SDNet

The code in this repository implements SDNet, a model-driven FOD reconstruction network, for further details see the accompanying paper at https://arxiv.org/pdf/2307.15273.pdf. The code in this repo is currently being updated to improve usability.

Abstract

Fibre orientation distribution (FOD) reconstruction using deep learning has the potential to produce accurate FODs from a reduced number of diffusion-weighted images (DWIs), decreasing total imaging time. Diffusion acquisition invariant representations of the DWI signals are typically used as input to these methods to ensure that they can be applied flexibly to data with different b-vectors and b-values; however, this means the network cannot condition its output directly on the DWI signal. In this work, we propose a spherical deconvolution network, a model-driven deep learning FOD reconstruction architecture, that ensures intermediate and output FODs produced by the network are consistent with the input DWI signals. Furthermore, we implement a fixel classification penalty within our loss function, encouraging the network to produce FODs that can subsequently be segmented into the correct number of fixels and improve downstream fixel-based analysis. Our results show that the model-based deep learning architecture achieves competitive performance compared to a state-of-the-art FOD super-resolution network, FOD-Net. Moreover, we show that the fixel classification penalty can be tuned to offer improved performance with respect to metrics that rely on accurately segmented of FODs.

Figure 1 SDNet achitecture

User Guide

File Structure

The directory the code comes in is named CSDNet_dir, and will have the following structure:

SDNet_dir
    └── data.py
    └── util.py
    └── inference checkpoints
        └── experiment_name
            └── inference
            └── runs
            └── model_saves
    └── models
        └── csdnet
    └── model_app
        ├── inference.py
        └── train.py

The network is designed to be trained on the WU-MINN human connectome project dataset, as a consequence the data directory is designed to be similar to the HCP data when downloaded. The main data directory, named hcp in this case, subjects labeled as per the HCP. The file path of this directory should be specified in options.py under the attribute data_dir. To specify the subjects in data_dir to be used for training, validation and testing adjust the train_subject_list, val_subject_list, and test_subject_list attributes.

.
└── hcp
    └── subject
        └── T1w
            └── T1w_acpc_dc_restore_1.25.nii.gz
            └── 5ttgen.nii.gz
            └── white_matter_mask.nii.gz
            └── Diffusion
                └── bvals
                └── bvecs 
                └── data.nii.gz
                └── nodif_brain_mask.nii.gz
                └── csf_response.txt
                └── gm_response.txt
                └── wm_response.txt
                └── csf.nii.gz
                └── gm.nii.gz
                └── wmfod.nii.gz
                └── gt_fod.nii.gz
                ├──fixel_directory
                    └── afd.nii.gz
                    └── peak_amp.nii.gz
                    └── index.nii.gz
                    └── directions.nii.gz
                    └── fixnet_targets
                        └── gt_threshold_fixels.nii.gz
                └── undersampled_fod
                    └── bvals
                    └── bvecs
                    └── data.nii.gz
                    └── normalised_data.nii.gz
                    └── csf_response.txt
                    └── gm_response.txt
                    └── wm_response.txt
                    └── csf.nii.gz
                    └── gm.nii.gz
                    └── wm.nii.gz
                └── tractseg *
                    └── peaks.nii.gz *
                    └── bundle_segmentations *
                        └── ** bundle_segmnetation_masks **

Network Training

Prior to network training, ensure the file paths have been set up as above and the options.py script contains the desired network configuration. To train the network run train.py.

Network Inference

Prior to using the network for inference, ensure the file paths have been set up as above and the options.py script contains the desired network configuration. To perform inference update the test_subject_list attribute in the options.py script and run the eval.py script.

Network Configuration:

Network configuration is performed by changing the attributes in options.py; documentation is included in the form of comments in this script.