Patients
This retrospective study protocol was approved by the hospital ethics committee [approval number: Medical Lun Shen (2023) No. 99], and written informed consent was waived.
A cohort of 256 women who underwent breast MRI scans at our institution between June 2021 and September 2023 was initially collected. Inclusion criteria were age > 18 years and the completion of two sets of multi-b-value DW images during breast MRI. Exclusion criteria were prior history of radiotherapy, chemotherapy, or surgery for breast lesions (N = 15); absence of surgery-based or needle biopsy-based pathological confirmation of the lesion (N = 22); incomplete breast MRI scan before surgery or within 2 weeks prior to needle biopsy (N = 8); severely poor MR image quality that substantially hindered interpretation due to artifacts (N = 6); and lesions with a diameter of < 5 mm (N = 27). Ultimately, 178 patients met the study criteria; the patient age range was 24 to 78 years (mean age, 49.52 ± 12.1 years). Based on the nature of the breast lesions, patients were categorized into a benign lesion group (N = 73) and a malignant lesion group (N = 105).
MRI data acquisition
A 3.0-T MR scanner (MAGNETOM Skyra, Siemens Healthineers, Erlangen, Germany) equipped with a 16-channel breast coil was used for imaging. The DWI sequence included 14 b-values and was conducted in two groups: conventional single-shot echo planar imaging (SSEPI-DWI) and SSEPI-DWI combined with SMS technology (SMS-SSEPI-DWI), both DWI sequences have identical scan positioning. Due to the value demonstrated in clinical practice by the diffusion of multiple b-values across a multitude of literature, we routinely scan two sets of multi-b-value sequences in our clinical work to assist in clinical diagnosis. Except for differences in repetition time, SMS acceleration factor, and scan time, parameters were largely consistent. Specific parameters are delineated in Table 1. Other scan sequences included T2-weighted imaging fat saturation with repetition time/echo time, 5000 ms/81 ms; field of view, 340 × 340 mm2; matrix size, 384 × 384; slice thickness/gap, 4.0 mm/1.0 mm; flip angle, 120 degrees; and acquisition time, 2 min 35 s. Transect T1-weighted imaging dynamic contrast-enhanced MRI (1 unenhanced and 6 enhanced sequence sets) included repetition time/echo time, 450 ms/1.58 ms; field of view, 340 × 340 mm2; matrix size, 384 × 256; slice thickness, 1 mm; flip angle, 12 degrees; and acquisition time, 7 min 20 s.
Image analysis
Two radiologists (WF and ZJ, with 8 and 15 years of experience in breast MRI diagnosis, respectively) independently analyzed the images. They were blinded to the pathological findings and DWI sequence parameters. After image acquisition, two sets of original DW images were uploaded to Body DiffusionLab (BoDiLab, Chengdu ZhongYing Medical Technology Co., Ltd., Chengdu, China) in the MR Station. Utilizing post-processing software, calculations were performed to derive results for each parameter of the DWI mono-exponential model and the FROC model.
-
(1)
ADC (DWI mono-exponential model) fitting formula:
$${S}_{b}/{S}_{0}= exp(-b\times ADC)$$
ADC: apparent diffusion coefficient, \({S}_{b}\): image signal intensity at b > 0 s/mm2, \(\:{S}_{0}\): image signal intensity at b = 0 s/mm2.
-
(2)
FROC model fitting formula:
$${S}_{b}/{S}_{0}=exp\left[-D{\mu\:}^{2\left(\beta -1\right)}{\left(\gamma {G}_{d}\delta \right)}^{2\beta }\left(\varDelta -\frac{2\beta -1}{2\beta+1}\delta \right)\right]$$
\(\:{S}_{b}/{S}_{0}\): same meaning as in the mono-exponential model; \({G}_{d}\), \(\delta\), and \(\varDelta\): amplitude, pulse width (25.66 ms), and gradient interval (30.13 ms) of the diffusion gradient, respectively [31]; D: diffusion coefficient (\(\mu\)m2/ms); \(\beta :\) intravoxel diffusion heterogeneity (unitless, 0 < \(\:\beta\:\) ≤ 1); and \(\:\mu\:\): spatial constant (\(\:\mu\:\)m). The Levenberg-Marquardt nonlinear fitting algorithm was used to fit the diffusion images of 14 b-values to the FROC model on a voxel-by-voxel basis, thereby generating three parameter maps.
Two readers independently performed whole-tumor VOI delineation on SSEPI-DWI images (b = 1000 s/mm²), using dynamic contrast-enhanced and T2-weighted imaging sequences as references. They manually outlined the volume of interest (VOI) layer by layer and saved these delineations. The software then automatically transferred these VOIs to each parameter map, yielding the calculation results. For the SMS-SSEPI-DWI sequence measurements, the same VOIs saved from SSEPI-DWI were used as the regions of interest, and the results were recorded and calculated. Each reader performed two measurements and averaged the lesion values. The final value for each parameter was derived from the averaged values of both readers.
The two readers independently assessed image artifacts, imaging sharpness, lesion conspicuity, and overall image quality for all images from the two sets of DWI sequences; they also assessed FROC-DWI-derived parameter maps (D map, β map, and µ map) and ADC maps using a 5-point Likert scale. Image artifacts were regarded as motion artifacts, susceptibility artifacts, and geometric distortions (1 = severe, 2 = moderate, 3 = mild, 4 = minimal, 5 = none). Imaging sharpness was determined according to the appearance of breast tissue edges (1 = severe blurring, 2 = moderate blurring, 3 = mild blurring, 4 = minimal blurring, 5 = sharp and no blurring). Lesion conspicuity was assessed based on the contrast between suspicious lesions and surrounding background tissue (1 = none, 2 = minimal, 3 = mild, 4 = moderate, 5 = severe). Overall image quality comprised a comprehensive consideration of image artifacts, imaging sharpness, and lesion conspicuity (1 = insufficient diagnostic, 2 = poor and definitely affecting interpretation, 3 = moderate and potentially affecting interpretation, 4 = good and not affecting interpretation, 5 = excellent) [32]. The average values assessed by the two readers were taken as the final result.
Additionally, regions of interest (ROI) were plotted where the lesions appeared largest in the SSEPI-DWI and SMS-SSEPI-DWI sequences, avoiding blood vessels and necrosis zone. The SNR and CNR for each b-value image in both sequences were calculated separately. The SNR was defined as the ratio of the mean signal intensity (\(\:{S}_{lesion}\)) of the lesion ROI to the standard deviation of the air background (\(\:{\sigma\:}_{Background}\)) [33]
$$\:SNR={S}_{lesion}/{\sigma\:}_{Background}$$
The following formula was used to calculate CNR:
$$\:CNR=\frac{{S}_{lesion}-{S}_{tissue}}{\sqrt{{{\sigma\:}_{lesion}}^{2}+{{\sigma\:}_{tissue}}^{2}}}$$
\(\:{\:\:\:\:\:\:\:\:\:S}_{lesion}\): mean signal intensity of lesion ROI, \(\:{S}_{tissue}\): mean signal intensity of normal breast tissue, \(\:{\sigma\:}_{lesion}\) and \(\:{\sigma\:}_{tissue}\): standard deviations of lesion ROI and normal breast tissue, respectively [34].
Statistical analysis
Statistical analyses were conducted using SPSS (version 23.0; SPSS, Inc., Chicago, IL, USA), MedCalc (version 20.0; MedCalc Software Ltd., Ostend, Belgium), and R (version 4.0.0; http://www.r-project.org/) softwares. Intraclass correlation coefficient (ICC) values, utilized to assess Intra- and inter-reader agreement, were categorized as follows: ICC ≤ 0.20, poor agreement; 0.21–0.40, fair agreement; 0.41–0.60, moderate agreement; 0.61–0.80, good agreement; and 0.81-1.00, excellent agreement [35]. Normality assessments were performed using the Kolmogorov-Smirnov test for all quantitative parameters. Comparative analyses of quantitative clinical data, FROC-DWI-derived parameters, and image quality scores were conducted using either independent-samples t-tests or the Mann-Whitney U test. For data that conform to a normal distribution, an independent samples t-test was applied for analysis. For data that do not conform to a normal distribution, a Mann-Whitney U test was used for analysis. The results were expressed as means ± standard deviations. Correlations between the two sets of DWI-derived parameters were evaluated by Spearman correlation analysis, with correlation coefficients (r) categorized as follows: r ≤ 0.24, little or no correlation; 0.25–0.49, fair correlation; 0.50–0.74, moderate correlation; and 0.75-1.00, good correlation [32]. Consistency between the two groups of FROC-DWI-derived parameters was assessed using Bland-Altman plots. Multivariate logistic regression was conducted to analyze FROC-DWI-derived quantitative parameters from both sets, thereby establishing a prediction model for distinguishing between benign and malignant breast lesions. Nomogram plots were generated based on the results of multivariate logistic regression, and optimal cutoff values were selected using the Youden index. The Delong test was utilized to identify significant differences in the area under the curve (AUC) of each receiver operating characteristic (ROC) curve [36]. Calibration with bootstrapped resampling was used to reduce the overfitting bias. Additionally, the Hosmer-Lemeshow goodness-of-fit test was performed to compare the predicted and actual response probabilities of the nomogram. The decision curve analysis (DCA) was also performed to quantify their clinical net benefits. The threshold for statistical significance was set to P < 0.05.
Add Comment