Module radiology_ai.preprocessing
Preprocessing functions, Transforms
and Pytorch Dataset
allowing to treat DICOM files and improve CT scan interpretations for Deep Neural Networks
Preprocessing
Raw images could be used by a deep learning model, however we can benefit from preprocessing. On this project the main three preprocess transformation are:
- KneeLocalizer: Center image only on region of interest (ROI) which normalize dataset.
- BackgroundPreprocess: Cleaning background and noisy parts which helps avoid model memorization due to characteristic background/noise.
- CLAHE Transform: Grayscale/Color histogram normalization which helps CNN model training.
See below some examples these transformations:
Expand source code Browse git
"""
Preprocessing functions, `Transforms` and `Pytorch Dataset` allowing to treat DICOM files and improve CT scan interpretations for Deep Neural Networks
.. include:: README.md
"""
Sub-modules
radiology_ai.preprocessing.dicom
-
Preprocessings used for DICOM treatments
radiology_ai.preprocessing.misc
-
Miscellaneous preprocessings
radiology_ai.preprocessing.transforms
-
Fastai Transformations to be used on training