Py3DFreeHandUS: A PURE PYTHON LIBRARY FOR 3D VOXEL-ARRAY RECONSTRUCTION BY USING 3D FREEHAND ULTRASOUND

INTRODUCTION and AIM 3D Freehand ultrasound (3DUS) is a technique that combines ultrasound images and position + orientation sensors (POS) with the aim to reconstruct large 3D anatomical parts. However, the few existing applications for applying this technique have at least one of the following disadvantages: i) not open-source; ii) only supporting data streams from a limited number of ultrasound or POS devices; iii) they are written in low-level languages such as C++, making rapid development and prototyping more difficult. We developed a pure Python library called Py3DFreeHandUS that solves all the above issues. PATIENTS/MATERIALS and METHODS Py3DFreeHandUS is written in Python 2.7 which has the same level of abstraction as MATLAB® but it is open-source and also suitable for scientific applications since there are available similar libraries. The pure Python implementation ensures the cross-platform aspect of the code without any compilation effort and achieves reasonable computation speed due, despite the heavy vectorization. Its inputs are independent to ultrasound systems and POS devices. The library is able to process data acquired simultaneously by US and POS devices, being input as DICOM and C3D files, respectively. In particular, it allows to perform the following operations: 1) Estimation of the pose of an US image with respect to POS (calibration as in Figure 1, [1]); 2) Estimation of the calibration quality by calculating Distance Accuracy and Reconstruction Precision [2]; 3) Creation of the 3D voxel-array (Figure 2) containing the US images repositioned in 3D space. It is possible to set the voxel-array axes aligned with global reference frame ones or rotate them to “follow” the swapping silhouette and so minimize the voxel-array dimensions. 4) Filling of empty voxels among slices by using Voxel Nearest Neighbour or Weighted Distance algorithm. 5) Exportation of grey values and US images scan silhouette voxel array to VTI file (VTK) for later visualization (e.g. Paraview, MeVisLab) RESULTS The present library is downloadable (https://github.com/u0078867/Py3DFreeHandUS). The calibration quality assessments were 1.9 mm and 3.9 mm for the distance accuracy and reconstruction precision, respectively. In Figure 2, a 3D reconstruction is showed. The average data processing time for each reconstruction was 5.9 min. DISCUSSION and CONCLUSIONS Py3DFreeHandUS implements state-of-the-art procedures for voxel-array reconstruction by using an open-source, rapid-development and high-level language such as Python. This reconstruction provides large 3D data sets which are useful to extract clinical features such as muscle morphology. Calibration quality assessment revealed satisfactory results, and the applied heavy vectorization allows users to speed-up computation.status: publishe