Segmentation of lung cancer cells in 3D CT Scans

Work done in a team of 2 during the Institut Polytechnique de Paris Medical Data (IPPMeD) hackathon at the Paris Saclay Cancer Cluster (PSCC). A presentation of the project can be found here and a full report here. The code is available on this repo.

We build a 3D U-Net model using Monai framework to segment lung cancer as a classification task. We use some preprocessing (cropping) and postprocessing (smoothing) techniques to improve the model performance. A sliding window approach is used to limit the size of the model and increase the accuracy.

We obtain a F1 score (Dice Metric) of 0.5 on our test set and 0.37 on the hackathon test set. We achieve the 2nd best score on the hackathon.

Image extracted from Monai documentation