A 3D human head dataset for non-coplanar keypoints detection

This MRI volumes ("nii.gz" format) were part of the IXI dataset (IXI Dataset – Brain Development (brain-development.org)). For our work, we focused on the T1-weighted images, which provided an higher degree of anatomical details. We chose and manually annotated 4 non-coplanar points inside these volumes, in order to train a 3D CNN to detect them. Such points could be exploited to perform alignment tasks. The annotation was carried out using 3D Slicer, and we encourage to use it for the visualization of the volumes and the relative annotations ("json" format). Below a description of the 4 keypoints:

Keypoint 1: The cerebral aqueduct in correspondence of the transverse plane slice where the mammillary bodies are two well defined little balls.

Keypoint 2: The point of contact between the two ventricles anterior horns before going into lateral ventricles (visualize on coronal plane)

Keypoint 3: The right eye center in correspondence of the largest diameter circle (on the coronal plane)

Keypoint 4: The left eye center in correspondence of the largest diameter circle (on the coronal plane)

The dataset consists of 507 volumes with related annotations.