NiChart Brain Ageing Scores

Overview

NiChart_BAScores includes the official implementation of NiChart's DeepSPARE-BA, our brain age regression model based on a ResNet-18. Users can train or perform inference on their own data using our CLI.

Weights can be found in the releases section.

The user can either use the tool directly in NiChart or use the CLI by first installing it:

Note

python 3.9 and higher required!

pip install -e .

Training

To train a model, you first need a directory that contains train, test and eval folders with niftii LPS oriented and DLICV aligned scans (see NiChart_DLMUSE if you want the full pipeline for DLICV alignment and flirt for linear registration) and a csv file that maps the MRID's to the target value. Then, a simple example of training a resnet18 on your data is by simply doing the following:

NiChart_BAScores train \
                 --in_dir [input directory] \
                 --out_dir [output directory] \
                 --mode [mode] \
                 --label_dict [path to .csv] \
                 --model_type [single, pairwise] \
                 --model [model name(for example: resnet18)] \
                 --target [target name(same as in label dict)] \

Note

Note that the niftii files must have the following suffix corresponding that they are preprocessed as it should: _T1_LPS_dlicv_aligned.nii.gz otherwise, the preprocessing pipeline must be applied first

Evaluation

Evaluation is performed by default as the last step in training, but, the user has the option to select any weight and model and evaluate it on a dataset. A simple example is the following:

NiChart_BAScores evaluate \
                 --in_dir [input directory] \
                 --mode [mode] \
                 --model [model name(for example: resnet18)] \
                 --model_type [single, pairwise] \
                 --model_weights [model_weights(.pth)] \
                 --label_dict [path to .csv] \
                 --target [target name(same as in label dict)] \
                 --plot_path [image name(.png)]

Note that in in_dir, there should exist a eval folder.

Inference

Last but not least, the user has the option to perform inference using any pretrained model:

NiChart_BAScores inference \
                 --in_dir [input directory] \
                 --out_dir [output directory] \
                 --csv [full path of the .csv output file] \
                 --model_type [single, pairwise] \
                 --model [model name(for example: resnet18)] \
                 --mode [mode]
                 --model_weights [model weights name(.pth)] \

Note that now in in_dir there should be only raw niftii files and not folders!

Note

You can see all the details about the arguments of the CLI in the tutorial folder.