Metadata-Version: 2.1
Name: lung-cancer-astra
Version: 0.1.5
Summary: 3D CT backbone (I3ResNet-152) feature extraction and low-risk cohort classifier; numpy/torch/pydicom only
Requires-Python: >=3.8
Description-Content-Type: text/markdown
Provides-Extra: ortam
Requires-Dist: torch; extra == "ortam"
Requires-Dist: numpy; extra == "ortam"
Requires-Dist: pydicom; extra == "ortam"

# lung-cancer-astra

Extracts 2048-dimensional features from 3D chest CT series with an I3ResNet-152 image backbone, then trains and evaluates a classifier within the cohort that a risk score labels as low risk.

- Dependencies: only `torch`, `numpy` and `pydicom` (expected to be installed already).
- Model weights are **not** included in the package; they are provided separately.
- The code contains no paths. All settings are read from a single `.env` file.

## Usage

```bash
pip install --no-deps lung-cancer-astra
python -m lung_cancer_astra --step kontrol --env /path/.env   # environment, settings, data check
python -m lung_cancer_astra --step manifest                   # label / score / DICOM list
python -m lung_cancer_astra --step cikar --gpus 0             # feature extraction (resumable)
python -m lung_cancer_astra --step birlestir                  # merge features into one file
python -m lung_cancer_astra --step egit                       # evaluation + final models
```

`--no-deps` keeps pip from touching the installed `torch` / `numpy` / `pydicom`.

## Configuration

See the header of `lung_cancer_astra/config.py` for the `.env` keys (prefix `LCA_`). Relative paths in `.env` are resolved against the folder that contains the `.env` file.

Step names, setting keys and in-code comments are currently in Turkish.
