junqiangchen/LUNA16-Lung-Nodule-Analysis-2016-Challenge — reverse-engineered prompt

Build me a Python project for the LUNA16 lung nodule challenge that can take CT scan data and help find possible lung nodules, then reduce false positives.

I want the workflow to feel practical, with scripts for preparing the dataset, making lung nodule masks from the annotations, checking CT slice and window information, creating 2D and 3D training patches, and saving image paths and labels into CSV files. It should also support candidate extraction, data augmentation, and splitting into train and test sets.

For the model side, include a 3D VNet style nodule detection pipeline and a 3D ResVGG or ResNet style classifier for false positive reduction. Add train and predict scripts for both parts, and save useful outputs like loss, accuracy, ROC, confusion matrix, metrics, and example segmentation results.

Use Python with TensorFlow, NumPy, SimpleITK, OpenCV, pandas, and scikit learn. Keep the code organized and runnable, with clear comments so I can follow the full preprocessing, training, and prediction flow.