Neuroharmony: A tool for harmonizing volumetric MRI data from unseen scanners

The model presented in Garcia-Dias, et al. (2020).

Documentation

neuroharmony.readthedocs.io

Install Neuroharmony.

pip install neuroharmony

Example of use:

Pre-trained Neuroharmony model

An example plot of how to load and apply pre-trained a Neuroharmony model.

import matplotlib.pyplot as plt
from neuroharmony.models.harmonization import fetch_trained_model, fetch_sample
import seaborn as sns

X = fetch_sample()
neuroharmony = fetch_trained_model()
x_harmonized = neuroharmony.transform(X)

rois = ['Left-Hippocampus',
        'lh_bankssts_volume',
        'lh_posteriorcingulate_volume',
        'lh_superiorfrontal_volume',
        'rh_frontalpole_volume',
        'rh_parsopercularis_volume',
        'rh_parstriangularis_volume',
        'rh_superiorfrontal_volume',
        'Right-Cerebellum-White-Matter',
        ]
fig, axes = plt.subplots(3, 3, figsize=(10, 10))
for roi, ax in zip(rois, axes.flatten()):
    ax.plot(neuroharmony.kde_data_[roi]['x'], neuroharmony.kde_data_[roi]['y'],
            color='#fcb85b', ls='--', label='ComBat harmonized training set')
    sns.kdeplot(X[roi], color='#f47376', ls=':', legend=False, ax=ax, label='Original test set')
    sns.kdeplot(x_harmonized[roi], color='#00bcab', ls='-', legend=False, ax=ax, label='Harmonized test set')
    ax.set_xlabel(roi, fontsize=13)
axes.flatten()[2].legend(ncol=3, bbox_to_anchor=(0.8, 1.175), fontsize=13)
axes.flatten()[3].set_ylabel('Density', fontsize=15)
plt.subplots_adjust(left=0.07, right=0.99,
                    bottom=0.05, top=0.96,
                    hspace=0.20, wspace=0.20)
plt.savefig('test.png', dpi=200)
plt.show()