls Jaipur Just Call 9521753030 Top Class Call Girl Service Available
Call Girls Jaipur Just Call 9521753030 Top Class Call Girl Service AvailableCall Girls Jaipur Just Call 9521753030 Top Class Call Girl Service Available
Call Girls Jaipur Just Call 9521753030 Top Class Call Girl Service Available
 
Independent Call Girls In Jaipur { 8445551418 } ✔ ANIKA MEHTA ✔ Get High Prof...
Independent Call Girls In Jaipur { 8445551418 } ✔ ANIKA MEHTA ✔ Get High Prof...Independent Call Girls In Jaipur { 8445551418 } ✔ ANIKA MEHTA ✔ Get High Prof...
Independent Call Girls In Jaipur { 8445551418 } ✔ ANIKA MEHTA ✔ Get High Prof...
 
Call Girls Amritsar Just Call 8250077686 Top Class Call Girl Service Available
Call Girls Amritsar Just Call 8250077686 Top Class Call Girl Service AvailableCall Girls Amritsar Just Call 8250077686 Top Class Call Girl Service Available
Call Girls Amritsar Just Call 8250077686 Top Class Call Girl Service Available
 
Call Girls Mumbai Just Call 8250077686 Top Class Call Girl Service Available
Call Girls Mumbai Just Call 8250077686 Top Class Call Girl Service AvailableCall Girls Mumbai Just Call 8250077686 Top Class Call Girl Service Available
Call Girls Mumbai Just Call 8250077686 Top Class Call Girl Service Available
 
Call Girls in Delhi Triveni Complex Escort Service(🔝))/WhatsApp 97111⇛47426
Call Girls in Delhi Triveni Complex Escort Service(🔝))/WhatsApp 97111⇛47426Call Girls in Delhi Triveni Complex Escort Service(🔝))/WhatsApp 97111⇛47426
Call Girls in Delhi Triveni Complex Escort Service(🔝))/WhatsApp 97111⇛47426
 
Top Rated Call Girls Kerala ☎ 8250092165👄 Delivery in 20 Mins Near Me
Top Rated Call Girls Kerala ☎ 8250092165👄 Delivery in 20 Mins Near MeTop Rated Call Girls Kerala ☎ 8250092165👄 Delivery in 20 Mins Near Me
Top Rated Call Girls Kerala ☎ 8250092165👄 Delivery in 20 Mins Near Me
 
Models Call Girls In Hyderabad 9630942363 Hyderabad Call Girl & Hyderabad Esc...
Models Call Girls In Hyderabad 9630942363 Hyderabad Call Girl & Hyderabad Esc...Models Call Girls In Hyderabad 9630942363 Hyderabad Call Girl & Hyderabad Esc...
Models Call Girls In Hyderabad 9630942363 Hyderabad Call Girl & Hyderabad Esc...
 
Premium Bangalore Call Girls Jigani Dail 6378878445 Escort Service For Hot Ma...
Premium Bangalore Call Girls Jigani Dail 6378878445 Escort Service For Hot Ma...Premium Bangalore Call Girls Jigani Dail 6378878445 Escort Service For Hot Ma...
Premium Bangalore Call Girls Jigani Dail 6378878445 Escort Service For Hot Ma...
 
💕SONAM KUMAR💕Premium Call Girls Jaipur ↘️9257276172 ↙️One Night Stand With Lo...
💕SONAM KUMAR💕Premium Call Girls Jaipur ↘️9257276172 ↙️One Night Stand With Lo...💕SONAM KUMAR💕Premium Call Girls Jaipur ↘️9257276172 ↙️One Night Stand With Lo...
💕SONAM KUMAR💕Premium Call Girls Jaipur ↘️9257276172 ↙️One Night Stand With Lo...
 
Call Girls Rishikesh Just Call 9667172968 Top Class Call Girl Service Available
Call Girls Rishikesh Just Call 9667172968 Top Class Call Girl Service AvailableCall Girls Rishikesh Just Call 9667172968 Top Class Call Girl Service Available
Call Girls Rishikesh Just Call 9667172968 Top Class Call Girl Service Available
 
Coimbatore Call Girls in Thudiyalur : 7427069034 High Profile Model Escorts |...
Coimbatore Call Girls in Thudiyalur : 7427069034 High Profile Model Escorts |...Coimbatore Call Girls in Thudiyalur : 7427069034 High Profile Model Escorts |...
Coimbatore Call Girls in Thudiyalur : 7427069034 High Profile Model Escorts |...
 

Learning-based vertebra segmentation, identification and partioning

  • 1. Learning-based vertebra segmentation, identification and partitioning DDH Open, 3 April 2019 Nikolas Lessmann
  • 2. Extrapulmonary diseases in lung cancer screening Cardiovascular disease Osteoporosis Lung cancer
  • 4. Potential new biomarkers • Height loss compared to previous screening round? • Density loss over time? • … CT-biomarkers for osteoporosis Bone mineral density (usually done with dual energy X-ray absorptiometry or QCT) Relative height loss (compared with neighboring vertebrae)
  • 5. • Visible in many different types of scans (different FOV, context) • Difficult to distinguish (anatomical label) • Difficult to separate from each other in low-resolution scans • How reliable are density measures in low-SNR scans? Challenges for automatic vertebra analysis in CT
  • 7. Vertebrae often need to be separated not only from the background, but also from each other Generic algorithm should not make assumptions about: • How many vertebrae are visible • Which vertebrae are visible • Their location within the image = instance segmentation An instance segmentation problem
  • 8. Dedicated loss functions, e.g., arXiv:1708.02551 Detection & segmentation pipelines (e.g., Mask RCNN, based on region proposals) Recurrent attention (instance by instance segm.) Instance segmentation with CNNs arXiv:1703.06870 arXiv:1605.09410
  • 9. • Only one class of objects (vertebra) • Size and shape are relatively consistent • Instances are located in a column, one next to the other A restricted instance segmentation problem
  • 10. Combined detection and segmentation • Segment one vertebra • Look for the next in the neighborhood The network needs a memory, like recurrent networks, to remember which vertebrae have already been found Iterative instance segmentation
  • 11. ROI (field of view of CNN) moves through the image volume
  • 12. Patch-based iterative instance segmentation Image Memory Segmentation 1. Iteration
  • 13. Patch-based iterative instance segmentation Image Memory Segmentation 2. Iteration
  • 14. Patch-based iterative instance segmentation Image Memory Segmentation 3. Iteration
  • 15. • Use prior knowledge that vertebrae are located next to each other • Search for spine, then follow the spine Traversal along the spine
  • 16. I = Image patch M = Memory patch S = Segmentation mask L = Anatomical label Prediction of label 1-24, global probabilistic model to find most plausible labeling C = Completeness score Vertebrae may be visible only partially in the image, relevant for following automatic steps Network architecture
  • 17. Label regression with probabilistic interpretation Prediction: 14.8 Interpretation: • 80% probability label 15 • 20% probability label 14 Maximum-likelihood approach Anatomical labeling Mean certainty per labeling sequence 0.24 0.56 0.89
  • 18. • Some vertebrae are often only partially contained in the scan • For some applications, that is important to know – biomarkers computed based on incompletely visible vertebrae can be unreliable Completeness score
  • 19. Sum of three error terms • Segmentation: Minimize FP and FN voxels • Labeling: L1 norm • Completeness: Binary crossentropy Loss function Weight map to improve segmentation accuracy at the boundary between vertebrae
  • 20. • Adam optimizer with fixed learning rate + high momentum (0.999) • Single patch per iteration (no mini-batches)  each patch has ~2,000,000 voxels • Also patches without vertebral bone are presented to the network • Memory state derived from reference segmentation • Data augmentation: elastic deformations, random Gaussian noise and smoothing, random cropping Training details
  • 21. • Manual and semi-automatic annotation of NLST (low-dose chest CT) scans • Several datasets from vertebra segmentation challenges Evaluation Lumbar spine CT Chest CT Lumbar spine MR (VB only)
  • 24. • We can do vertebra segmentation in CT • We can do vertebral body segmentation in MR • We cannot do vertebral body segmentation in CT… (New training set? Transfer learning? GANs?) Motivation
  • 25. Alternative idea: Let a CNN fit a continuous surface that represents the boundary Surface fitting
  • 26. • Splines are defined by a set of control points • Splines are smooth between control points Think of a stick or a thin metal sheet that is fixated at several points and resists bending due to its rigidity  CNN could predict location of control points Fitting a thin-plate spline with a CNN
  • 27. How to backpropagate through this? Flowchart
  • 28. Differentiable TPS surface? Surface “height” across grid Configuration of grid points w.r.t. control points Configuration of control points w.r.t. control points Control point “heights” (network predicition)
  • 29. If we had vertebra and vertebral body segmentations of the same subjects (and maybe even the same scans), we could train a simpler approach … We have segmentation masks from • Different subjects • Different scans • Different modalities even How to train without matching segmentation pairs?
  • 30. Segmentation masks do not match, but we can learn something from this data – statistical approach, capture plausible shapes • Do not use image intensities at all • Train convolutional autoencoder on vertebral body segmentation masks and use as out-of-distribution classifier Shape encoding with autoencoders
  • 31. Flowchart Trained with vertebra segmentation masks from CT scans (around 600 NLST scans, > 6000 vertebrae) Previously trained with vertebral body segmentation masks from MR scans (around 300 UMCU scans, >1000 vertebral bodies)
  • 32. Reconstruction error + several regularizing terms • Volume penalty: Volume Vertebra – Volume Vertebral Body • “Fights” the reconstruction loss by pushing the surface to the right (posterior) • Empty segmentation masks are perfectly reconstructed by the AE… • Curvature penalty: abs(det(Hessian)) • Prevents that surface is very wobbly (seems not really necessary) Loss function
  • 34. Manual annotation of vertebrae and vertebral bodies in 5 chest CT scans Results
  • 35. Measurements of the height of the vertebral bodies and correlation with extracted volume Average of 3 measurements in 389 scans Results
  • 36. We can… • Find and segment the vertebrae • Label them anatomically • Determine if they are completely or partially visible • Partition them into body and posterior elements Next step: Application (fracture prediction, fracture classification, …) Also possible: use partitioning results to train direct vertebral body segmentation Conclusions and future work