Embed
Email

maps_ mash-ups and metadata

Document Sample

Shared by: wuzhenguang
Categories
Tags
Stats
views:
0
posted:
1/8/2012
language:
pages:
49
maps, mash‐ups and metadata



medical imaging and mapping 

medical imaging and mapping

compared and contrasted

topics

• … just what is a map anyway?

• 21st Century mapping

• Mashups = meta‐analysis?

• When it goes wrong

• Particle Radiotherapy

• Pathology and ‐omics

• Comparison of map‐based workflows

• Concluding remarks

what is a map (OED)

what is a map (OED)

• 1 a. A drawing or other representation of the earth's surface 

or a part of it made on a flat surface, showing the distribution 

of physical or geographical features (and often also including 

of physical or geographical features (and often also including

socio‐economic, political, agricultural, meteorological, etc., 

• …

• 3 a. A diagram or collection of data showing the spatial 

distribution of something or the relative positions of its 

g p

components. Freq. with distinguishing word.

more functionally…

more functionally

• A diagram derived from an image or a survey, with

– components labelled, coloured or otherwise annotated according to 

one or more legends

one or more legends

– (optionally) associated with a coordinate system

– further metadata for scale, provenance, projection…

• A legend links a conceptual model appropriate to the diagram 

with features in the diagram through the use of symbols or 

colour

• A coordinate system allows registration of the diagram with 

lit th fd t d t d t

reality or other sources of data and metadata

mash ups

‘mash ups’

• Dataset joining from unconnected data sources facilitated by 

common keys, labels or coordinates unanticipated by the 

originators of the data

originators of the data

• Most commonly map based

Coordinate systems ubiquitous, easily identifiable within text, often 

– Coordinate systems ubiquitous easily identifiable within text often

accurate

• Latitude and longitude

• Postal code

– Well documented services and APIs

Compelling: collision of mobile devices, high bandwidth networks, 

– Compelling: collision of mobile devices high‐bandwidth networks

GPS, satellite imagery 

• A user defined legend, often with ‘crowd sourced’ data

• A form of ‘meta‐analysis’

It commonly happens that there is no one conclusive study 

but 'meta‐analysis' gives you a combined result from every 

randomised trial that has been done around the world. For 

example, the drug Tamoxifen ‐ an oestrogen blocker that may 

prevent breast cancer cells growing ‐ was the object of forty

prevent breast cancer cells growing  was the object of forty‐

two studies world‐wide ‐ of which only four or five had shown 

significant benefits. But this did not mean that Tamoxifen did 

t t t i tb t 'Wh t ll th

not protect against breast cancer. 'When we put all the 

studies together it was blindingly obvious that it does ‐ you 

don't have to be a medical statistician to see that. Nor do you 

y

need to be an economist to see the advantages of saving tens 

of thousands of lives with this inexpensive drug’



Prof. Richard Gray, University of Birmingham

early breast cancer trialists’ 

collaborative group



• Initiated in 1983

• Hundreds of institutions worldwide

• Consensus on 30 variables, a data model and submission 

format

Analyse data every 5 years

• A l d 5

• Computable data and follow‐up for 200,000 cases in the 2000 

review

• Rock‐solid evidence base for the treatment of early breast 

cancer

polluter zip code ‐> Google map

polluter zip code > Google map

• The US EPA took the zip code of their polluter register and 

plotted it on Google Maps

Anomalous heavy polluter appeared in downtown San 

• A l h ll t di d t S

Francisco…

the address for a noted polluter was actually that of its legal

• … the address for a noted polluter was actually that of its legal 

representatives

caveats

• Metadata is expensive, so how good is free metadata?

• Often rely on implicit data standards – no information on 

li

compliance or accuracy

• Often lexical matching – unreliable 

Crowd sourced information – no provenance

• C d di f i

• Originator of data may be biased

h h ll f h l d df h

• Might not have all of the contextual metadata required for the 

proper interpretation of the data

PARTICLE RADIOTHERAPY

tumour



100







80

SOBP

Dose (%)









60

D









40

P i ti k

Pristine peak



20









50 100 150

Depth (mm)

combining fields

combining fields



80 50







60 150 0

150







40 0









y

X‐Rays Protons/Ions

Paravertebral Epithelioid Sarcoma

Intensity Modulated Protons (IMPT) vs. 

Intensity Modulated Photons (IMRT) 7 (field)

y ( ) ( )



IMPT IMRT

greater precision and control means…

greater precision and control means

• We can:

– irradiate tissue that is close to important organs

boost the tumour dose without increasing the dose to the surrounding 

– boost the tumour dose without increasing the dose to the surrounding

tissues

– reduce immediate side effects of radiotherapy

• Ocular; CNS; sarcomas esp. where resection is difficult or 

incomplete; paediatric cancers; prostate; breast tumours…

MAPPING IN RADIOTHERAPY

the problem

the problem

• Based on the consensus of opinion, 32% of radically treated 

patients should receive inverse‐planned IMRT and 22% 

forward‐planned IMRT, making a total of 55%. In fact, 2% 

forward planned IMRT making a total of 55% In fact 2%

receive inverse‐planned IMRT and 11% the less complex 

p , ,

forward‐planned IMRT. Thus, with an estimated 75,948 radical 

treatments being carried out with megavoltage radiotherapy, 

the professional opinion is that 41,421 of patients would 

benefit from treatment with IMRT. In fact, only 9775 were so 

b f f h f l

treated in 2008; a shortfall of 32,497 patients treated instead 

with conventional radiotherapy.

with conventional radiotherapy

– Survey of the availability and use of advanced radiotherapy technology 

in the UK.  Mayles et al, Clin Oncol (R Coll Radiol). 2010 Oct;22(8):636‐

42. Epub 2010 Jul 27.

radiotherapy underutilised

radiotherapy underutilised

The machines are in the trust, but have insufficient resources 

• Th hi i h b h i ffi i

to plan advanced treatments – even for IMRT

Compounded by need to plan in four dimensions (time)

• Compounded by need to plan in four dimensions (time)

– tumours shrink (we hope)

– tumours move (lung cancer)

• Need automation

– organ recognition

– delineation of target and treatment margins

delineation of target and treatment margins

– better calculations of dose

– comparison of multiple plans and multiple modalities

– calculation of risk factors such as malignant inductance probabilities

– through complex calculations and Monte Carlo simulations

ICRP Publication 110 Phantoms

125 119 119 119 109 109 109 109 109 109 109 109 109 109 109 109 109 109 109 

109 109 109 109  28  28  28  28  28  28  28  28  28 109 109 109 109 109 ...

Elemental 

Composition

N H Na S

4.2 3.6 0.3 0.3 Mg

0.2

P

9.4

O

44.8

C

15.9

Ca

21.3



Tissue №

voxel Mineral bone Blood/bone 

marrow content

Organ № 1% blood, 0% red 

28 ‐ Femora, upper  bone marrow, 0% 

half, cortical yellow bone marrow, 

100%  bone



Mass Density

1.92gcm‐3

ICRP Publication 110 Phantoms









ICRP Adult Male Phantom ‐ from left to right: anterior surface, coronal slice showing mass density, mass density 

integrated over coronal slices, calcium content integrated over coronal slices

Monte Carlo simulations

• Using the Monte Carlo code MCNPX  Remainder Tissues

1.2

(transports photons, protons and 

( p p ,p 1.0









)

efficient (Gy/Gy)

heavy ions) 0.8

0.6



• Implemented phantom geometry 0.4









Coe

0.2 JJ

• Benchmarked by comparing x‐ray  0.0 DW





organ dose conversion coefficients  0.01 0.1

Photon Energy (MeV)

1





with Jan Jansen of Health Protection 

Agency

Can perform simulations and extract 

• C f i l ti d t t

3D dose data – useful for comparisons 

with treatment planning

with treatment planning

conversion to DICOM

• Can replace each voxel with an 

estimated Hounsfield Units (HU) 

( )

value, then export to DICOM 

using CERR toolkit for Matlab

• Allows comparison of treatment 

planning systems with Monte  ICRP 110 Organs

(Soft Tissue Region)

( g )

Carlo simulations, investigation of 

C l i l ti i ti ti f









pping Power

1.2

HU vs electron density/proton  1.15

stopping power

stopping power









at 100MeVV

Relative Proton Stop

1.1

– Potentially requires more care for  1.05

protons given sharp dose gradients

1







e

0.95

-100 0 100 200 300

CT Number (Hounsfield Units)





25

system diagram



XiO TPS MATLAB MCNPX

Dan s 

Dan’s

CERR

code



Clinical

DICOM planC

l C Ph t

Phantom

Monte Carlo

Monte Carlo

Treatment

Simulations

Plans

Dan’s 

CERR

code









l l ki difi i h h

• Also looking at modifications to the phantom – i i

inserting a tumour, 

adding finer scale structure to the lungs and bone

MIP Calculations

MIP Calculations

• Heatmaps for the probability of malignant induction for 

different treatments (with the Mayo Clinic)

atypical meningioma

– Subtotally resected atypical meningioma

– Compare plans for different types of radiotherapy

• 3D Conformal RT (standard treatment)

• IMRT

• RapidArc (helical RT)

• Spot scanning protons

Spot scanning protons

– Account for fractionation

Dose (GyE) ‐log10(Surviving Fraction)



protons

p

IMRT

Cells Transformed per cm3



x10‐3

protons

p









x10‐3

IMRT

• (show the video…)

multimodality image workspace for 

contouring

Commercial 

Commercial

TPS

CT

CT

CT

CT Push data back

Push data back 

CT

CT into commercial 

TPS vis DICOM‐RT





PET PET

CT

CT

CT

CT

CT

MRI Fast image  Automatic segmentation 

registration of normal tissue structures



P it f t t

Persistence of structures 

CT

CT across image modalities

CT

CT

CT

PET

AccelRT

/

• a Cluster/Grid‐enabled treatment planning workbench for 

DICOM images

automatic draft delineation of tumour and adjacent tissues in four 

– automatic draft delineation of tumour and adjacent tissues in four

dimensions

– ‘find‐one‐like‐it’ searches across image repositories

– declarative extension for new services

• led by clinicians at University of Cambridge/Addenbrookes

• segmentation, registration services: Cambridge eScience

• workbench, service declaration, metadata management: 

PTCRi

PATHOLOGY IMAGE PROCESSING

automation of scoring

automation of scoring

• Pathologist annotation is a bottleneck and expensive

• Use astronomy image analysis to provide metadata to the 

tissue microarray maps

ti i

• Publish the image set, together with legend and metadata for 

further meta‐analysis and DAN studies

further meta analysis and DAN studies

• Requires automatically generated maps and annotations of 

the microarrays

the microarrays

TMA definitions in a metadata registry

TMA definitions in a metadata registry









4

Construct TMA and 

Create EXCEL map 







ARIOL server

TMA database

1. Transform to XML  SQL Server

DEEP ZOOM  2. Import slidemap

slidemap

i

images  ARIOL WEB

stored on  SERVICE 

file system (AWS)





7. Export  4. Export images+ 3. Accept requests for 

scores  metadata images

6. Manual on‐line scoring



JPEG images

5. Rename images





Batch Deep Zoom 

Converter program





Statistical  Pathgrid

Image  analysis image 

User

Scorer analysis

Manual score: Intensity 3, Proportion 4 = 7

ER positive









ER negative Manual score: Intensity 1, Proportion 1 = 2

10

Pathgrid automated score for ER in highly correlated with the 

Allred manual score.

Allred manual score.

‐ summary of 1571 core images (1183 ER positive, 388 ER negative)

‐ 92.4% sensitivity, 93.7% specificity









8



7

6

5



4









ER+

3



2









ER‐

0









11

Pathgrid automated score for HER2 in highly correlated with the 

manual score

Herceptest manual score

‐ summary of 1775 core images (1573 HER2‐, 115 HER2 borderline,  87 HER2+)

‐ 96.5% sensitivity  97.6% specificity









3 HER2+

2 borderline



1

HER2‐

HER2

0

unscored









15

CONCLUDING REMARKS

getting from A to B in the 21st Century

getting from A to B in the 21



Workflow Route planning

• Image acquisition • Satellite

• Map registration • Automatic

g

• Feature recognition y

• Aided by dataset

• Dataset joining • Features of interest, traffic

• Human annotation / p

• Crowd sourced/subscription

• Making the decision • Automated algorithms

• Execution g

• GPS guidance

the workflow remains the same

… the workflow remains the same

Workflow Treatment planning

• Image acquisition • CT, PET, Ultrasound

• Map registration • Tattooed reference point

g

• Feature recognition • Want to be automatic

• Dataset joining • Elemental composition

• Human annotation p

• Expert sourced

• Making the decision • Manual treatment planning

• Execution • Servo actuated relative to 

reference point

the workflow remains the same

… the workflow remains the same

Workflow Tissue sub study ‐> evidence

• Image acquisition • Photograph of microarray

• Map registration • Spot coordinates

g

• Feature recognition g g g

• Astrogrid image recognition

• Dataset joining • Clinical data about subject

• Human annotation p (p g )

• Expert sourced (pathologist)

• Making the decision • Statistical analysis

• Execution p p

• Paper publication

common functions and workflow

common functions and workflow

• Many process are the same at some level of abstraction from 

a software perspective

Facilitates the development and reuse of common software

– Facilitates the development and reuse of common software

• Allows us to define the workflow and transformations at a 

higher level from user perspective

higher level from user perspective

– more automation of the analysis process

– but science will always want to be cutting edge

• Explains why we are all here already sharing and reusing 

software

acknowledgements

• Raj Jena (Addenbrookes)

Lorna Morris (CRI, Cambridge)

• Lorna Morris (CRI, Cambridge)

• Bleddyn Jones, Claire Timlin, Dan Warren 

(PTCRi)

thank you

Steve Harris

James Martin Research Fellow

James Martin Research Fellow

Oxford University Computing Laboratory

Parks Road, Oxford. OX1 3QD

Tel: 01865 610653

Fax: 01865 283531

F 01865 283531



steve.harris@comlab.ox.ac.uk

steve.harris@ptcri.ox.ac.uk

http://www.comlab.ox.ac.uk/people/Steve.Harris/

htt // l b k/ l /St H i/

http:/www.ptcri.ox.ac.uk/people/#PicSH



Related docs
Other docs by wuzhenguang
Is Air Quality a Problem in My Home
Views: 8  |  Downloads: 0
IHRM Chapter 6
Views: 9  |  Downloads: 0
37.10593
Views: 7  |  Downloads: 0
December_break
Views: 8  |  Downloads: 0
Lectures for 2nd Edition
Views: 9  |  Downloads: 0
Google Chart
Views: 30  |  Downloads: 0
By registering with docstoc.com you agree to our
privacy policy

You are almost ready to download!

You are almost ready to download!