This guide is still under construction

Fitting¶

Dosma supports cpu-parallelizable quantitative fitting based on scipy.optimize.curve_fit.

To perform generic fitting to any array-like object using an arbitrary model function, we can use dosma.utils.fits.curve_fit(). For example, we can use this function to fit an array to a monoexponential model y = a * exp(b * x) using a maximum of 4 workers:

>>> from dosma.utils.fits import curve_fit, monoexponential
>>> curve_fit(monoexponential, x, y, num_workers=4)

Recently, quantitative MRI (qMRI) has enabled computing voxel-wise relaxation parameter maps (e.g. T₂, T₁\({\rho}\), etc.). We can fit a monoexponential model for each voxel across these registered_images, where tc0 is the initial guess for parameter \(\frac{1}{b}\) in the monoexponential model:

>>> from dosma.utils.fits import MonoExponentialFit
>>> tc0 = 30.0
>>> echo_times = np.asarray([10.0, 20.0, 50.0])
>>> fitter = MonoExponentialFit(echo_times, images, tc0=tc0, num_workers=4)
>>> quant_map, r2_map = fitter.fit()

Custom model functions can also be provided and used with curve_fit. Note that for fitting quantitative parameters using exponential models, exponential parameters are the multiplicative inverse of what is defined by default in these models. You can account for this by passing the inverse as arguments. The command below is equivalent to the fitter above:

>>> curve_fit(monoexponential, 1/echo_times, [x.volume for images.volume], p0=(1.0, 1/tc0), num_workers=4)