3 Summary iW-Net 9 The authors proposed an end-to-enddeep learning scheme shown in the figure below. In the paper, they describe a significant improvement of the challenging segmentation of small and non-solid nodules and include tables and additional figures not reported here. This network and the idea of using the physics-inspired weight map as both a feature map and a loss function component represent an important step towards solving both aforementioned issues of user interaction and inter-observer variability. We could definitely extend such an approach to several other segmentation tasks and hopefully we’ll keep seeing similar successful attempts to build a powerful communication between Deep Learning models and radiologists. Finally, it’s worth mentioning that due to reduction in size of the network compared to state-of-the-art methods, the authors manage to get a more robust batch normalization, and their average inference time is down to 0.12 ± 0.08(s). Beyond these points, the strongest impact of iW-Net is its ability to manually correct the segmentations without external algorithms. Conclusion "We could definitely extend such an approach to several other segmentation tasks" "The strongest impact of iW-Net is its ability to manually correct the segmentations without external algorithms."