Quantitative evaluation of automated robot-assisted volumetric breast ultrasound

Adding volumetric ultrasound (3DUS) to MRI improves cancer detection rate in a breast. However, the fusion of 3DUS and MRI is challenging due to the different position (supine versus prone, respectively) and high deformation of a breast in the existing 3DUS scanners. In this study, we present and quantitatively evaluate a novel robot-assisted breast scanning system for 3D US acquisitions. Since the breast is scanned in a prone position with minimal and measurable deformation, this can facilitate MRI - 3DUS fusion. For quantitative evaluation, a breast-shaped rigid polyvinyl-alcohol phantom was constructed containing spherical lesions (15 mm diameter) with different echogenicity (-22 dB, -5 dB, -4 dB, 3 dB, 7 dB). First, the phantom was scanned with a Siemens Skyra 3T MRI (Siemens Healthcare, Erlangen, Germany). Next, the phantom was scanned by a robotic arm (KUKA, Augsburg, Germany) (Fig. 1a.) following a pre-planned spiral trajectory, determined from the MRI volume. The flange of the robot was equipped with an L10-5v US transducer attached to a P500 system (Siemens, Mountain View, CA, US). 2D B-mode US data were acquired for 3 minutes at 12 fps. The imaging depth was 5 cm, and the focal depth was set at 2 cm. A volume of 77 × 77 × 69 mm3 was reconstructed with an isotropic sampling distance of 0.2 mm utilizing a voxel nearest neighbor method with a subsequent 'hole filling' step, i.e. interpolation. The contrast-to-noise ratio (CNR) and signal-to-noise ratio (SNR) were calculated per lesion and compared to reference values from the originally acquired 2D B-mode images. The 3D US volume was registered to the MRI volume by using 4 lesions as landmark, using rigid registration. The distance between the centers of the remaining lesion in MRI and 3DUS after the registration was calculated. The measurement was repeated for 5 combinations between lesions and landmarks. The average distance was used as a measure of registration accuracy. For the reconstructed volume only 24 % of the data were obtained by interpolation. On average, the CNR and SNR were 24% and 6% higher, respectively, for the 3DUS compared to the reference. The registration accuracy was 3.4 mm; hence, the presented scanning approach enables ultrasound breast 3D imaging in prone position facilitating MRI - 3D US fusion.