This challenge was a difficult detection problem because of the respiratory and circulatory systems network, leading to potential confusion due to high contrast images.

Our approach was to implement a pre-processing function that handles original images and annotations, isolating the lungs from the rest of the body using Hounsfield Units (HU) using a 3D U-NET segmentation algorithm and .postprocessing This detection allowed us to automatically reject artifacts detected outside the lungs.

This step was followed by a nodule detection model, based on the « Retina U-Net 3D » architecture  https://www.nature.com/articles/s41591-019-0447-x , that accepted the preprocessed image input and predict the detected nodules by generating the coordinates of a rectangular parallelepipedal box, but also intensity attributes and segmentation results.

Finally, we used a classification model (SVM support vector machine after feature extraction), to classify the patient as pathological having at least one nodule greater than 100 mm3 or normal.

Thanks to IBM team we used 3 IBM Power AC922 servers with Nvidia V100 32 Go  graphic cards (GPU – Graphics Processing Unit with respectively 4, 4 et 6, for a total of 14 GPUs)

https://www.ibm.com/us-en/marketplace/power-systems-ac922 .

That processing power together with the capacity to handle large models allowed to process successfully the test images in the required time with the best result of the ten teams in the challenge.