This study was approved by the institutional review board (IRB) of Sichuan Provincial People’s Hospitaland obtained informed consent from every woman participant. During November 2018 and March 2021, a total of 153 patients were initially scanned with a DWI sequence. The inclusion criteria were (1) placenta previa and suspected PAS disorders based on clinical risk factors or previous US results; (2) singleton pregnancy; (3) fetal development coincides with gestational age. Patients were excluded for the following reasons (1) chronic hypertension, pre-existing renal disease, and diabetes mellitus; (2) inadequate surgical records; (3) suspected placental insufficiency; (4) severe artifacts on MRI images. Finally, a total of 73 patients (mean age 32.21 ± 4.62 years, range 22–45 years) were enrolled (Fig. 1). The average gestational age was 32 weeks (range 26–36 weeks).
MRI protocols
The MRI examination was performed on a 1.5T MR scanner (Aera, Siemens Healthineers, Erlangen, Germany) using a 16-channel body matrix coil. Conventional MR sequences including HASTE, True-FISP, T1WI and DWI were scanned. The image protocols were (1) axial, coronal, and sagittal halffourier acquisition single-shot turbo spin echo (HASTE): field of view (FOV) 420×80mm, 5 mm thick section, 20% gap, matrix 272×320, scan duration 50 s; (2) axial, coronal and sagittal true fast imaging in steady-state precession (TRUFISP): FOV of 420×80 mm, 5 mm thick section, 30% gap, matrix 234×384, and a scan duration of 48s; (3) 3D-volumetric interpolated breath-hold examination (3D-VIBE): FOV 400 mm, 5mm thick section, 20% gap, matrix 180×320, scan duration 8 s; (4) Diffusion weighted imaging: FOV 390mm, 5mm thick section,matrix 192 × 120, parallel imaging acceleration factor 2, b values ranging from 0 to 1600 s/mm2 (b = 0, 50, 100, 150, 200, 400, 600, 800, 1000, 1200, and 1600 s/mm2),scan duration 7 min 29 s.
Image processing and analysis
The ROI delineation and calculation of DWI, DKI, IVIM parameters were performed using research software IMAgenGINE (Vusion Tech) [24].
A monoexponential fit was used to calculate ADC and eADC with b values of 0 and 1000 s/mm2. The following equations were used:
Sb/S0 = exp (-b×ADC),
Sb/S0 = eADC = exp [−(b × ADC)].
where Sb and S0 are the signal intensities in the diffusion gradient factors of b and 0, respectively.
A non-Gaussian model was used to calculate MD and MK with 6 b-values (b = 0, 400, 800, 1000, 1200, and 1600 s/mm2) fitting the following equation [25, 26]:
Sb/S0 = exp (− b×MD + b2×MD2×MK/6),
where Sb and S0 are the signal intensities acquired with the diffusion gradient factors of b and 0, respectively; MD is the mean diffusivity representing the corrected ADC; MK is the diffusion kurtosis.
A biexponential fit was used to compute D, D*, and f with 8 b-values (b = 0, 50, 100, 150, 200, 400, 600, 800 s/mm2) fitting the following equation [27, 28]:
Sb/S0=(1-f) exp (− b×D) + f exp [− b×(D + D*)],
where Sb and S0 are the signal intensities in the diffusion gradient factors of b and 0, respectively; f is the perfusion fraction; D is the diffusion coefficient; D* is the pseudo-diffusion coefficient.
All ROIs were drawn independently by 1 radiologist including as large parts of the placenta as possible. According to the maternity records of patients, all accreta lesions in our patients with invasive placentas were in the lower uterine segment. ROIs of the accreta lesions in invasive placentas (AL) and lower 1/3 part of the placenta in noninvasive placentas (LP) were drawn, respectively. (Fig. 2). The ROIs were automatically replicated to all diffusion parameter maps. The mean, minimum, and maximum ADC (ADC mean, ADC min, and ADC max), eADC (eADC mean, eADC min, and eADC max), MD (MD mean, MD min, and MD max), MK (MK mean, MK min, and MK max), D (D mean, D min, and D max), D* (D* mean, D* min, and D* max), and f (f mean, f min, and f max) values were automatically calculated.
Image analysis
All MRI images were reviewed by 1 radiologists with 8 years of experience in obstetric imaging. The reader was blind to ultrasound diagnosis, surgical and pathological findings.
MRI features of PAS disorders were recorded as presence or absence, including T2 dark bands, placental heterogeneity, abnormal intraplacental vascularity, placental cervical protrusion sign, focal exophytic mass, placental recess, placental bulge, abnormal vascularization of the placental bed,myometrial interruption, bladder tenting, bladder vessel sign and parametrial vessel sign [29–31].
Reference standard
The diagnosis of invasive placentas is made intraoperatively and pathologically according to the FIGO clinical classification[32]. Invasive placentas include placenta increta and placenta percreta. Placenta percreta is diagnosed when the placental tissue invades the uterine serosa and surrounding organs, including the broad ligament, vaginal wall, and bladder visually. Placenta increta is diagnosed when gentle cord traction results in the uterus being pulled inwards without separation of the placenta, with increased vascularity around the placental bed, bluish or purple colouring and distention of the placental bed.
Pathological examination was performed on uterine specimens following hysterectomy or on placental tissue from the removed placenta following conservative treatment. Placenta percreta is diagnosed when hysterectomy specimens showing villous tissue within or breaching the uterine serosa, invading the bladder wall tissue and other pelvic tissues. Placenta increta is diagnosed when hysterectomy specimens or partial myometrial resection of the increta area showing placental villi within the muscular fibers or in the lumen of the deep uterine vasculature.
Statistical analysis
The diffusion parameters of the placenta may change with the increase of gestational weeks. Therefore, all DWI parameters were corrected using the following equation before further statistical analysis [33]:
$${y}_{corrected}=y-\varvec{\beta }{\varvec{X}}_{\varvec{a}\varvec{g}\varvec{e}}=y-\left(\begin{array}{c}{\beta }_{0}\\ {\beta }_{c}\end{array}\right){\varvec{X}}_{\varvec{a}\varvec{g}\varvec{e}}$$
where \({y}_{corrected}\) is the value after correction; \(y\) is the original value; \({\varvec{X}}_{\varvec{a}\varvec{g}\varvec{e}}\) is the gestational week vector; \({\beta }_{0}\) and \({\beta }_{c}\) are the constant and the linear regression coefficient in linear fitting, respectively.
Student t-test, Mann-Whitney U-test, and χ2 test were used to compare the difference of clinical features between patients with invasive and noninvasive placentas. Mann-Whitney U-test was used to compare the difference of DWI parameters between AL in patients with invasive placentas and LP in patients with noninvasive placentas. χ2 test and receiver operating characteristic (ROC) analysis were performed to estimate the discriminative ability of MRI features and to evaluate the diagnostic performance of significant DWI parameters in predicting invasive placentas. Youden index and corresponding sensitivity, specificity, positive and negative likelihood ratios were calculated. Significant DWI parameters and MRI features showing the highest Youden index were included for differentiation. A stepwise multivariate logistic regression analysis was performed to identify independent risk factors of invasive placentas.P values < 0.05 were considered statistically significant. All analyses were performed on a Win 10 desktop with Python 3.6, scipy 1.5.2, and sklearn 0.24.1.