Understanding IVIM Theory

Intravoxel Incoherent Motion (IVIM) imaging separates tissue diffusion from microvascular perfusion using diffusion-weighted MRI.

The Core Concept

In a single imaging voxel, water molecules move in two different ways:

1. True diffusion: Random Brownian motion in tissue
2. Pseudo-diffusion: Motion in blood within capillaries

These create distinct signal decay patterns at different b-values.

The Two-compartment model

Tissue Compartment

Water molecules undergo true diffusion:

• Random thermal motion
• Constrained by cell membranes
• Relatively slow (D ≈ 1 × 10⁻³ mm²/s)

Vascular Compartment

Blood water undergoes pseudo-diffusion:

• Flow through randomly oriented capillaries
• Appears like fast diffusion
• Much faster (D* ≈ 10-20 × 10⁻³ mm²/s)

Signal Model

Bi-Exponential Equation

The IVIM signal model is:

\[ \frac{S(b)}{S_0} = f \cdot e^{-b \cdot D^*} + (1-f) \cdot e^{-b \cdot D} \]

Where:

Parameter	Description	Typical Value
S(b)	Signal at b-value	measured
S₀	Signal at b=0	measured
f	Perfusion fraction	0.05-0.30
D*	Pseudo-diffusion coefficient	5-100 × 10⁻³ mm²/s
D	Tissue diffusion coefficient	0.5-2.0 × 10⁻³ mm²/s

Signal Behavior

At different b-values:

Parameter Interpretation

Diffusion Coefficient (D)

D reflects true tissue diffusion:

• Measures water mobility in tissue
• Influenced by cellularity, viscosity, membrane permeability
• Not affected by blood flow

Clinical meaning:

• Low D: High cellularity (tumors), restricted diffusion (stroke)
• High D: Low cellularity, necrosis, edema

Pseudo-Diffusion Coefficient (D*)

D reflects microvascular blood flow*:

• Related to blood velocity and capillary geometry
• Much faster than tissue diffusion
• Large variability (10-100 × 10⁻³ mm²/s)

Why it's "pseudo-diffusion":

• Blood flow in randomly oriented capillaries
• Appears as diffusion-like signal decay
• But mechanism is flow, not thermal motion

Perfusion Fraction (f)

f represents the vascular signal fraction:

\[ f = \frac{V_{blood}}{V_{blood} + V_{tissue}} \cdot \frac{T_{2,blood}}{T_{2,tissue}} \]

• Approximately equals blood volume fraction
• Weighted by T2 differences
• Typical values: 5-30% depending on tissue

Mathematical Details

Relationship to Conventional DWI

At high b-values (b > 200 s/mm²), perfusion contribution is negligible:

\[ \frac{S(b)}{S_0} \approx (1-f) \cdot e^{-b \cdot D} \]

This is why:

• ADC (apparent diffusion coefficient) at high b ≈ D
• But ADC at low b is contaminated by perfusion

Segmented Fitting

A practical fitting approach:

1. Step 1: Fit D from high b-values only

Using b > 200 s/mm²: $$ \ln(S/S_0) = -b \cdot D + \ln(1-f) $$

1. Step 2: Fix D, fit f and D* from low b-values

Using all b-values with D fixed: $$ S/S_0 = f \cdot e^{-b \cdot D^*} + (1-f) \cdot e^{-b \cdot D} $$

Full Fitting

Simultaneously fit all three parameters:

• More accurate when SNR is high
• Requires good initial estimates
• May have convergence issues

B-Value Selection

Optimal Sampling

For reliable IVIM:

B-value range	Purpose	Recommended values
0	Reference signal	0
10-50	Perfusion sensitivity	10, 20, 30, 40, 50
100-200	Transition region	100, 150, 200
300-800	Pure diffusion	400, 600, 800

Minimum: 4 b-values (but limited reliability) Recommended: 8-10 b-values

Why Multiple Low B-Values?

The perfusion effect decays rapidly:

• At b = 50: ~60% of perfusion signal remains
• At b = 100: ~37% remains
• At b = 200: ~14% remains

More low b-values → better D* estimation

Physical Constraints

D* > D Constraint

Physically, D* must be greater than D:

• Blood flow is faster than tissue diffusion
• If D* < D, model assumptions are violated
• osipy enforces this constraint

Parameter Bounds

Physiologically motivated bounds:

D:   [0.1, 5.0] × 10⁻³ mm²/s
D*:  [2.0, 100.0] × 10⁻³ mm²/s
f:   [0.0, 0.7]

Simplified Models

Mono-Exponential (No IVIM)

Ignores perfusion:

\[ S(b)/S_0 = e^{-b \cdot ADC} \]

• ADC = D when b is high
• ADC > D when low b-values included (perfusion contamination)

Simplified IVIM

When D* >> D, can use:

\[ S(b)/S_0 \approx (1-f) \cdot e^{-b \cdot D} + f \cdot \delta(b=0) \]

Where the perfusion term is essentially a step function.

Applications

Oncology

IVIM parameters can indicate:

• Tumor cellularity (low D = high cellularity)
• Tumor vascularity (high f = more vessels)
• Treatment response (changes in D, f)

Liver

Particularly useful because:

• Dual blood supply (portal + arterial)
• High perfusion fraction
• Non-invasive assessment of fibrosis

Brain

Challenging due to:

• Low f (small blood volume)
• White matter has very low perfusion
• But useful for stroke, tumors

Limitations

SNR Requirements

IVIM fitting is noise-sensitive:

• f and D* have high uncertainty
• D is less sensitive to noise (estimated from high b-values only)
• Requires multiple averages

Motion Sensitivity

DWI is motion-sensitive:

• Breathing affects abdominal IVIM
• Cardiac pulsation affects brain
• Motion correction recommended

Model Assumptions

The bi-exponential model assumes:

• Two distinct compartments
• No exchange between compartments
• Mono-exponential decay in each

These may be violated in:

• Pathological tissues
• Tissues with multiple compartments
• Very short or long b-value ranges

Relationship to Other Techniques

IVIM vs ASL

Both measure perfusion without contrast:

Aspect	IVIM	ASL
Measures	f, D*	CBF directly
Units	fraction, mm²/s	ml/100g/min
Technique	Diffusion-weighted	Spin labeling
Best for	Liver, tumors	Brain

IVIM vs DCE

Aspect	IVIM	DCE
Contrast	None	Gadolinium
Parameters	D, D*, f	Ktrans, ve, vp
Perfusion info	Indirect (f)	Direct
Safety	No contrast risk	Contrast concerns

References

1. Le Bihan D et al. "Separation of diffusion and perfusion in intravoxel incoherent motion MR imaging." Radiology 1988.

2. Koh DM et al. "Intravoxel incoherent motion in body diffusion-weighted MRI: Reality and challenges." AJR 2011.

3. Federau C. "Intravoxel incoherent motion MRI as a means to measure in vivo perfusion: A review of the evidence." NMR Biomed 2017;30(11):e3780.

See Also

• IVIM Tutorial
• How to Handle Fitting Failures