Classification of Dynamic Breast MR Image Data - Biomedical Signal Analysis: Contemporary Methods and Applications - page 278

MR imaging

MRI was performed with a 1.5 T system (Magnetom Vision, Siemens,
Erlangen, Germany) equipped with a dedicated surface coil to enable si-
multaneous imaging of both breasts. The patients were placed in a prone
position. First, transversal images were acquired with a STIR (short
TI inversion recovery) sequence (TR=5600 ms, TE=60 ms, FA=90 ◦ ,
IT=150 ms, matrix size 256

256 pixels, slice thickness 4 mm). Then a
dynamic T1 weighted gradient echo sequence (3-D fast, low, angle-shot
sequence) was performed (TR=12 ms, TE=5 ms, FA=25 ◦ ) in transver-
sal slice orientation with a matrix size of 256

×

× 256 pixels and an effective

slice thickness of 4 mm.
The dynamic study consisted of six measurements with an interval
of 83 sec. The first frame was acquired before injection of paramag-
netic contrast agent (gadopentatate dimeglumine, 0.1 mmol/kg body
weight; Magnevist TM , Schering, Berlin, Germany) and immediately fol-
lowed by the five other measurements. Rigid image registration by the
AIR method [288] as a preprocessing step was used. As this did not
correct for nonlinear deformations, only data sets without relevant mo-
tion artifacts were included. The initial localization of suspicious breast
lesions was performed by computing difference images (i.e., subtract-
ing the image data of the first acquisition from the fourth acquisition).
As a preprocessing step to clustering, each raw gray-level time series
S ( τ ) ,τ

was transformed into a pixel time course (PTC)
of relative signal reduction x ( τ ) for each voxel, the precontrast scan at
τ = 1 serving as reference. Based on this implicit normalization, no
significant effect of magnetic field inhomogeneities on the segmentation
results was observed.

∈{ 1 , ··· , 6 }

Data clustering

The employed classifier (the minimal free energy vector quantization
neural network) is according to grouping image pixels together based on
the similarity of their intensity profiles in time (i.e., their time courses).
Let n denote the number of subsequent scans in a dynamic contrast-
enhanced breast MRI study, and let K be the number of pixels in
each scan. μ ∈{ 1 , ··· ,K } , that is, the sequence of signal values

x μ (1) , , x μ ( n ) , can be interpreted as a vector x μ ( i ) R n in the

{ ··· } ∈

n -dimensional feature of possible PTCs at each pixel.