You are on page 1of 1

Computer Aided Engineering in Surgery KU Leuven

Reading Assignment 1 Restricted Surface Matching Date: March 29th 17


Name: Phin Yuan Ting Student ID: R0648529

Questions
1. How are the model set and the dataset referred to in this paper?
The model set is referring to the therapeutic object (part of patients anatomy to be
operated on). While the dataset is referred to virtual objects (the patients image)
obtained from computed tomography (CT), MRI or ultrasound.
The mathematical transformation between the coordinate systems of these two
modalities is termed registration.

2. What is the overall improvement that is aimed at?


Although current medical imaging and surgical tool tracking have high level of
accuracy, a small error in the transformation between different modalities can lead
to large inaccuracies in the registration and CAS system.
The aim is to find a robust and highly accurate registration technique. This paper
combines techniques from existing paired-points matching and surface matching to
create a new method called restricted surface-matching (RSM), where a disparity
function f(T) is introduced.

3. What is the "disparity function" in this case (describe in words) and which
innovation is claimed?
It is a measure to indicate how much the surface points deviate from the surface.
Large values of f(T) indicate a strong deviation of surface points, and thus poor
transformation T.
It also restricts the optimization algorithm to sticking to the vicinity of initial
transformation by introducing a penalty function w(di). This enables poor
transformations T to be discarded very effectively, thus minimizing computing
requirement.
The innovation involves a new heuristic search algorithm, where a hybrid approach
that combines evolution strategy (genetic algorithm) and a local minimum search
algorithm is used. It is an adaptive generator for possible transformations that solve
the registration problem.
With this new algorithm, no significant deviation from reference transformation was
found in over 100,000 calculations. Furthermore, no re-calculations were required,
and the transformation time was reduced by a factor of three.

You might also like