Attached code for the paper MARVEL: MR Fingerprinting with Additional micRoVascular Estimates using bidirectional LSTMs
This repository contains attached code to our MICCAI 2024 paper MARVEL (preprint and published versions). Please do not hesitate to contact us in case of any trouble dealing with our code.
For network training, we used a Python 3.8 virtual environment with dependencies listed in the network_requirements.txt.
For matching, we used a Python 3.11 virtual environment with dependencies listed in the match_requirements.txt.
To train the network, use train_BiLSTM.py. The code requires the following functions that you need to implement depending on your data pipeline:
-
load_dictionnary(path): function that loads your dictionary and returns two numpy arrays, one with its parameter values of shape(n_signals, n_parameters)and the second with its signals of shape(n_signals, n_pulses) -
load_vessl_distrib(path): function that loads your vessel distributions and returns two numpy arrays, one with its vascular parameter values of shape(n_distributions, n_parameters)and the second with its frequency df distribution coefficients of shape(n_distributions, n_df) -
convol_base_dico(DICO_base_params, DICO_base_signals, vasc_params, vasc_distribs): function that convolves the base dictionary along the df axis to produce the vascular dictionary. Returns two numpy arrays, one with its parameter values of shape(n_signals, n_parameters)and the second with its signals of shape(n_signals, n_pulses)
If you use our code, please cite :
Barrier, A., Coudert, T., Delphin, A., Lemasson, B., Christen, T. (2024). MARVEL: MR Fingerprinting with Additional micRoVascular Estimates Using Bidirectional LSTMs. In: Linguraru, M.G., et al. Medical Image Computing and Computer Assisted Intervention – MICCAI 2024. MICCAI 2024. Lecture Notes in Computer Science, vol 15002. Springer, Cham. https://doi.org/10.1007/978-3-031-72069-7_25