Kinetic Modeling

Edward Di BellaDmitri RiabkovHarshali BalSathya Vijayakumar

Introduction to kinetic modeling

Challenges1. Formulation/selection of physiological model2. Obtaining accurate input functions3. Noise issues - optimal data groupings4. Noise issues - optimal reconstruction, non-linear optimization, use of constraints

Applications

Outline

Examples of tracer kinetics:

•Intravascular (99mTc-albumin, MS-325)

•Extracellular (gadolinium-DTPA)

•Intracellular (99mTc-Sestamibi, 99mTc-Teboroxime)

•More complex uptake and washout characteristics, including changes of state, binding (18FDG)

Introduction

•Area under curve

•Upslope

•Percent Enhancement

Semi-quantitative Models

Two Compartments

Tissue - Interstitial (ve)

Vascular - redblood cells

Tissue - Intracellular

Vascular - plasmacapillary

k

)/exp()( evEFtEFth

E = extraction, F = flow, t =time

=volume of extravascular spaceev

Axial concentration gradient

Vascular - plasma (Cp)Vascular - redblood cells (Hct)

capillary

Tissue

Model with capillary transit time

cec

c

TtvTtEFFE

TtFth

)/)(exp()(

Canine Study (LAD occlusion)

28 beats later FIESTA imageEarly contrast 8 regions

Myocardial Perfusion with Contrast MRI

(a) (b) (c)

(e)

27

(d)Region number

Infarct - volume of distribution changes

Tissue - Intracellular

Vascular - plasma (Cp)

Tissue - ve

Vascular - redblood cells

Normal Infarct

Myocardial viability

MRI and PET Viability

(Upper panels) Non-viable apical region shown with FDG-PET, left, and contrast enhanced MRI, right. (Lower panels) Non-viable inferoposterior region shown with FDG-PET, left, and contrast enhanced MRI, right. Arrows indicate regions of non-viability.

• MRI

– Saturation– Flow effects

Input Function - challenges

• Dynamic PET and SPECT

– Blood binding

– Temporal sampling rate

• Solution: Blind estimation of kinetic parameters

Static SPECT:

Dynamic SPECT:

20-30s 90-100s 490-500s

290-300s

University of Utah

Cluster Analysis for segmentation

Segmentation and formation of parametric imagesfor dynamic cardiac SPECT

• Approach: segment 4D short-axis data with clustering andobtain parametric maps– clustering:

p Cnpn

p

a min

an is vector of values for location n

p is the average curve of membersof pth cluster Cp

Results

Clustered blood input more smoothcompared to manually chosen ROIs

Smoother blood input may result inless variance in parameter estimatesfrom compartment model fitting

Results

Clustered wash-in images Summed short-axis images

201Tl canine study

Teboroxime patient study

SUMMARYKinetic modeling is a very useful approach with many applications

Numerous important and interesting research areas:– Acquisition

– Automated robust processing

– Input function

– Modeling

– Visualization and use of parametric images