scch is an initiative of scch is located in from image analysis research to software for clinical,...
TRANSCRIPT
SCCH is an initiative of SCCH is located in
From image analysis research to softwarefor clinical, biomedical and healthcare environments
Dr. Julian Mattes
+43 7236 3343 [email protected]
2© Software Competence Center Hagenberg GmbH, MATTES Medical Imaging GmbH
Introduction
3© Software Competence Center Hagenberg GmbH, MATTES Medical Imaging GmbH
Introduction
4© Software Competence Center Hagenberg GmbH, MATTES Medical Imaging GmbH
Introduction
5© Software Competence Center Hagenberg GmbH, MATTES Medical Imaging GmbH
Aortic stent grafts
Endovascular treatment of an aortic aneurysm
Abdominal aortic aneurysm
Stentgraft thoracic
Stentgraft abdominal
6© Software Competence Center Hagenberg GmbH, MATTES Medical Imaging GmbH
Medical Problem
Complications due to a stentgraft
Problems: Leak by (local) „displacement“ Leak by fatigue of material Other leaks Occlusions due to kinks
Reasons: Changes in the aneurismatic sac Elongation of the aorta Technical fatigue (local) displacement Technical fatigue leak
Stentgraft abdominal
7© Software Competence Center Hagenberg GmbH, MATTES Medical Imaging GmbH
Medical Problem
Data / Aim of the analysis
Stentgraft abdominal
1 Patient, 2 CT volume scans, interval: 11 months
Goal:Recognition of an initiating complication
Individual risk assessment Preventive endovascular re-intervention
8© Software Competence Center Hagenberg GmbH, MATTES Medical Imaging GmbH
Software Solution
Approach for early recognizing complications
Software for
A highly automated generation of the stent‘s surface
Basis for 3D-Measurements and superimpositions
Superimposition of the stent surface for different points
in time
Representation of the migration with respect to a
reference structure
9© Software Competence Center Hagenberg GmbH, MATTES Medical Imaging GmbH
Software Solution
Deformation analysis
Superimposition of the stent‘s surface for different points in time
Quantitative value for the amount of deformation
Mean length of motion vectors: 5,65 mm
10© Software Competence Center Hagenberg GmbH, MATTES Medical Imaging GmbH
State / Further requirements
Software
allows reproducing conventionally measured parameter with
the same precision and reliability (J.Mattes et al. EJR 2012 81(3)496-501,Epub 2011)
helps avoiding to overlook complications
helps better recognizing and assessing initiating complica-
tions second reader for a preventive re-intervention
Requirements for industrial exploitation
a study of a larger patient population
Integration of additional anatomical structures
Minimal user effort, integration in the radiologist’s work flow
Image analysis application fields
Röntgen
MR
USPET
Pre-operative planning of paths of surgical tools (e.g., on MRI). Markers are fixed on patient and visible in the pre-operative image.
During operation the fixed markers are localized by the same system as the diodes on the surgical tool. They are co-registered with the markers on the pre-operative image.
“Seminal paper” (Russ Taylor): Benabid, Cinquin, Lavallee, Le Bas, Demongeot, Rougement (1987): A computer driven robot for stereotactic surgery connected to CAT-scan magnetic resonance imaging. Technological design and preliminary results. Applied neurophysiology, 50: 153-154
Computer aided diagnosis
www.praxim.fr
Computer assisted interventions: planning, navigation
12© Software Competence Center Hagenberg GmbH, 2015
Overview
Introduction Example: Shape change of an implant – aortic stent grafts
General aspects Innovation chain Exploitation strategy Market Software quality
Further examples Image analysis in histopathology Cellular motion and anti-cancer drug investigation A foot scanner technology
13© Software Competence Center Hagenberg GmbH, 2015
General aspects
14
Innovation chain
Reproduced from Tirillum Power Wind Corporation
15© Software Competence Center Hagenberg GmbH, 2015
Innovation chain
Example stent grafts Technology push new technique non-rigid surface registration Market pull need for quantification tool, integrated in work flow
Reproduced from Tirillum Power Wind Corporation
16© Software Competence Center Hagenberg GmbH, 2015
Exploitation strategy
Development on demand (market pull) Example foot scanner
Development on own initiative Often driven by technology push Within subsidized project (possibly with industrial partners) Own patents Exploitation via royalties for software, patents Combination with services Exploitation via start up
Patents Software patents??
17© Software Competence Center Hagenberg GmbH, 2015
Market
Knowledge about market Competitors / potential partners State of products Size of market Framework requirements Market entry barriers (certification, access to PACS)
18© Software Competence Center Hagenberg GmbH, 2015
Software quality
Certification EU directive for medical products (EN 62304, EN ISO 13485,
EN ISO 14971, EN 62366) Software development life cycle (software development
process) Requirement gathering and analysis (documentation of
specifications) Design Implementation or coding Testing Deployment Maintenance
Documentation, review
19© Software Competence Center Hagenberg GmbH, 2015
Examples
20© Software Competence Center Hagenberg GmbH, 2015
Examples
o Shape change of an implant – aortic stent grafts Image analysis in histopathology Cellular motion and anti-cancer drug investigation A foot scanner technology
21© Software Competence Center Hagenberg GmbH
Image analysis in histopathology
Project TisQuant: Procedures for highly automated assessment and classification of cells in tissue sections by means of spatial marker profiles
Challenge: Association of individual cells and stained cellular markers
Assessment of drug treatment Assessment of tumor agressivity 7 partners from Germany and Austria
3 scientific partners: SCCH, University of Vienna, DKFZ (German Cancer Research Center)
4 SMEs (2 from Germany and 2 from Austria)
2 partners are end users (DKFZ and the SME LabDia from Vienna)
TisQuant – Key components
22© Software Competence Center Hagenberg GmbH
Technical / Image analysis Automated cell detection/segmentation User interaction work flows Generation of ground truth Quatitative analysis and classification
Experimental research Identification and establishing of a membrane marker
System integration Integration of the system of 3 SME partners Testing under lab conditions
Neuroblastoma, IHC stained bright field image (part)
Acquired within TisQuant (Labdia, TG)
23© Software Competence Center Hagenberg GmbH, 2015
Automated cell segmentation
Detektion/Segmentierung Zellkerne – klassisch histologisch
ForegroundMap
Binary Image(Cell Clumps)
Final Segmentation
FG / BGSegmentation
HistologicalImage (RGB)
Recursive ClumpDecomposition
24© Software Competence Center Hagenberg GmbH, 2015
Automated cell segmentation
Detektion/Segmentierung Zellkerne – klassisch histologisch
Dorfer et al. (2016) Submission to ISBI in preparation
25© Software Competence Center Hagenberg GmbH, 2015
Automated cell segmentation
Detektion/Segmentierung Zellkerne – klassisch histologisch Recursive Waterflow
Dorfer et al. (2016) Submission to ISBI in preparation
Zerlegungs-Baum erzeugt durch rekursives Clump-Splitting
User interaction work flows
26© Software Competence Center Hagenberg GmbH
Determination of segmentation parameters
Classification of cells
Supported ground truth labelling
Kromp et al. (2015) Computer Vision Winter Workshop Schloss Seggau, 2015
European Patent Pending
27© Software Competence Center Hagenberg GmbH, 2015
Examples
o Shape change of an implant – aortic stent graftso Image analysis in histopathology Cellular motion and anti-cancer drug investigation A foot scanner technology
28© Software Competence Center Hagenberg GmbH, 2012
Cellular motion
Cellular motion is relevant for the metastatic risk related to a tumour wound healing processes in developmental biology
Anti-migratory anti-cancer strategies reduced cellular motion as a response to an anti-cancer drug as
cetuximab increased cellular motion when the epidermal growth factor
(EGF) binds to its receptor on the cell membrane cetuximab blocks this binding
Investigating the mechanism of the motile behaviour of cells How alters EGF the motile behaviour of a cell?
29© Software Competence Center Hagenberg GmbH, 2012
CANCERMOTISYS
Cellular motion in the project
“ CANCERMOTISYS – Systems Biology of drug effects on the motility of gastric cancer cell lines“
www.cancermotisys.eu
Quantitative motion analysis:The epidermal growth factor (EGF) is stimulating the motile behavior of cancer cells. This can be prohibited by the anti-body cetuximab. In order to demonstrate such differences in motion precise quantitative measurements are necessary.
Molecular basis of cellular motion:Differences in gene expression are correlated to changes in the motile behavior
30
Cell detection – problem
Cell detection in two problematic situations:1. Cells may agglomerate and form clusters.2. Cells may have a complex appearance with high variation in
between cells.
© Software Competence Center Hagenberg 2012
Input-Data: IAPA, TU München, AG Prof. B. Luber Gastric carcinoma cell line MKN1
Cluster of 5 cells.
Cell with complex appearance.
31
Cell detection – approach
Cell detection – an approach based on machine learning and texture partitioning
© Software Competence Center Hagenberg 2012
Input data
Texture partitioning Filtered and joined seeds Evaluation – blue: detections, yellow: ground truth
Dark seedsBright seeds
T. Kazmar et al. 32nd IC IEEE EMBS 2010
32
Cell detection – results
© Software Competence Center Hagenberg 2012
Results – including very difficult cases: If a cell is automatically labelled this is correct in 91% of the cases (precision) 83% of all cells are correctly detected (recall)
Evaluation on more than 300 single images with up to 80 cells each.
Results:
: all cells labelled manually
: automated detection
: cells not detected
: over-detection
All 5 cells have been correctly detected; no over-detection.
Correct detection of the cell with complex appearance; no over-detection.
Input-Data: IAPA, TU München, AG Prof. B. Luber Gastric carcinoma cell line MKN1
33
Motion parameters
© Software Competence Center Hagenberg GmbH 33
Radial EffectivityMean velocity Mean area
Motion parameters derived from cell tracking.
Time resolved parameters
Instantaneous velocity
34
Example
© Software Competence Center Hagenberg GmbH 34
unbehandeltEGF-behandelt
untreated EGF-treated
35© Software Competence Center Hagenberg GmbH, 2015
Examples
o Shape change of an implant – aortic stent graftso Image analysis in histopathologyo Cellular motion and anti-cancer drug investigation A foot scanner technology
BootDoc foot analyzer
Device for measuring foot sizes based on photograph Aim
no cover requested, works if client leaves socks on
Previous device
BootDoc foot analyzer
dark field
Solution: New hardware Dark field around feet LED lights in stripes (red, white) Fish-eye lens Allows keeping box small
Software fast image analysis based on
MSER detection varying exposure time to get
the best results possible life video stream is available
unlike with competing scanners
length = 268.9/264.0 mmfront width = 98.8/99.9 mmheel width = 74.3/74.0 mmfoot type = 8.3/8.7 mm
naked feetblue/red/green socks
image undistortionbackground suppression
color adjustmentcandidate contour detection
best contour selectionprecise measurementsfoot image extraction
BootDoc foot analyzer
BootDoc foot analyzer
Works with or without socks!
automatic measurements length front width heel width foot type (pronation/supination)
precision under +-1.5mm disadvantages of the predecessor model solved
uncomfortable foot cover removed analysis even for feet with socks on mirror-less construction with no interfering
reflections large area for feet available smaller box thanks to fish-eye lens used
40© Software Competence Center Hagenberg
Conclusion
Mixture of technology push and market pull
Question task formulated by biologist/user
Precise and verificable requirement specification
Exploitation model royalties
Good market knowledge, company partner
Accept user interaction but keep it efficient
41© Software Competence Center Hagenberg
Acknowledgment
Universitätsklinik für Radiodiagnostik, Innsbruck
Iris Steingruber-Chemelli, Andreas Chemelli, Werner Jaschke
Institut für allgemeine Pathologie und pathologische Anatomie, Technische Universität München (München)
Birgit Luber, Julian Kneisel, Simone Keller
Labdia Labordiagnostik GmbH
Peter Ambros, Florian Kromp
BootDoc/Wintersteiger GmbH
Andreas Lachinger
Software Competence Center Hagenberg
Tomas Kazmar, Matej Smid, Matthias Dorfer