Annual challenge organised by the french association of radiologists. 2019 edition took place from 14/09 to 13/10, teams were presented with x3 datasets:
- TC images of lungs for cancer nodules detection (classification)
- TC images of the brain for prediction of multiple sclerosis' level (regrassion: score prediction 1-20)
- Calculation of the muscle surface for sarcopenia (segmentation) Present notebooks are a part of lung cancer classification project.
Learning dataset consisted of ~650 3D DICOM images with an typical average resolution of 512x512x350. Images were processed for NN 224x224x3 input to use weights pretrained on the ImageNet. An example of a normalized x3 2D slices stacked to RGB image:
Link to organiser's web-site: https://jfr.radiologie.fr/presidents-word