DESCRIPTION:

BronchoPose dataset: a bronchoscopy navigation synthetic dataset based on real anatomies to enable fair comparison among methods. The presented database is composed by trajectories across airways extracted from Computerized Tomography (CT) scans. The anatomy of airways was extracted using a self-developed segmentation method. The trajectories of the database include the navigation path (spatial coordinates), the camera pose, and the simulated images of the virtual bronchoscopy. Virtual airways models are simulated using our own platform developed in C++ and VTK. From the virtual models, bronchoscope trajectories are simulated from the trachea entrance up until the 4 levels and cover the upper-right (ur), lower-right (br), upper-left (ul), and lower-left (bl) lobes. Trajectories are generated from the central navigation path through the luminal central line traversed using the arch-length parameter. Different increments in this parameter allow the simulation of varying velocities across the path. For each central path, different variations, both, in position and camera orientation are generated. Finally, paths with neighboring variations are randomly combined along the navigation arc-length parameter in order to simulate realistic trajectories. In total, our dataset has 876 trajectories per patient and lobe, amounting to a total of 842,712 frames.

CONTENT:

The dataset contains 2 folders by previously decompressing 4 .zip files:

1.Segmentations:

Contains the airway masks (with 1 as airway) of 6 different patients (Lens3.mat, P18.mat, P20.mat, P21.mat, P25.mat, P30.mat) in matlab volumes. load in Python: import scipy.io mat = scipy.io.loadmat('P18.mat') vol = mat['DistalGlobalTh'] vol.shape = (401, 401, 693) #X,Y,Z

1. VirtualNavigations:

Contains 6 folders, one for each patient:

Lens3_INSP_SIN --> Lens3 LENS_P18_14_01_2016_INSP_CPAP --> P18 LENS_P20_14_01_2016_INSP_CPAP --> P20 LENS_P21_04_02_2016_INSP_CPAP --> P21 LENS_P25_25_02_2016_INSP_CPAP --> P25 LENS_P30_14_04_2016_INSP_CPAP --> P30

2.1 Each patient folder:

For each patient there are 4 central navigation paths and their variations defined in a .csv file with all navigations metadatas generated with VTK. For each central camera navigation position, 875 variations (images) are generated combining position and rotation changes.
Folders are named with lung lobule part Upper left (ul), UpperRight (ur), BottomLeft (bl), bottomRight (br) with all the generated png images.

2.2 CSV files metadata content:

.csv file with all navigations metadatas generated with VTK

2.2.1 CSV fields:

patient --> patient id lobe --> path lung lobule: Upper left (ul), UpperRight (ur), BottomLeft (bl), bottomRight (br) level --> this field is incorrect, don't use. pos_x,pos_y,pos_z --> current camera position up_x,up_y,up_z --> camera up vector shift_x,shift_y,shift_z --> camera position shift from central navigation. To get central position (pos_x,pos_y,pos_z)-(shift_x,shift_y,shift_z). q_w,q_x,q_y,q_z --> current quaternion using GetQuaternion() from VTK qw_base,qx_base,qy_base,qz_base --> quaternion of the camera position from its central navigation. Rx,Ry,Rz --> current rotation of X,Y and Z axis. Rx_base,Ry_base,Rz_base --> rotation of X,Y and Z axis of the camera position from its central navigation. filename --> current generated png image base_filename --> png image of the camera position from its central navigation.

2.2.2 CSV clarifications:

1.Central navigation path from different lobules (in this case ul) can be extracted by selecting csv rows that satisfaies the following restrictions:

If lobe=='ul' and (shift_x,shift_y,shift_z)=(0,0,0) and (q_w,q_x,q_y,q_z) = (qw_base,qx_base,qy_base,qz_base) and filename=base_filename then this is a central camera postion whithout variations.

2.Filename nomenclature:

[patient id][lobe][level][SkelPoint][#PoseVariation].png

Example: LENS_P18_14_01_2016_INSP_CPAP_br_0_360_0.png patient id : LENS_P18_14_01_2016_INSP_CPAP (same name has the patient folder) lobe : br level : 0 (not useful, don't use) SkelPoint : 360 (skel point extracted from a skeletonitzation of the volumen segementation).

PoseVariation : from 0 to 875 different pose (camera position and orientation) variations for each camera central navigation.

     If #PoseVariation=0 then this is an image of central camera postion whithout variations.

1. Bronchial Tree Objects & Camera Pose Alignment

3.1 Six Obj files of the 6 corresponding patients. 3.2 Readme file: This document explains how to use the provided .obj files (bronchial tree geometry) and .csv files (camera trajectories) to reconstruct the original synthetic views, without requiring access to the original rendering code.

Cite this dataset:

Borrego-Carazo, J., Sanchez, C., Castells-Rufas, D., Carrabina, J., & Gil, D. (2023). BronchoPose: an analysis of data and model configuration for vision-based bronchoscopy pose estimation. Computer Methods and Programs in Biomedicine, 228, 107241. doi: https://doi.org/10.1016/j.cmpb.2022.107241

Paper code available: https://github.com/BCJuan/BronchoTrack