Discussion - Robotized Transcranial Magnetic Stimulation - page 149

9.1 Robust Real-Time Robot/World Calibration
First of all, we have presented a new method for performing the calibration
between robot and tracking system in a robust online fashion ( Chap. 4 ). It uses a
marker attached to the robot's third link. With the transform from this marker to
the fourth joint, estimated beforehand, we can calculate the system's calibration
using one measurement of the tracking system and the forward calculation from
the robot's base to its fourth joint.
Our experimental results have shown that this calibration is suitable for the use
in the robotized TMS system. The mean calibration error is 1.36 mm. It is not as
accurate as the currently used QR24 algorithm [ 6 ] (mean calibration error of
0.88 mm), but more accurate than the standard hand-eye calibration method
proposed by Tsai and Lenz [ 13 , 14 ] (mean calibration error of 1.94 mm), when
evaluated on a grid of different tracking system positions ( Sect. 4.3.2.1 ). On the
other hand, we have found that this online calibration method features the lowest
error distributions when we perform calibration in different regions of the robot's
workspace ( Sect. 4.3.2.2 ). Apart from this, we also found that the calibration
method using the additional marker is accurate (mean error of 0.16 mm) and stable
(mean variation of 0.34 mm), see Sect. 4.3.1 .
For the robotized TMS application many transforms with possible errors are
combined for the final coil positioning by the robot (cf. Sect. 1.3 ) . These are:
• a Computed Tomography (CT)- or Magnetic Resonance Imaging (MRI)-scan
with the calculated head contour,
• a tracked headband with passive marker spheres,
• registration between headband and virtual head based on measurements with a
pointer,
• coil calibration also performed with a pointer, and
• the robot/camera calibration.
Consequently, we have measured the impact of the different robot calibrations on
the overall accuracy of the TMS application.
With our test within a realistic robotized TMS application ( Sect. 4.3.2.3 ) , we
have shown that the presented online calibration method is sufficient and ade-
quately precise for use in the robotized TMS system. Neuro-navigated TMS
systems (without robot) are state-of-the-art in TMS research (see Sect. 1.2 ). The
accuracy of these systems is in the range of 5-6 mm [ 12 ]. Therefore, the accuracy
of the robotized TMS system using the online calibration approach features more
than twice the accuracy (2.21 mm) of the navigated TMS systems. Only the
application of the QR24 algorithm for hand-eye calibration provides results that
are approximately 0.4 mm more precise. However, this might degenerate during
application. In contrast, online calibration has the advantage of maintaining the
accuracy throughout application as the calibration can be updated and checked
during application.