Population
120 women (mean age = 55 years old (28-88)) who underwent breast MRI between July 18th 2016 and March 31th 2017 in whom an abnormal enhancing lesion was identified with subsequent pathological analysis (n=179: 69 benign,
7 borderline,
103 malignant lesions) were retrospectively and consicutively included,
regarless of MRI indication.
MR acquisition (Fig.1)
Patients were imaged in prone position on a 1.5 T GE Optima MR 450w GEM system using a dedicated 8-channel breast array coil (GE,
Milwaukee,
USA).
T1 and T2-weighted non fat-satured axial sequences were acquired before contrast administration.
Dynamic contrast-enhanced T1-weighted fat-satured gradient-echo sequences (VIBRANT,
GE) were acquired before and four times after bolus injection of Gadolinium chelate.
The ULTRAFAST sequence named DISCO (Differential Subsampling with Cartesian Ordering ; T1W Gradient-echo sequence undersampling k-space) sequence was performed before the first phase of the T1 VIBRANT sequence.
DISCO sequence acquisition began simultaneously to the contrast injection; eleven temporal DISCO ranks were thus acquired during 1 min 18 sec (7.7 sec each rank) in order to assess early lesion enhancement.
Reading protocols (Fig2)
Full standard protocol
The two readers,
blinded to clinical and pathological data,
were asked to classify lesions according to the Bi-rads MR lexicon based on morphology and enhancement characteristics (obtained from curves type 1,2,3).
FAST protocol
A month later,
readers were asked to read the FAST protocol consisted in T2-W,
T1-W and the first T1-W fat-satured VIBRANT after contrast injection.They were ask to classify lesions according to Bi-rads classification,
blinded thought from any dynamic enhancement parameters.
Abbreviated protocols
Readers were asked to determine in which of the leven ranks on the ULTRAFAST sequence the lesion beacame first visible.
Then,
to perform semi-quantitative analysis,
Region of Interrest (ROI) were drawn on the ULTRAFAST sequence in each lesion detected.
Enhancement curves were extracted from AW server sofware by a research engineer (GE,
Milwaukee,
USA).
The following parameters were extracted from curves : Enhancement integral (EI,%) ; Maximal Slope of Increase (MSI,
%/sec) ; Maximim of Enhancement (Rmax,
%) ; Time of Maximum of Enhancement (Rmax Timing,
sec) ; Wash-in Rate (WIR,
sec) ; EA (Enhancement Amplitude,
%) ; Time of Half Rising (THR,
sec)
Statistical analysis
Analysis was performed on statistics sofware MedCalc (Ostend,
Belgium).
Descriptive anlysis was performed using a non-parametric Mann-Whithney test for non-continuous variables.
To determine whether 'continous values' could differentiate malignant from benign lesions,
thresholds were determined by constructing a receiver operating characteristic (ROC) curve.
We calculated odds ratios (OR) for predicting malignancy with 95% confidence intervals and p-values for each of the predictor variables for malignancy.
Quadratic κ coefficients were calculated to assess intraobserver agreements for lesion characterization between the two protocols (FAST and Full protocol),
regardless of dynamic data.
Quadratic Κ coefficients were calculated to assess interobserver agreement for enhancement rank determination on ULTRAFAST acquisition.
Univariate analysis was used to calculate odds ratios for each of the predictor variables for malignancy.
Then,
step by step logistic regression analysis was performed to determine which criteria was significantly associated with malignancy.
A receiver operating (ROC) curve analysis was performed to compare the results of interpretations based on the full gold standard protocol versus the abbreviated protocol.