[Feature] <Add reasonable initial guesses>

[etpeterson]

Feature description

The IVIM parameters can differ greatly depending on body part. Valid initial guesses are important to successful fitting as they speed up fits and improve final accuracy.

Describe the solution

This could be based on user input and a table of body part to mean value or some other simple and fast algorithm to generate an initial guess.

Describe alternatives

No response

Additional context

No response

Are you working on this?

None

[LSUDOKO]

Can you Assign me this issue

[etpeterson]

Can you Assign me this issue

Done! Thanks

[LSUDOKO]

VIM Parameter Characteristics by Tissue Type :

• Different tissues exhibit characteristic ranges for D (true diffusion), D* (pseudo-diffusion), and f (perfusion fraction)
• Example typical values:
  • Brain: D≈0.7×10⁻³ mm²/s, D*≈7×10⁻³ mm²/s, f≈0.1

Implementation Approach :

• Create a lookup table/dictionary mapping body parts to typical parameter values
• Use these values as initial guesses for curve fitting algorithms
• The forward model (shown in your code) remains unchanged as it's the mathematical model

Helps :

• Better initial guesses lead to:
  • Faster convergence during fitting
  • Reduced risk of getting stuck in local minima
  • More physiologically plausible results
• Particularly important for IVIM where parameters can be correlated

Integration with Existing Code :

• Your forward model ivim_parameters_to_signal stays the same
• The initial guesses would be used when calling fitting routines (like scipy.optimize.curve_fit or torch optimizers)
• The guesses would be passed as the p0 parameter in most fitting functions

What do you think of this approach ?

[etpeterson]

That sounds great. It would be nice to have the initial guess be either numbers as it is now or a string like "Brain" or "Liver" that then gets looked up and replaced by numbers from your lookup table.

[etpeterson]

Hey, just checking if there's a pull request coming soon?

VIM Parameter Characteristics by Tissue Type :

• Different tissues exhibit characteristic ranges for D (true diffusion), D* (pseudo-diffusion), and f (perfusion fraction)
• Example typical values:
  • Brain: D≈0.7×10⁻³ mm²/s, D*≈7×10⁻³ mm²/s, f≈0.1

Implementation Approach :

• Create a lookup table/dictionary mapping body parts to typical parameter values
• Use these values as initial guesses for curve fitting algorithms
• The forward model (shown in your code) remains unchanged as it's the mathematical model

Helps :

• Better initial guesses lead to:
  • Faster convergence during fitting
  • Reduced risk of getting stuck in local minima
  • More physiologically plausible results
• Particularly important for IVIM where parameters can be correlated

Integration with Existing Code :

• Your forward model ivim_parameters_to_signal stays the same
• The initial guesses would be used when calling fitting routines (like scipy.optimize.curve_fit or torch optimizers)
• The guesses would be passed as the p0 parameter in most fitting functions

What do you think of this approach ?

[LSUDOKO]

Yes sir
I am Trying