1

[Medical] Image Computing:A Vision of the Future

Medical Physics and Bioengineering,

Andrew Todd-PokropekUniversity College London,A.Todd@ucl.ac.ukwww.medphys.ucl.ac.uk

Welcome to (just outside) Szeged

Geometrical Problem

• What geometrical shape can you form from12 Pentagons20 Hexagons

With sides of equal length

2

The Vision?

McCoy

Medical Imaging

Temporal resolution

Spatial resolution

(mm)

EEG / MEG

fMRIPET

1 ms 1 second 1 min 10 min

8

6

4

2

10

MRIaMRId

NIR

MRS

• Overview– Teamwork– Vision- How can we obtain images– The past– The present– The future?

+digression on PCA/ICA

Medical Physics and Biomedical Engineering

MedPhys&BioEng

HealthCareMedicalScience

EngineeringBasicScience

Exchange of Information

• From Basic Science: for new ideas and techniques• From Engineering: for design, practicality and function• From Medical Science: for applicability, biological

soundness• From Healthcare: for technology transfer and

evaluation (and benefit to patients).

Multi-disciplinary teams are required

3

Acknowledgements

Colleagues and Collaborators:CMIC (in particular Dave Hawkes), CS, Medical Physics and BioengineeringMIAS IRC (UCL, Oxford, Imperial, Manchester)INSERM U494Harvard, Georgetown, Leuven, INRIA,EPSRC, MRC, Welcome, Siemens, Philips, GSK, and many others.

VISION (IN HEALTHCARE)

• Vision (the use of images) is a major source of information (in healthcare)

• We can use images to extract (clinical) information affecting diagnosis and therapy

• We can use image to show complex data in a more easily perceived manner (data visualisation)

‘The eyes (and ears) of medicine’ (The Times)

Other types of visual data:Epidemiological Studies

Philippines

Notavailableorzero

Populationatrisk(%), 2000<55-1010-2020-4040-8080-9090-100

Evolution of AIDSEvolution of AIDS

MalariaMalaria

A TGC

CC GCCG

AC

CA

AA TT

TC

CCC

ACC

CAC

AAC

CCA

E coli genome

M. leprae

M. musculus

A. fulgidus

Genome signature

Chaos Game Representation(CGR)

But of course more conventional medical images

CTCT

MRMR

VisualVisual

XrayXray

Gamma

X-raysMicrowavesRadio

IR / visible / UV

4

Any imaging procedure requires• A probe (sound, light, X-rays, etc)• A perturbation (energy absorbed)• A measurement (a detector)

•Image formation is often an inverse problem

An example from NIR Optical Tomography

• MONSTIR is a 32 channel time-resolved system for measuring these temporal point spread functions (TPSFs).

Fibre switch

AckAck E E HIllmanHIllman

The future

• It is very hard to make predictions

Especially about the future[Yogi Berra]

• So let’s look at the past first.

The first medical physicist?Hippocrates (~460-377 BC), Founder of medicine

Has theory that says health is a balance between blood,phlegm, black bile, yellow bile. Disease is an imbalancebetween these “humours”

Each “humour” has a property - blood carries heat. Too muchbody heat is due to too much blood and therefore causesillness.

Can we get an image of body heat to work out where theillness is?

Hippocratic Thermography

For pleurisy:

“Soak a piece of fine linen in warm moist finely titrated Eretrian earth; then wrap this all the way round his thorax, and, wherever it first dries, there you must cauterise or incise, as close to the diaphragm as possible, but sparing the diaphragm itself”

Photographs of Hippocraticthermogram taken at 2 minute intervals after aregion of the back has beenheated using red pepperextract.

Can identify hot spot at 5-10 minutes

5

Dissection• Dissection is not medical imaging but it was the only real

way to “see” inside the body before x-rays

• Human dissection first used systematically in Renaissance in new universities (c.1300).

• Continued and got more sophisticated until 1800s.

• Still used now to teach medical students and for post mortems, though there are attempts to use MRI for non-invasive post mortems.

Human thoraxLeonardo da Vinci

Virtual HumanNLM

First steps in endoscopy

• Hippocrates described rectal speculum, Romans used vaginal speculum

• 19th Century: rigid tubes with light source and mirrors for gynaecology

• 1869: Panteloni used a rigid endoscope to find and cauterise a uterine growth

• 1869: Kussmaul used 50 cm rigid endoscope on professional sword swallower

Colon segmentation

Key question: how muchdoes the tumour invadeThe colonic wall.

Ack Sorantin

Wilhelm Conrad Röntgen (1845-1923)

Discovery of X-rays

Quote from Lord Kelvin‘X-rays will turn out to be a hoax’

Röntgen’s wife’s hand

6

Equipment

Lead battery

Mechanical interrupter

(20 Hz)

Discharge tube Spark inductor 50 kV

Vacuum pump

Communicating the discovery• Discovered → 8th November 1895• “Eine neue Art von Strahlen” → 28th December 1895• Wrote to colleagues → 1st January 1896• Reported in newspaper → 5th January 1896• First clinical x-ray → 7th January 1896• Reported in Lancet → 11th January 1896• Public demonstration → 23rd January 1896• Breast cancer treatment → 29th January 1896• Fluoroscopy → March 1896• Anti X-ray underwear → March 1896• Patient’s hair falls out → April 1896• 1000 papers published on x-rays in 1896• Nobel Prize → 1901

First clinical UCL x-ray• X-ray image of needle in

woman’s finger taken by a UCL Professor of Chemistry, Norman Collie, in 1896

• Probably the first clinical x-ray in Britain

• Norman Collie (1859-1942) was also a respected mountaineer with 21 first ascents in Canadian Rockies

A short digression about PCA and ICA

• What is PCA really about• How big is the Covariance matrix• How does ICA differ from PCA

How to perform PCA

• Build covariance matrix as data T . Data• After centering about mean• Diagonalize to give -• Eigenvalues (estimates of variance)• Eigenvectors (principal axes)

• Almost the same as KL and Factor analysis• Note about correlation

PCA as regression

• Regress x on y or y on x• Regress to minimise

perpendicular distance

• Data matrix is n examples of m variables

• Covariance matrix is mxm

7

PCA ICA

Digital subtraction angiography

Before After

Subtraction

Computed tomographyPlanar x-ray: 2D projection of 3D data

CT: 2D slice through 3D data

State of the art

New tools-Interactive 3-D display

AckAck A A LinneyLinney

MRI

8

Cerebral Vascular Accident

Right Breast Left breast Right Breast Left breastBlood VolumeBlood Saturation

OpticalTomography

Imaging of blood filled tubes in intralipid by photoacousticimaging (P. Beard, BORL, Medical Physics, UCL)

• Excitation: 800nm (6.7mJ/cm2); pulse duration=8ns• Lateral scanning step-size: 140µm

14mm x 14mm

6262µµmm100100µµmm 300300µµmm

Examples of EIT development work at UCL

Perspex rod

Images reconstructed using :

homogenous sphere

full shelled head

1.1 nm-1

1.4 nm-1

1.7 nm-1

transmission image at 17.4 keV

Diffraction images at different momentum transfer values

Excised breast tissuephantom

DEBI

Application to in vitro breast tissue samples

conventionalmammography

SR phase contrastimaging on screen-filmsystem

SAME DELIVERED DOSE(1.5 mGy)

F. Arfelli et al. Phys. Med. Biol. 43 (1998) 2845-52

9

Jun Li et al., J. Anat 202, 463 (2003)

X-ray Phase Imaging (Ian Robinson)

F Pfeiffer & Christian David, Paul Scherrer Institut

F Pfeiffer & Christian David, Paul Scherrer Institut

X–ray imaging

New Flat–field correction method

Raw dataStep 1Step 2Step 3Step 4

Ground beetle

Flat-field correction for hardening of the beam

Granja (Medipix project)

Tim Watson, Richard Cook (GKT)Tony Wilson (Oxford)

In Vivo Micro- Endoscopy

Vasculature of the ventral aspect of the posterior third

of the tongue.

Molecular Imaging

• Viewing physiological processes– Direct uptake of tracers– Indirectly via activated receptors– Major applications

• Oncology• Cardiology• Neurology

– Impact on gene expression and therapy

The process The use of reporter genes

10

Small animal PET scanners

PET Optical

250m

m

Rat bone scan

90mm

400g rat 30g mouse

• Rat: 400g, 65x250mm• 5mCi Tc99m-MDP• 30min. scan, 1.5h p.i.• Apt3, Ø=2.0mm• scan range 250mm

zoom of rat spine

Synthesis at all “biological levels”Synthesis at all “biological levels”

Bioinformatics

Medical ImagingMedicalInformatics

Public HealthInformatics

Integrating biomedical data for better health

EU – FP7

Cardiac acquisition (and registration)

Clinical Issues

Spectra in the internal carotidMyocardial functional imaging

Automatic Estimation of regions of interest from

1st passage in NMR

Estimation of contraction (Vx) intra-myocardial

NMR velocity estimation Markovian approach

Imaging ofcontraction - perfusion

Hibernation - Stunning

Left Ventricle Right ventricle

Myocard

11

Diffusion Weighted ImagingTractography

Ack Clark

ab

c fe

g

Ack Benali

A Project –Ack Unité 494 INSERM, LENA UPR 640 CNRS, Unité 483 INSERM, Centre MEG, Service de Neuroradiologie, Hôpital Pitié-Salpêtrière

•in collaboration with existing projects

Pre-operative

fMRI

Post-operative

fMRI

Per-operativeCortical stimulationPre-operative

multi-modality visualisation

Establishing a clincal interface Integrating statistical

models *How to make itHow to make itpractical/ routinepractical/ routine

Functional images and surgery

Association maps

Before surgery

Ativation maps

After surgery

Malignant melanoma with normal liver CT & US

Courtesy of: Liselotte Højgaard,MD DMSc, Annika Eigtved, MD ph.d., Anne Kiil Berthelsen, MD. PET & Cyclotron Unit, Dept. Nuclear Medicine, Rigshospitalet, University of Copenhagen.

PET / CT

RF Ablation of Lung Tumours

Images Courtesy of Bill Lees

Combined X-ray and MR guided ElectroPhysiology

MR X-RAY

Rhode et al TMI 2003 and 2005; Razavi et al, Lancet 2003, Guy’s Hospital, London

12

Real-time GuidanceMR surface displayed on live x-ray at 3 frames per secondHelical ablation catheter in left upper PV

Rhode et al TMI 2003 and 2005; Razavi et al, Lancet 2003, Guy’s Hospital, London

Motion modelling in lung radiotherapy

Blackall et al WCOMP 2003, McClelland et al SPIE Med Imag 2005, ESTRO 2005

Movement of point landmarks in motion model

• Tumour• Oesophagus• • • Nodes

MAGI system in the Operating Room:

Overlay of 3D preoperative image dataon stereo field of view of binocular operating microscope

(Edwards et al IEEE-Trans Med Imag 2000)

Edwards et al, Imperial College London

13

Robotically Assisted Lung BiopsyUnder CT Fluoroscopy

Kevin Cleary et al Georgetown University, Washington

Robotically Assisted Lung Radiotherapy Using Optical

tracking (Accuray’s Synchrony)

The Future of Medicine?

• Preventive medicine– Including environment

• Personalised Medicine – Using the –omics (genomics/ proteomics …)

• Keyhole/ robotic surgery– Implanted devices

• Nanotechnology– Biolab– MEMS (micro electro-mechanical systems)– Lab-on-a-chip micro-arrays and diagnosis– Drug production

• Complex Systems– Mathematical biology– Modelling and simulation

Summary

• Extra-ordinary progress• But not quite Star-Trek yet

• But then- we still can’t fly faster than light.

Many thanks to collaborators

The End?

14