# Get started with ACDC

## ACDC : from raw data to 128x128 tensor

Each patient has a 4d (sucession of 3d volumes) image file.
This file is at NII format.


Monai or torchio offer method to open such files.

Moreover, two instant are saved for each patient : End-systole, End-diastole.

1 patient =(4d file, 3D ES file, 3D ED file, info.cfg)
For each diastol and sistol 3D volumes, we have segmentation mask for 4 classes (background, Myocradium, Left Ventricule, Right Ventricule).
info.cfg gives metadata.


For the first patient, at a precise instant, for a precise slice :

In order to load efficently a precise patient, we use ``get_scan_and_mask``.
Then for each slide, we identify bouding-boxes (by underlying non-zero area in each slice).
Finaly we resize each mask and 3d-volume to a unique size (H=128, W=128)

The function get_dataset load and preprocess each patient.
Then we save in a dataset module each ES/ED slice as an element of the dataset (annotated with the corresponding mask, a metadata about condition of the patient and an other for ES or ED)


## Use ACDCDataModule