Correlating Clinical Scores with Anatomical Electrodes Locations for Assessing Deep Brain Stimulation.

Full text

Page 1

Correlating clinical scores with anatomical electrodes
locations for assessing deep brain stimulation
Florent Lalys
1,2,3, Claire Haegelen
1,2,3,4, Alexandre Abadie
1,2,3,
Pierre Jannin
1,2,3
1
INSERM, U746, Faculty of Medicine CS 34317, F-35043 Rennes, France
2
INRIA, VisAGeS Unit/Project, F-35042 Rennes, France
3
University of Rennes I, CNRS, UMR 6074, IRISA, F-35042 Rennes, France
4
Department of Neurosurgery, Pontchaillou University Hospital, F-35043 Rennes, France
Abstract. Movement disorders in patients with Parkinson's disease may require
functional surgery, when medical therapy isn’t effective. In Deep Brain
Stimulation (DBS), electrodes are implanted within the brain to stimulate deep
structures such as SubThalamic Nucleus (STN). This paper describes
successive steps for constructing digital atlases gathering patient’s location of
electrode contacts and clinical scores. Three motor and three neuro-
psychological scores were integrated in the study. Correlations between active
contacts localization and clinical data were carried out using an adapted
Hierarchical Ascendant Classification and have enabled the extraction of
clusters aiming to suggest optimum sites for therapeutic STN DBS. The
postero-superior region has been found to be effective for motor score
improvement whereas the antero-inferior region revealed noticeable neuro-
psychological scores deterioration. Comparison with existing results has shown
that such atlases are very promising for understanding phenomena better.

Keywords: Deep Brain Stimulation, digital atlases, subthalamic nucleus,
Parkinson disease
1 Introduction
Parkinson's disease (PD) is recognized as one of the most common neurological
disorders, affecting 1% of people older than 60 years. The pathology is an age-related
deterioration of certain nerve systems, which affects the patient's movement, balance,
and muscle control. Major symptoms are indeed characterized by abnormalities of
motor functions, several of which predominate, but all do not necessarily occur in all
individuals. While these symptoms related to PD can be treated with medication
therapy, few patients have side effects or treatments for which it is not effective. In
inserm-00617006, version 1 - 25 Aug 2011
Author manuscript, published in "IPCAI 2011, Berlin : Germany (2011)"
DOI : 10.1007/978-3-642-21504-9_11

Page 2

such cases, Deep Brain Stimulation (DBS) [1] might be necessary, according to strict
patient inclusion criteria. The target for DBS in Parkinson’s disease has been moved
from the ventro-lateral thalamic nucleus (Vlc) to the medial global pallidus (Gpm)
and the Sub-Thalamic Nucleus (STN). Among these three deep brain structures, the
STN became the mean target of high-frequency DBS in patients with PD and severe
disabled symptoms [2]. Even if STN DBS has proved its efficiency for motor
symptoms improvement, several studies have reported adverse-events after DBS
surgery affecting cognitive functions, emotion or behavior [3,4,5]. It remains
interrogations about contacts location that provide the largest motor improvement
while producing the least clinical side effects.
Preoperative definition of the target is based on the relief of symptoms and results
of previous implantations only. Its precise localization from patient images is
unfeasible mainly due to contrast and spatial resolutions limitations. Consequently,
the surgeon needs additional information and knowledge for indirect identification of
such small-targeted structures. Some are explicit such as anatomical atlases adapted to
the patient images thanks to image registration. Some are implicit knowledge, such as
the one acquired from learning, literature, and previous surgical cases. Different
strategies have recently studied methods for making explicit this implicit information
and knowledge. The concept of probabilistic functional atlases has been introduced
[6]. After a step of normalisation within a common space, effective contacts are
linked to a large panel of parameters acquired during the various stages of the
procedure. The spatial normalisation allows performing retrospective studies, where
statistical techniques are used to study anatomical or functional variability between
patients. Response to stimulation, electro-physiological recordings, and clinical scores
related to motor or cognitive evolution are all possible information that can be
integrated in such atlases [7,8,9,10]. This fusion of information permits to understand
functional organization within deep brain structures that helps for the identification of
the optimal zone of therapy.
As far as we know, no functional atlases were proposed yet for representing the
relationships between the anatomy and a large panel of clinical scores. While most of
these atlases are using a single motor score for modelling the global outcome of the
patient, we proposed in this paper to extend this research by adding motor and
neuropsychological scores. We thus introduce the concept of anatomo-clinical atlases
and describe the methods for their computation.
2 Materials and Methods
In this study, the objective was to correlate anatomical position of electrode contacts
with clinical outcomes in STN DBS. After an automatic segmentation of electrode
contacts for each patient, the crucial step was to perform an accurate patient-to-
template registration, allowing expressing contacts coordinates in a common
reference space. Then the integration of clinical scores obtained pre and post-
operatively permitted the extraction of representative anatomo-clinical clusters. We
describe in this section every stage of our procedure, from the extraction of electrode's
contacts to the creation of atlases.
inserm-00617006, version 1 - 25 Aug 2011

Page 3

2.1. Data
The study population consisted of 23 patients (10 women and 13 men, mean age 60
years old) with idiopathic PD who had bilateral STN DBS according to selected
inclusion criteria at the University Hospital of Rennes (France). Patients had pre-
operative 3-T T1-weighted MR (1 mm x 1 mm x 1 mm, Philips Medical Systems), pre
and post-operative CT scans (0.44 mm x 0.44 mm x 0.6 mm in post-operative
acquisitions and 0.5 mm x 0.5 mm x 0.6 mm in pre-operative acquisitions, GE
Healthcare VCT 64). All images were denoised with the Non-local means algorithm
[11], and a bias correction algorithm [12] based on intensity values was also applied
on T1 MR images.
2.2. Contact localization and registration workflow
Electrode contacts coordinates have to be first extracted from patient's post-operative
images, and then warped into a common space to perform a retrospective study for
population comparison. The spatial coordinates of each electrode contact were
determined based on black artefacts from CT images, corresponding to the center of
the actual contact [13]. The segmentation process included an intensity threshold, a
search for the artefacts, and constraints on neighborhood. An additional step of
extrapolation was necessary when just two or three contacts were identified from the
four (due to the resolution of the CT scan).
We then used, as a common space, a mono-subject high-resolution 3T MRI brain
template, constructed and assessed in [14]. The registration workflow was described
and validated within the validation part of the template in the context of DBS in [15]:
the post-operative CT was first linearly registered to the pre-operative MRI. An affine
MR-to-template transformation was then computed, followed by a local affine
registration computed on a region of interest within deep brain structures. With this
procedure the contacts positions could be precisely warp into our template.
2.3. Clinical scores
From numerous clinical scores assessed in our population, we studied UPDRS III
(Unified PD Rating Scale [16], part III, taking into account akinesia, tremor, rigidity)
widely used to assess bilateral motor improvement of patients, the Schwab & England
[17] scale, and the Hoehn & Yahr [18] scale. We also studied 3 neuro-psychological
scores that are often assessed through the STROOP score [19] that tests the global
cognitive efficiency, and the verbal fluency tests [20] (both categorical and phonemic)
that determine the ease with which patients can produce words. For each score,
patients were tested without medication just before surgery (DBS OFF) and three
months after it under stimulation (DBS ON). We took the difference between DBS
ON and DBS OFF to compute a value that represents for all scores the degree of
improvement or worsening of the patient.
inserm-00617006, version 1 - 25 Aug 2011

Page 4

2.4. Atlases
Atlases were computed as the 3D visualization of active contacts that are represented
by a color code related to the patient clinical scores values. Due to errors or missing in
report scoring, 15 patients were incorporated in the motor studies, and 22 in the
neuro-psychological ones. Left and right contacts are shown on atlases. The x-axis
represents the left-right direction, the y-axis represents the antero-posterior direction,
and the z-axis represents the caudo-cranial direction.
2.5. Creation of clusters
After atlas construction, we performed a segmentation step with the help of non-
supervised techniques. The Hierarchical Ascendant Classification (HAC) was used on
clinical scores and coordinates to search homogeneous groups of patients. Feature
vectors are composed of 4 features: the value of the x-axis, y-axis, z-axis, and the
score value:
Score) z, y,(x,=X
. The HAC operates by successively merging pairs
of existing clusters, where the next pair of clusters to be merged is chosen as the pair
with the smallest distance. This linkage between clusters a and b was performed with
the Ward criterion along with the weighted Euclidean distance:

ba
M
=k
bkakk
ba
n+n
)xx(w
nn= b)(a,d
∑
−
1
2
(1)
where
∑
=i
a
n
ai
a
a
x
n
=x
1
1
is the centroid of cluster a (resp. b),
a n (resp.
b n ) is the
number of objects in cluster a (resp. b), and
1
1
=w=w=w
32
and
k
w are the weights, specified by
124
3
4=w
. The dendrogram was cut in order to obtain two
or three clusters for each clinical score. Validation of the non-supervised
classification was performed with an ANOVA test. Different clusters were computed
in order to correlate clinical scores and anatomical location of electrode contacts.
Compare to clustering on clinical scores only, adding the coordinates permits better
definition of clusters and better appreciation of the efficiency of DBS according to
contacts locations.
3 Results
For all following figures, we superimposed an ellipse to improve clinical
representation and interpretation. This ellipse was extracted from the segmentation of
the STN by an expert, and has sensibly the same dimension of a standard STN (10 x 6
inserm-00617006, version 1 - 25 Aug 2011

Page 5

x 3 mm). For each clinical score, analyses were made for both hemispheres, but all
clusters found included same patients. This assumes a symmetry of the contact
coordinates between the left and the right sides of patients.





Fig. 1. UPDRS III analysis, with the dendrogram (left hemisphere), the cluster display
and the ANOVA validation. Coordinates are defined in mm in the template reference
space. Euclidean distances are between centroïds of clusters.

Fig. 1. shows an improvement of UPDRS III in the postero-superior region (cluster 3
in red including 2 patients), whereas the worst scores are situating mainly outside the
STN and in the antero-inferior region (
p
6
10−
=
). H&Y and S&E scores showed
5.10−
=p
).
similar results, but with a fuzzy definition of clusters (

From Fig. 2., we can see a deterioration of the categorical fluency in the posterior
region (green), and an improvement in the antero region (blue) (
phonemic fluency we found a general deterioration for all patients, without apparent
separation of clusters. The analysis of the STROOP score indicated score
improvement in the postero-superior region, and deterioration in the antero-inferior
region (
10−
=p
).

2
4
10−
=p
). For the
5
inserm-00617006, version 1 - 25 Aug 2011