BRIQA: Balanced Reweighting in Image Quality Assessment of Pediatric Brain MRI

• Competition: LISA 2025 Challenge – Task 1 (MRI Quality Assessment)
• Result: 2nd-place solution
• Authors: Alya Almsouti*, Ainur Khamitova*, Darya Taratynova*, Mohammad Yaqub
  (* equal contribution)
• Paper: TBA

For questions, reach out at darya.taratynova@mbzuai.ac.ae

Structure

• qc_pipeline/ – encoder, transforms, utilities, and training/testing modules.
• scripts/ – thin shell wrappers: env.sh, train.sh, test.sh.
• requirements.txt – dependencies needed by the package.

Setup

1. Install deps (Python ≥3.9):

   pip install -r requirements.txt

2. Edit scripts/env.sh to match your data layout, then source it. This sets PYTHONPATH and the env vars read by the helpers:

   source scripts/env.sh

Training

Run the helper after overriding any variables you need:

CLASS_NAME=Noise \
BATCH_STRATEGY=rotating \
OUTPUT_DIR=./results/noise_rot \
scripts/train.sh

The script calls python -m qc_pipeline.training with the env-driven arguments.

Testing

Generate predictions with the best checkpoint:

TEST_DATA_DIR=/path/to/test/images \
WEIGHT_DIR=./results/noise_rot \
scripts/test.sh

Outputs land in ${TEST_OUTPUT_DIR} (defaults to the weight directory).

Direct Python Usage

Alternatively execute the modules yourself:

python -m qc_pipeline.training --help
python -m qc_pipeline.testing --help