Soumya Ghose1, Rakesh Shiradkar1, Jhimli Mitra1, Rajat Thawani1, Mirabela Rusu2, Michael Feldman3, Amar Gupta4, Andrei Purysko5, Lee Ponsky6, and Anant Madabhushi1
1Department of Biomedical Engineering, Case Western Reserve University, Cleveland, OH, United States, 2GE Global Research, 3Perelman School of Medicine, University of Pennsylvania, 4Diagnostic Radiology, Cleveland Clinic Foundation, Cleveland, OH, 5Diagnostic Radiology, Cleveland Clinic Foundation, Cleveland, OH, United States, 6Urology, Case Western Reserve University School of Medicine
Synopsis
In a single center IRB approved retrospective study, statistically significant differences in the shape of the
prostate gland were observed between BCR+ and BCR- populations.
Introduction
Despite advancement in surgical
procedures and radiation therapy, there may be treatment failure or biochemical
recurrence (BCR) of prostate cancer in
estimated 30-35% of the treated
prostate cancer (PCa) patients within 10 years
of definitive therapy1. Patients
developing BCR are at higher risk of mortality due to disease progression. Hence
early prediction of BCR may enable the use of more aggressive or neo-adjuvant therapies and may potentially improve patient outcome. BCR is often associated with aggressive cancers, resulting in an abnormal
bulge in the prostate gland 2.
In many instances extracapsular
(ECap) spread which is found to be
predictive of BCR is associated with an irregular
bulge in the prostate, focal capsular retraction and or
thickening broad capsular tumor contact3. These findings
beg the question whether there are quantitative differences in the shape between
the prostate capsules for patients who will
(BCR+) and will not go on to develop
BCR (BCR-) following definitive
therapy (i.e. surgery or radiation). In
this study we seek to explore this question
by quantitatively analyzing and comparing the shape of prostate capsule as captured on pre-operative T2w MRI
between BCR+ and BCR- patients.Materials and Methods
A single center IRB approved study included 874 patients who were referred
for prostate MRI between 2008 and 2014. The inclusion criteria were as follows: availability of complete image datasets (T1w, T2w and ADC map); no treatment for PCa before
MRI; presence of clinically localized PCa; availability
of Gleason score; and data available
for post-treatment PSA and follow-up for at least 3 years in patients without
BCR. Of the 874 patients in the registry, 77
patients fit these criteria. To reduce
statistical bias, the following rules were employed
for cohort selection: a) equal number of patients
in BCR+ and BCR- groups;
b) similar Gleason scores in
both groups; and c) similar tumor stages in both groups. This reduced the final number of study patients to 50. Of these 50 patients, 25
were in the BCR+ group and 25 were in the BCR- group. The prostate capsule was
manually segmented on T2w images by an experienced genitourinary radiologist with more than 7 years
of experience in reading prostate MRI.
To perform a statistical comparison of prostate gland shapes between
BCR+ and BCR- cases, all prostate MRIs were aligned to a common space. Registration to the representative template was performed in two stages
using an initial affine
registration4, followed
by a non-rigid B-spline5 based registration. Two atlases, A+ and A-, were constructed, one each for the BCR+ and the BCR- groups respectively. Each of A+ and A- were affine registered to bring
both the atlases
to a common space for statistical comparison. All registered prostate capsules of both the BCR+ and BCR-
groups are transformed into a signed distance function using Danielsson
distance map function6. As opposed to the binary representation of a
mask where each voxel within the prostate capsule is 1 and 0 outside, each
voxel determines the distance of a given voxel from the capsule boundary. The
function has positive values for voxels inside the prostate capsule, it
decreases in value as the voxel approaches the boundary where the signed
distance function is zero, and it takes negative values outside of the prostate
capsule. The statistical comparisons of
BCR+ and BCR- is now done using the corresponding signed distance
representations of each registered prostate capsule. A non-parametric General
Linear Model (GLM) based t-test8 with
permutation testing and corrections for multiple comparison testing was used to
find statistically significant prostate capsule
shape differences between
A+ and A-. The effect of ECap spread
on prostate capsule shape was also analyzed.Results
The comparison of A+
and A- atlases revealed statistically significant differences (p<0.05) in the shape
of the prostate capsule.
Statistically significant shape differences in the prostate capsule were observed
between BCR+ without ECap spread and BCR- without ECap spread groups. Statistically significant shape differences in prostate shapes between
BCR+ with ECap spread and BCR- with ECap spread groups was also observed.
These results are presented in Figure 1. Between the A+ and A- atlases, statistically significant shape differences were identified on the anterior, posterior
and the lateral side of the
prostate. Comparing patients with ECap [AM1] spread, statistically
significant differences were observed
primarily towards the anterior, lateral and posterior parts of the prostate. Comparing
patients without ECap spread, statistically significant shape
differences were identified towards the posterior and lateral sides of the prostate.
[AM1]Change extent everywhere to spreadAcknowledgements
No acknowledgement found.References
1. Boorjian, S. A. et al. Long-term risk of clinical
progression after biochemical recurrence following radical
prostatectomy: the
impact of time from surgery to recurrence. Eur. Urol. 59, 893–899 (2011).
2. Koca, O., et
al. The factors predicting biochemical recurrence in patients with radical prostatectomy. Archivio Italiano
di Urologia e Andrologia 87, 270–275
(2016).
3. Davis, R. et al. Accuracy of multiparametric magnetic
resonance imaging for extracapsular extension for prostate cancer
in community practice. Clinical
Genitourinary Cancer (2016).
4. Ourselin, S., et al. Reconstructing a 3d structure
from serial histological sections.
Image and vision computing 19,
25–31 (2001).
5. Rueckert, D. et
al. Nonrigid registration using free-form deformations: application to
breast MR images. Medical Imaging, IEEE Transactions on 18,
712–721 (1999).
6. Danielsson, P.-E. Euclidean distance mapping. Computer Graphics and Image Processing 14, 227 – 248 (1980).
7. Winkler, A. M., Ridgway, G. R., Webster, M. A., Smith, S. M. & Nichols,
T. E. Permutation inference
for the general linear
model. Neuroimage 92, 381–397 (2014).