Evaluation scripts for MICCAI 2022 MELA Challenge: Mediastinal Lesion Analysis.
MELA-Challenge/
requirements.txt Required packages for evaluation
MELA/
evaluation_mediastinum.py Functions for model evaluation
Run the following in command line to install the required packages. First create a specific Anaconda environment and activate it:
conda create -n mela python=3.7
conda activate mela
And then install required packages using pip:
pip install -r requirements.txt
You can download the competition data by first Join the challenge then visit the Dataset page.
We use this script to evaluate your test set submission online. You can evaluate your own prediction locally as well. The evaluation script has very specific requirements on the submission format. Please make sure that these requirements are followed or your submission won't be graded.
To evaluate your prediction locally, you need to prepare the ground-truth and prediction csv file. Take validation dataset as an example. After the train/validation data is downloaded, you should unzip it and find the following ground-truth csv file:
mela-val-gt.csv
Your prediction should be organized to a .csv file:
mela-val-pred.csv
The prediction info .csv should have eight columns: public_id
(patient ID), coordX
, coordY
, coordZ
(prediction coordinates marking the bounding boxes of lesions), x_length
, y_length
, z_length
(length of the predicted bounding boxes) and probability
(detection confidence), e.g.:
public_id | coordX | coordY | coordZ | x_length | y_length | z_length | probability |
---|---|---|---|---|---|---|---|
mela_0881 | 15.1 | 12.54 | 54.22 | 21 | 22 | 20.2 | 0.75 |
mela_0881 | 115.1 | 152.54 | 85.16 | 10 | 32.1 | 15.3 | 0.26 |
mela_0882 | 24.31 | 10.05 | 100.1 | 21 | 22 | 20.2 | 0.66 |
mela_0882 | 105.32 | 52.94 | 55.16 | 16 | 19.88 | 14 | 0.35 |
... | |||||||
mela_1100 | 111 | 85.2 | 65.55 | 20 | 35 | 37.25 | 0.55 |
mela_1100 | 52.25 | 73 | 88.26 | 18 | 20.56 | 40.25 | 0.27 |
Each row in the prediction .csv represents one predicted bounding box of lesion area. The public_id should be in the same format as in the provided .nii file names.
After setting all of the above, you can evaluate your prediction through the following command line:
python -m MELA/evaluation_mediastinum --gt_dir <ground_truth_csv> --pred_dir <prediction_csv>