Deep Learning for enhanced 3D landmark positioning: addressing biases in traditional registration methods

In this study, we developed a 2D DL architecture to automate the placement of 10 selected landmarks on 463 mouse skulls. These landmarks were chosen for their diverse locations and definitions, and, for each of them, the predicted landmark positions was compared to predictions of deformable registration approaches such as ALPACA. Euclidean distances to ground truth and individual shape conservation were compared between the two approaches. Images .tif files and associated manual landmarks are provided with this metadata file, along with DL model weights and R codes used to perform the statistics in the associated publication.