Luna 16 competition : false positive reduction ( project report : computer-aided diagnosis in medical imaging )

We propose a method for automatic false-positive reduction of a list of candidate nodules, extracted from lung CT-scans, using a convolutional neural network. Batch normalisation was applied to reduce overfitting. Data augmentation on the positive set of candidates was used to balance the training set. Training and testing was performed on the LUNA16 competition data set. The resulting AUC score was 0.9944 on average per fold in 10-fold cross-validation.