csa-atrophy

Evaluate the sensitivity of atrophy detection with SCT. The algorithm works as follows:

• Consider subject I --> sI
• Applies a rescaling on the native image (e.g. 1, 0.95, 0.8) --> rX
• Applies random affine transformation --> tY
• Segment the cord
• Compute CSA --> CSA(sI, rX, tY)

How to run

This code has been tested using Python 3.7.

Download (or git clone) this repository:

git clone https://github.com/sct-pipeline/csa-atrophy.git
cd csa-atrophy

Installation: csa-atrophy requires specific python packages for computing statistics and processing images. If not already present on the computer's python environment such packages will automatically be installed by running pip command:

pip install -e .

Download the Spine Generic Multi-Subject dataset.

Edit the file config_sct_run_batch.yml according to your setup. Notable flags include:

• path_data: If you downloaded the spine-generic data at another location, make sure to update the path;
• include_list: If you only want to run the script in a few subjects, list them here. Example: include_list: ['sub-unf04', 'sub-unf05']

See sct_run_batch -h to look at the available options.

Run the analysis:

sct_run_batch -config config_sct_run_batch.yml

:note: desired subjects using flag -include and in parallel processing using flag -jobs.

To output statistics, run in Dataset

csa_rescale_stat -i csa_atrophy_results/results -o csa_atrophy_results -config config_script.yml -fig

Quality Control

After running the analysis, check your Quality Control (QC) report by opening the file qc/index.html. Use the “Search” feature of the QC report to quickly jump to segmentations or labeling results. If you spot issues (wrong labeling), add their filenames in the 'config_correction.yml' file (see https://spine-generic.rtfd.io/en/latest/analysis-pipeline.html for further indications). Then, manually create labels in the cord at the level of inter-vertebral discs C1-C2, C2-C3, ..., C4-C5 with the command:

manual_correction -config config_correction.yml -path-in csa_atrophy_results/data_processed -path-out PATH_DATA

The bash script outputs all manual labelings to the derivatives directory in the dataset path defined in path_data. It is now possible to re-run the whole process. With the command below labeling will use the manual corrections that are present in the derivatives/ folder of the dataset, otherwise labeling will be done automatically.

sct_run_batch -config config_sct_run_batch.yml

Statistics

After everything is done, compute stats: Per-subject stat: Panda dataframe df_sub:

• intra-subject MEAN: MEAN[CSA(sI, rX, :)] --> MEAN_intra(sI, rX): df_sub['mean']
• intra-subject STD: STD[CSA(sI, rX, :)] --> STD_intra(sI, rX): df_sub['std']
• intra-subject COV: STD[CSA(sI, rX, :)] / MEAN[CSA(sI, rX, :)] --> COV_intra(sI, rX): df_sub['cov']
• rescale_estimated_subject MEAN: MEAN[ CSA(sI, rX, :) ] / MEAN[ CSA(sI, 1, :) ] --> MEAN_rescale_estimated_subject(sI, rX): df_sub['rescale_estimated']
• intra-subject error MEAN: MEAN[CSA(sI, rX, :)] - (rX^2 * MEAN[CSA(sI, 1, :)]) --> MEAN_error_intra(sI, rX): df_sub['error']
• intra-subject error in percentage MEAN: [MEAN[CSA(sI, rX, :)] - (rX^2 * MEAN[CSA(sI, 1, :)])] / MEAN[CSA(sI, rX, :)] --> MEAN_error_intra_perc(sI, rX): df_sub['perc_error']
• intra-subject difference MEAN: [MEAN[CSA(sI, 1, :)] - (rX^2 * MEAN[CSA(sI, rX, :)])] --> MEAN_diff(sI, rX): df_sub['diff']

Across-subject stats: Panda dataframe df_rescale

• intra-subject STD: MEAN[STD_intra(:, rX)] --> STD_intra(rX): df_rescale['std_intra']
• intra-subject COV: MEAN[COV_intra_sub(:, rX)] --> COV_intra(rX): df_rescale['cov']
• inter-subject STD: STD[MEAN_intra(:, rX)] --> STD_inter(rX): df_rescale['std_inter']
• rescale_estimated (across subjects) MEAN: MEAN[MEAN_rescale_estimated_subject(:, rX)] --> MEAN_rescale_estimated(rX): df_rescale['mean_rescale_estimated']
• rescale_estimated (across subjects) STD: STD[MEAN_rescale_estimated_subject(:, rX)] --> STD_rescale_estimated(rX): df_rescale['std_rescale_estimated']
• error in percentage (across subjects) MEAN: MEAN[MEAN_error_intra(:, rX)]
• error in percentage (across subjects) STD: STD[MEAN_error_intra(:, rX)]
• intra-subject difference STD: STD[MEAN_diff(:, rX)]

Power analysis:

• sample size for a between subject study: [(z(uncertainty) + z(power))^2 * (2 * STD[MEAN(:, rX)]^2)] / [MEAN[CSA(sI, 1, :)] - MEAN[CSA(sI, rX, :)]]
• sample size for a within subject study: [(z(uncertainty) + z(power))^2 * (STD[MEAN_diff(:, rX)]^2)] / [MEAN[CSA(sI, 1, :)] - MEAN[CSA(sI, rX, :)]]

Plot results:

• STD_intersub
• Mean and STD inter-subject error percentage in function of rescaling
• sample size: minimum number of patients to detect an atrophy of X with Y% power and Z% uncertainty
• CSA values boxplot in function of rescaling
• Error values boxplot in function of rescaling