Computer-Aided Diagnosis Parameters of Invasive Carcinoma of No Special Type on 3T MRI: Correlation with Pathologic Immunohistochemical Markers

Abstract

Purpose

To investigate the correlation between computer-aided diagnosis (CAD) parameters in 3-tesla (T) MRI and pathologic immunohistochemical (IHC) markers in invasive carcinoma of no special type (NST).

Materials and Methods

A total of 94 female who were diagnosed with NST carcinoma and underwent 3T MRI using CAD, from January 2018 to April 2019, were included. The relationship between angiovolume, curve peak, and early and late profiles of dynamic enhancement from CAD with pathologic IHC markers and molecular subtypes were retrospectively investigated using Dwass, Steel, Critchlow-Fligner multiple comparison analysis, and univariate binary logistic regression analysis.

Results

In NST carcinoma, a higher angiovolume was observed in tumors of higher nuclear and histologic grades and in lymph node (LN) (+), estrogen receptor (ER) (−), progesterone receptor (PR) (−), human epidermal growth factor 2 (HER2) (+), and Ki−67 (+) tumors. A high rate of delayed washout and a low rate of delayed persistence were observed in Ki−67 (+) tumors. In the binary logistic regression analysis of NST carcinoma, a high angiovolume was significantly associated with a high nuclear and histologic grade, LN (+), ER (−), PR (−), HER2 (+) status, and non-luminal subtypes. A high rate of washout and a low rate of persistence were also significantly correlated with the Ki-67 (+) status.

Conclusion

Angiovolume and delayed washout/persistent rate from CAD parameters in contrast enhanced breast MRI correlated with predictive IHC markers. These results suggest that CAD parameters could be used as clinical prognostic, predictive factors.

INTRODUCTION

Breast MRI has a high sensitivity of approximately 90% for breast cancer and is used as a complementary imaging modality to mammography and ultrasonography in the assessment of breast disease (1). Dynamic contrast-enhanced MRI (DCE-MRI) offers morphological and functional information, with outstanding sensitivity and variable specificity for breast cancer diagnosis (2,3,4). In addition, pharmacokinetic MRI parameters have been investigated for the differentiation of breast cancer subtypes (5). Recently, preoperative breast MRI has been highlighted not only as a diagnostic imaging method but also as a prognostic tool that can indicate different prognostic factors and surgical outcomes in invasive breast cancer (6,7).

Computer-aided diagnosis (CAD) is widely used in breast MRI and provides improved sensitivity and specificity for readers with varying levels of experience by providing automated analysis of enhancement kinetics. By excluding lesions with a low threshold enhancement and reducing the interpretation time by automating the process, the false-positive rate for CAD is reduced by 25% at a 50% threshold and by 50% at a 100% threshold for enhancement compared with that for analysis performed by a radiologist (8).

Invasive breast cancer includes various histologic subtypes and they have a variety of unique morphological characteristics and MR enhancement patterns. Invasive carcinoma of no special type (NST) accounts for majority of breast cancers and has different clinical courses and prognoses according to pathologic immunohistochemical (IHC) markers. After the 2013 St. Gallen meeting, five subtypes (luminal A-like, luminal B-like human epidermal growth factor 2 [HER2] negative, luminal B-like HER2 positive, HER2 overexpression, and triple negative [TN]) were suggested, and the criteria of molecular subtypes was further specified (9). Contemporary treatment plans are increasingly guided by the molecular classification of breast cancer (10). To know the status of IHC markers and molecular subtypes preoperatively is helpful for deciding the treatment plan and predicting the patient prognosis. There were already a few studies which have investigated the correlation between CAD features and IHC markers, they have shown that CAD features could be used as prognostic markers (11,12,13). Differently from previous studies, we analyzed the invasive carcinoma of NST, the most common type of invasive breast cancer for excluding heterogeneity of subtypes.

The purpose of this study was to retrospectively investigate the correlation between CAD parameters in preoperative 3-tesla (3T) MRI with pathologic IHC markers in invasive carcinoma of NST.

MATERIALS AND METHODS

PATIENT POPULATION

The Institutional Review Board approved this study and required neither patient approval nor patient informed consent for the review of their images and records (IRB No. XC17REDI0052).

From January 2018 to April 2019, data from 450 patients who underwent breast MRI were examined. We excluded patients who underwent MRI examinations for benign breast diseases or chemotherapy evaluation and those who underwent MRI post-operation. Then, patients with specific types of invasive breast cancer such as invasive lobular carcinomas (n = 2), invasive cribriform carcinoma (n = 1), invasive medullary carcinoma (n = 1), invasive papillary carcinoma (n = 1) and invasive mucinous carcinoma (n = 1) were excluded for homogeneity of research. Consequently, 94 women (mean age, 58.84 years; range, 39–86 years) diagnosed with NST carcinoma who had undergone 3T breast MRI using the CAD system were retrospectively investigated. Surgical excision and pathologic study were performed in all patients.

DATA ACQUISITION AND ANALYSIS

MRI TECHNIQUE

All MRI examinations were performed using two 3T systems (MAGNETOM Skyra, Siemens Medical Solutions, Erlangen, Germany and Ingenia, Philips Medical Systems, Best, the Netherlands) with a dedicated array breast coil in the axial orientation. Patients were placed in the prone position.

MRI images from the Ingenia scanner used the following sequences: an axial, fat-suppressed, fast spin-echo T2-weighted imaging sequence (repetition time/echo time [TR/TE], 100032/70 ms; flip angle, 90°; 150 slices; field of view [FOV], 320 mm; matrix, 512 × 512; 1-mm slice thickness; no gap) and an axial diffusion-weighted sequence. Pre- and postcontrast dynamic axial T1-weighted three-dimensional, fat-suppressed, fat-spoiled gradient-echo sequences (TR/TE, 4.7/2.0 ms; flip angle, 12°; 1.0-mm slice thickness) were obtained before and at 76, 136, 196, 256, 316, 376, and 436 seconds after gadoterate meglumine injection.

MRI images from the MAGNETOM Skyra scanner were acquired using the following sequences: an axial, turbo spin-echo T2-weighted imaging sequence (TR/TE, 10190/76 ms; flip angle, 148°; 52 slices; FOV, 360 mm; matrix, 384 × 384; 2.5-mm slice thickness) and pre- and postcontrast axial T1-weighted fast low-angle shot three-dimensional, volumetric interpolated breath-hold examination sequences (TR/TE, 4.6/1.7 ms; flip angle, 10°; 0.9-mm slice thickness) obtained before and at 67, 127, 187, 247, and 367 seconds after gadoterate meglumine injection.

A bolus of 0.1 mmoL/kg gadoterate meglumine (Dotarem; Guerbet, Villepinte, France) was intravenously injected by using an MRI-compatible power injector at a rate of 2 mL/s, followed immediately by a 20 mL saline flush.

CAD SYSTEM

The CAD system (CADstream, version 6.0; Confirma, Kirkland, WA, USA) received all T1-weighted images, and breast tumors were automatically segmented into three dimensions. Then, CAD parameters such as angiovolume and tumor location were assessed for each lesion and retrospectively processed to generate kinetic parameters of dynamic contrast enhancement. The CAD systems compared pixel signal intensity values on the precontrast images and postcontrast series. A 50% threshold was set to balance the sensitivity and specificity (8,14). If the pixel value was greater than the threshold, the pixel was shown in color, and a color overlay was rendered to the lesions meeting threshold enhancement criteria according to changes in the pixel values between the early contrast-enhanced and delayed contrast-enhanced series; the results were indicated as follows: persistent type (increased pixel signal intensity of greater than 10% from the early postcontrast series, color coded blue); washout type (decreased pixel signal intensity at the last postcontrast series greater than 10% from the early postcontrast series, color coded red); and plateau type (change in neither direction by more than 10%, color coded yellow). Thus, we used the following CAD parameters: curve peak (the highest pixel signal intensity in the early phase postcontrast series); angiovolume (the total volume of the enhancing lesion); early enhancement profiles (the rates of rapid [> 100%] or medium [50%–100%] enhancing components within a tumor in the early phase postcontrast series); and delayed enhancement profiles (the rates of washout, plateau, and persistent types in the delayed phase postcontrast series) (Fig. 1A, B).

Fig. 1. MR images with a computer-aided detection color overlay map in a 47-year-old female with histologic and nuclear grade 3, Ki-67 positive, estrogen receptor and/or progesterone receptor positive, luminal B subtype invasive carcinoma of no special type in the right breast.

A. Postcontrast T1-weighted image in the early phase shows an irregular and heterogenously enhancing mass in the outer portion of the right breast.

B. Areas in red, yellow, and blue indicate washout delayed enhancement, plateau-delayed enhancement, and persistent delayed enhancement patterns, respectively. Kinetic curve shows a rapid early enhancement and a delayed washout-type curve. The initial peak enhancement value is 282%. With respect to the delayed-phase enhancement, 57% of the mass represents washout, 21% of the mass represents a persistent-type curve, and 21% represents a plateau-type curve.

PATHOLOGIC IMMUNOHISTOCHEMICAL MARKERS

Pathologic IHC markers included nuclear and histologic grade, pathologic tumor size, vascular and lymphatic invasion status, and regional lymph node (LN), estrogen receptor (ER), progesterone receptor (PR), HER2, Ki-67, cytokeratin 5/6 (CK5/6) and epidermal growth factor receptor (EGFR) status. In our assessment, ER or PR (+) status was indicated by stained nuclei in > 1% of cancer cells in 10 high-power fields. The HER2 staining intensity was scored as 0, 1+, 2+, or 3+. Tumors with scores of 3+ were classified as HER2 (+), whereas those with scores of 0 or 1+ were classified as HER2 (−). Tumors with scores of 2+ were further investigated with fluorescence in situ hybridization to determine the HER2 status. For the Ki-67 index status, we used a cutoff value of 15% for classification into negative and positive groups (15). For molecular subtype assessment, samples were classified into four groups according to hormone receptor status (ER and/or PR), HER2 status and Ki-67 index as follows: luminal A, hormone receptor (+), HER2 (−), and low levels of the Ki-67; luminal B, hormone receptor (+) and either HER2 (+) or HER2 (−) with high levels of Ki-67; TN, hormone receptor (−) and HER2 (−); and HER2-overexpression, hormone receptor (−) and HER2 (+). All values of pathologic IHC markers and pathologic tumor size were based on pathological reports from the electronic medical records of our institution.

STATISTICAL ANALYSIS

Dwass, Steel, Critchlow-Fligner multiple comparison analysis was used to compare the means of all pairs of CAD parameters and IHC markers with a nonparametric, pairwise comparison method. Spearman correlation analysis was used to assess correlations between CAD parameters and pathologic IHC markers. The Spearman correlation coefficient ranges from −1 to +1; +1 indicates a perfect positive association of ranks, zero indicates no association between ranks, and −1 indicates a perfect negative association of ranks.

Univariate binary logistic regression analysis was performed to study independent predictors by using dichotomized IHC markers as dependent variables and CAD parameters as covariates. Histologic and nuclear grades were grouped as low (grade I) or high (grade II or III). Molecular subtypes were classified as luminal (luminal A and luminal B) or nonluminal (HER2-overexpression and TN).

All data analyses were performed using SAS (version 9.4, SAS Institute Inc., Cary, NC, USA), and a p value below 0.05 was considered significant.

RESULTS

CAD parameters on 3T-MRI of invasive carcinoma of NST are summarized in Table 1.

Table 1. CAD Parameters of Invasive Carcinoma, NST (NST Carcinoma).

CAD = computer-aided diagnosis, max = maximum, min = minimum, NST = no special type, SD = standard deviation

CORRELATION BETWEEN CAD AND PATHOLOGIC AND IMMUNOHISTOCHEMICAL PARAMETERS

High mean angiovolume was related to ER (−), PR (−), and HER2 (+) status (p = 0.002, p = 0.009, p = 0.004, respectively). In addition, a higher mean angiovolume was observed in tumors of higher nuclear and histologic grade (p = 0.001 and p < 0.001, respectively) and in tumors of the LN (+), Ki-67 (+) and HER2-overexpression subtype (p = 0.006, p = 0.035, and p = 0.006, respectively). A high rate of washout and a low rate of persistent enhancement were observed in NST carcinoma with Ki-67 (+) (p < 0.001 and p = 0.002, respectively). The curve peak showed a tendency similar to that of angiovolume, but the correlation was not significant. Although a higher mean angiovolume was observed in CK5/6 (+) and EGFR (+) tumors, the correlations between these factors were not significant (Table 2). Angiovolume showed a moderate correlation with pathologic tumor size (r = 0.632, p < 0.001). The curve peak showed a weak correlation with pathologic tumor size (Table 3).

Table 2. IHC Markers and CAD Parameters of NST Carcinoma.

^*†Different letters indicate significant differences between groups based on the Dwass, Steel, Critchlow-Fligner multiple comparison analysis.

CAD = computer-aided diagnosis, EGFR = epidermal growth factor receptor, ER = estrogen receptor, HER2 = human epidermal growth factor 2, IHC = immunohistochemistry, LA = luminal A, LB = luminal B, NST = no special type, PR = progesterone receptor, TN = triple negative

Table 3. Spearman Correlation between Pathology Size and Computer-Aided Diagnosis Parameters of NST Carcinoma.

NST = no special type

BINARY LOGISTIC REGRESSION ANALYSIS

High angiovolume was independently correlated with high nuclear grade (grade 2 and 3, odds ratio [OR] = 2.055, 95% confidence interval [CI]: 1.048–4.028, p = 0.036), high histologic grade (grade 2 and 3, OR = 2.175, 95% CI: 1.152–4.107, p = 0.017) and LN (+) status (OR = 0.802, 95% CI: 0.668–0.963, p = 0.018) in invasive carcinoma of NST (Table 4). Additionally, high angiovolume correlated with ER (−) (OR = 1.287, 95% CI: 1.077–1.538, p = 0.006), PR (−) (OR = 1.214, 95% CI: 1.025–1.439, p = 0.025), and HER2 (+) status (OR = 0.755, 95% CI: 0.616–0.926, p = 0.007) (Fig. 2A, B). A higher washout rate of NST carcinoma correlated with Ki-67 (+) status (OR = 0.966, 95% CI: 0.944–0.989, p = 0.004). Although a high angiovolume correlated with Ki-67 (+) status, this correlation was not significant (OR = 0.844, 95% CI: 0.704–1.013, p = 0.069).

Table 4. Results of the Univariate Binary Logistic Regression Analysis between Immunohistochemistry Markers and Computer-Aided Diagnosis Parameters.

^*Nuclear grade was grouped into 1 and 2 + 3.

^†Histologic grade was grouped into 1 and 2 + 3.

^‡Subtype was grouped into luminal (luminal A and B subtypes) and nonluminal (HER2 overexpression and triple negative subtypes).

CI = confidence interval, ER = estrogen receptor, HER2 = human epidermal growth factor 2, LN = lymph node, NST = no special type, PR = progesterone receptor

Fig. 2. MR images with a computer-aided detection color overlay map in a 61-year-old female with high angiovolume (6.1 cc) and non-luminal subtype (HER2-overexpression subtype; estrogen receptor-negative, progesterone receptor-negative, HER2-positive and Ki-67-positive) invasive carcinoma of no special type in the left breast, showing histologic and nuclear grade 3. Regional lymph node was metastasis-positive.

A. Postcontrast T1-weighted image in the early phase shows an irregular and heterogeneously enhancing mass in the upper inner quadrant of the left breast.

B. Kinetic curve graph shows a rapid early enhancement and a delayed washout-type curve. The initial peak enhancement value is 521%. With respect to the delayed-phase enhancement, 25% of the mass represents washout, 57% of the mass represents a persistent-type curve, and 17% represents a plateau-type curve.

HER2 = human epidermal growth factor 2

DISCUSSION

Our study may suggest that a high angiovolume on CAD is associated with not only pathology size but also high nuclear and histologic grade; LN (+), ER (−), PR (−), and HER2 (+) status; and nonluminal subtypes in NST carcinoma. A high rate of washout and a low rate of persistent enhancement on CAD correlated with Ki-67 (+) status, too.

Angiogenesis, the rapid increase in the formation of blood vessels, is required for the supply of sufficient oxygen and nutrition for breast tumor growth, and breast cancer cells require persistent nourishment and oxygen supply through the vascular network of capillaries in the system (16). It has been documented that tumor angiogenesis is a critical factor in breast tumor growth and aggressiveness, and contrast enhancement is thought to reflect tumor angiogenesis (17). Therefore, an increased level of angiogenesis is associated with decreased survival in breast cancer patients (18). The CAD program reconstructs the enhancing component within a tumor meeting a given threshold and can be used to calculate the volume of this component, which is referred to as angiovolume. We hypothesized that this value can indirectly represent tumor angiogenesis. Consequently, we suggest that angiovolume on CAD could provide vital prognostic information in breast cancer patients. Angiovolume is already known to correlate with the response to neoadjuvant chemotherapy and the survival rate (6). In addition, angiovolume may be related to pathologic tumor size and high Ki-67 expression (12). In our demonstration, a higher mean angiovolume was observed in Ki-67 (+) NST carcinoma than Ki-67 (−) NST carcinoma. However, this difference was not significant in the binary logistic regression analysis. We also found that high angiovolume correlated with not only pathologic tumor size but also high nuclear and histologic grade; LN (+), ER (−), PR (−), and HER2 (+) status; and nonluminal subtypes. According to Bharti et al. (19), the mean microvessel density on pathologic specimens was the highest in ER (−) and PR (−) samples, and a significant correlation was found between ER status and mean vascular density. Song et al. (20) reported that circumscribed mass margin, associated nonmass enhancement on MR images, high histologic grade, high Ki-67 index, and older age could be associated with HER2 positivity.

The curve peak is the highest enhancing pixel in the early phase of enhancement on dynamic enhancement images. Although the curve peak shows a similar tendency to angiovolume in our study, the correlation between the two parameters was not significant.

For the time-signal intensity curve (TIC) pattern, early rapid enhancement with delayed washout is generally regarded as a suspicious pattern in the area of greatest enhancement in a manually drawn region of interest within a tumor on breast MRI. Few studies have investigated the correlation of the TIC and the molecular subtypes of breast cancer. A study reported that the TN breast cancer subtype was significantly associated with high histologic grade and a persistent enhancement pattern on breast MRI. This difference in the TIC pattern between the TN breast cancer subtype and other breast cancer subtypes may be due to the heterogeneity of the internal enhancement of TN breast cancer subtype (21). Our study automatically obtained the TIC and the rates of early and delayed enhancement profiles from the whole lesion by CAD instead of a single TIC pattern. We did not identify significant differences in the rates of persistent or washout patterns between subtypes.

Ki-67 is a nuclear protein associated with cellular proliferation and a powerful prognostic marker in breast cancer. It is helpful in assessing the risk of recurrence for luminal subtype breast cancers (22). Song et al. (12) reported that kinetic features such as peak enhancement, angiovolume, and a delayed plateau pattern showed positive correlations with the Ki-67 index. Tumors that proliferate faster show faster enhancement on DCE-MRI (23). Another study found that a washout curve may predict a higher Ki-67 index (24). Our study also revealed that a high rate of delayed washout and a low rate of delayed persistent enhancement in NST carcinoma correlated with Ki-67 (+) status, as in previous studies.

CK5/6 and EGFR are adverse prognostic markers in TN breast cancer (25). While cancers positive for these markers showed a higher mean angiovolume than those negative for these markers, the correlation was not significant in this study.

Invasive cancers are known to present rapid enhancement kinetic curves. Nam et al. (26) suggested that higher histologic grade and ER (−), PR (−) and p53 (+) status are associated with a higher rate of rapid enhancement components on early phase enhancement. However, in our study, early enhancement profiles showed no significant association with any IHC markers.

Our study has some limitations. First, there were relatively small numbers of HER2-overexpression cases (n = 15) compared with the sample sizes of the other groups. Second, our study was performed retrospectively in a single center. These results should be supported in larger group studies and prospective studies. Moreover, morphologic characteristics of the lesion according to standardized Breast Imaging-Reporting and Data System descriptors were not considered. Combining CAD parameters with descriptors such as spiculated or irregular margins and shape strongly indicating malignancy could lead to closer correlations of these features with IHC markers, and these results could finally suggest a more precise clinical course. Further comparison studies of each CAD feature and IHC markers of cancer may help explain these correlations.

In conclusion, our results suggest that the angiovolume and delayed washout/persistent rate from CAD parameters of invasive carcinoma of NST are the most useful predictive factor reflecting IHC markers of breast cancer preoperatively.