176 / 1082
S163
ESTRO 36 2017
_______________________________________________________________________________________________
Conclusion
Acute toxicity following single dose IMRT-based
preoperative APBI appears to be mild. Clinically feasible
optimal and acceptable OAR constraints were developed
for future studies. Long-term follow-up is required for
validation and the skin dosimetry constraints, in
particular, may need additional modification based on
long-term outcomes. Chronic toxicity will be available at
the 36
th
ESTRO conference.
[1] Olivotto et al. 2013 [2] van der Leij et al. 2014
OC-0317 MR radiomics and fractal dimension in
cervical cancer predicting pathological complete
response
N. Dinapoli
1
, A.L. Valentini
2
, A. Pesce
1
, R . Gatta
1
, C.
Masciocchi
1
, V. Ninivaggi
2
, B. Gui
2
, G. Matt ana
1
, M.A.
Gambacorta
1
, R. Autorino
1
, M. Campitelli
1
, A. Testa
3
, G.
Chiloiro
1
, J. Lenkowicz
1
, C. Casà
1
, G. Scambia
3
, L.
Bonomo
2
, V. Valentini
1
1
Università Cattolica del Sacro Cuore -Policlinico A.
Gemelli, Radiation Oncology Department , Rome, Italy
2
Università Cattolica del Sacro Cuore -Policlinico A.
Gemelli, Radiology Department, Rome, Italy
3
Università Cattolica del Sacro Cuore -Policlinico A.
Gemelli, Gynecology Department, Rome, Italy
Purpose or Objective
Standard treatment in cervical cancer (CC) is largely based
on chemoradiation (CRT) for locally advanced stages. The
role of surgery after a first time of CRT is limited to
selected protocols. In such cases it is possible to assess the
pathological complete response (PCR). A fractal is an
object having geometric shape that can be divided into
subparts, each of which is a reduced copy of the whole.
This property, known as self-similarity, can be measured
by the Fractal Dimension (FD). Aim of this study is finding
an enhanced by FD radiomics signature detecting patients
with PCR.
Material and Methods
Pathologically proven CC patients underwent to CRT after
MR definition of the local stage. T2 High-Resolution images
were used for delineating GTV. A home-made software
was used for assessing radiomics features after pre-
processing images by Laplacian of Gaussian (LoG) filter
and tuning its σ parameter, and for extracting fractal
dimension (FD) on raw images. FD was extracted at
different threshold levels considering all deciles between
10% and 100% of signal intervals inside the GTV volume
analyzed slice by slice by using the box-counting
technique. Stepwise logistic regression (SLR) models were
used for predicting PCR using FIGO Stage, 1
st
order
features (skewness, kurtosis, entropy) and different
thresholds for FD. AUC of ROC and calibration, were
determined for internal validation.
Results
177 patients were retrospectively recruited (FIGO stage Ib
– IVb). FD was the only significant predictor for the whole
dataset in SLR (MaxDF@10%: P-Value<0.001, MinDF@0-
90%: P-Value<0.05, see fig. 1). AUC of ROC was 0.70 with
not significant deviance at calibration. After selecting
patients according to the MR pixel spacing (PS) the most
numerous group was found in patients having PS=0.58mm
(69 patients). A new SLR model showed significance of
entropy (σ=2.36mm, P-Value<0.05), kurtosis (σ=1.23mm,
P-Value<0.01) and MinDF@0-90% (P-Value<0.01), with AUC