This retrospective study included patients with type II TIC breast lesions who underwent preoperative MRI and breast lesion resection in the Department of Breast Surgery between January 2019 and July 2022.

The inclusion criteria were 1) patients with breast dynamic enhancement MRI examination in the study hospital; 2) complete medical information; 3) imaging diagnosis of type II TIC breast lesion; 4) no needle biopsy, radiotherapy, or chemotherapy before the MRI examination; and 5) available postoperative pathological examination. The patients with poor imaging quality were excluded, e.g., motion artifacts due to the inability of the patients to remain still or metal artifacts due to chest or upper arm skin tattoos with inks containing minerals and metals.

This study was approved by the ethics committee of Shanghai Fourth People’s Hospital (approval ID: 2021-047-001). The requirement for individual informed consent was waived by the committee due to the retrospective nature of the study.

All examinations were performed using a 3.0 T MR scanner (Prisma, Siemens, Erlangen, Germany) with a standard 18-channel dedicated bilateral breast coil. Breast MR protocol included routine non-contrast T1W, T2-weighted (T2W), diffusion-weighted imaging (DWI), and DCE-T1WI. DWI was acquired using a single-shot echo planar imaging (EPI) sequence on the transverse plane. The b-values were set to 0 or 1,000 s/mm². The apparent diffusion coefficient (ADC) mapping, along with slice selection, phase encoding, and frequency encoding directions, was computed on the workstation. DCE-MRI was acquired using a fast low-angle shot (FLASH) sequence at the transverse plane with the number of excitations set to 1. The MR plain scan was first acquired before contrast injection. Then, the contrast [gadopentetate dimeglumine (Gd-DTPA)] was injected into the cubital vein at 2.5 mL/s, and the residual contrast agent in the tube was flushed with 20 mL of normal saline. Meanwhile, the DCE scanning program was started. The DCE scanning was repeated five times, with each round lasting approximately 1 minute.

The images were transferred to the picture archiving and communication system (PACS; United-Imaging Healthcare, Wuhan, China). Multiple semi-quantitative parameters were collected, including signal intensity of pre-contrast (SIpre), peak signal intensity (SIp), signal intensity of wash-in (SIwi), slope of peak (Sp), slope of wash-in (Swi), time to peak (Tp), time of wash-in (Twi), enhancement rate of peak (ERp), and enhancement rate of wash-in (ERwi). SIpre was the SI before enhancement. SIp was the peak SI in the whole period. Tp was the duration between contrast injection and the SI of the ROI reaching SIp. SIwi was the maximal SI in the wash-in period. Twi was the duration between contrast injection and the SI of the ROI reaching SIwi. Sp, ERp, Swi, and ERwi were calculated according to the following equations.

DCE-T1WI images were transferred to the Siemens post-processing workstation (mean curve: Siemens, Erlangen, Germany) to obtain the TIC. Post-processing was performed by two senior radiologists with >15 years of experience in breast cancer MRI, both blinded to the pathological diagnosis. Round ROIs for TIC were placed independently by the two radiologists, and they were as large as possible and covered the most prominent area of enhancement in the lesion. The size of the ROI had to be >0.3 cm². ROI delineation had to avoid obvious necrosis, hemorrhage, and cystic changes. The software automatically generated the TIC. In case of disagreement, the final ROI selection and radiological diagnosis were made after discussion. Three time points (time point of contrast injection, Twi, and Tp) were selected, and four parameter mappings [wash-in, washout, maximum intensity projection over time (MIPt), and positive enhancement integral (PEI)] of breast perfusion pseudo-color images were generated. The signals of the whole lesion, the peripheral and central parts of the lesion, and the opposite side of the normal breast were documented.

Pathological diagnosis was taken as the reference in this study. The levels of estrogen receptor (ER), progesterone receptor (PR), C-erbB2, Ki67, CK7, SMMS-1, P63, and E-cad index were determined by immunohistochemical examination (IHC) of the resected breast cancer specimens.

Stata 17.0 (Stata Corporation, College Station, TX, USA) was used in this study. The continuous variables were tested for the normality of their distribution using the Shapiro–Wilk test. Normally distributed continuous variables were presented as means ± standard deviations (SDs) and analyzed using Student’s t-test (two groups) or ANOVA (three or more groups). Non-normally distributed continuous variables were presented as medians (lower quartiles, upper quartiles) and analyzed using the Kruskal–Wallis H-test (two groups) or ANOVA (three or more groups).

Univariable and multivariable Cox regression analyses were performed to investigate the association between various DCE-T1WI features and the benign/malignant classification of breast lesions. Variables with a p< 0.05 in the univariable analysis were included in the multivariable analysis and nomogram model development. The performance of the nomogram was validated using sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), accuracy, and area under the receiver operating characteristic curve (AUROC). A decision curve analysis (DCA) was performed. In this study, p-values were retained to three decimal places, and p< 0.05 was considered statistically significant.

Ninety-eight patients with type II TIC lesions were included in this retrospective study. Their median age was 54 years; all patients were female (100%), 25.5% had hypertension, 11.2% had diabetes, and 17.4% had hyperlipidemia. Table 1 shows the clinical and DCE-T1WI characteristics of the patients. Age was significantly different between the benign and malignant groups (40.0 vs. 64.0 years, p< 0.001), as well as boundary enhancement (13.3% vs. 43.4%, p = 0.001) and ADC (1.31 ± 0.14 vs. 0.86 ± 0.18 × 10⁻³, p< 0.001). For semi-quantitative DCE-T1WI parameters, seven of nine features showed significant differences between the benign and malignant groups (all p< 0.05); only SIp and Twi were not significantly different (p = 0.131 and p = 0.135, respectively).

Table 2 shows the clinical and DCE-T1WI features of the patients with different classes. The ERp was lower in grade III lesions compared with grade I–II lesions (median, 1.99 vs. 2.46, p = 0.020), without significant differences for the other parameters. Figure 1 shows a classic case of female patients with grade III invasive ductal carcinoma of the left breast.

Ten features were included in the univariable analyses ( Table 3 ). All peak-related DCE-T1WI features exhibited associations with breast lesion malignancy, with p = 0.031, p< 0.001, p< 0.001, and p = 0.022 for SIp, Tp, Sp, and ERp, respectively. However, for the wash-in-related features, Twi and ERwi were not associated with malignancy (p = 0.238 and p = 0.182), and the type of TIC was not associated either (p = 0.374).

After the univariable analysis, seven DCE-T1WI features, namely, SIpre, SIp, Tp, Sp, ERp, SIwi, and Swi (all with p< 0.05), were included in the multivariable analysis. Only Tp was statistically significant in the multivariable analysis, with an odds ratio of 0.95 [95% confidence interval (CI) of 0.93 to 0.97, p = 0.014], exhibiting a negative association with malignancy ( Figure 2 , Table 2 ).

The nomogram with the selected features is shown in Figure 3 . The sensitivity, specificity, PPV, NPV, and accuracy of the nomogram were 0.827, 0.761, 0.795, 0.796, and 0.796, respectively. The ROC of the nomogram is shown in Figure 4 . The AUROC was 0.862. The DCA is shown in Figure 5 .