Comparison of False-Positive Versus True-Positive Findings on Contrast-Enhanced Digital Mammography

Abstract

BACKGROUND.

Contrast-enhanced digital mammography (CEDM) has been shown to outperform standard mammography while performing comparably to contrast-enhanced MRI.

OBJECTIVE.

The purpose of our study was to compare imaging characteristics of false-positive and true-positive findings on CEDM.

METHODS.

This retrospective study included women who underwent baseline screening CEDM between January 2013 and December 2018 assessed as BI-RADS category 0, 3, 4, or 5 and who underwent biopsy with histopathologic diagnosis or had a 2-year imaging follow-up. Lesion characteristics were extracted from CEDM reports. A true-positive finding was defined as a lesion in which biopsy yielded malignancy. A false-positive finding was defined as a lesion in which biopsy yielded benign or benign high-risk pathology or in which 2-year imaging follow-up was negative.

RESULTS.

Of 157 patients (median age, 52 years), 24 had a total of 26 true-positive lesions, and 133 had a total of 147 false-positive lesions. Of the 26 true-positive lesions, one (4%) exhibited only a mammographic finding on low-iodine images, 13 (50%) exhibited only a contrast finding on iodine images, and 12 (46%) exhibited both a mammographic finding on low-energy images and a contrast finding on iodine images. A true-positive result was more likely (p = .02) for lesions present on both low-energy images and iodine images (31%) than on low-energy images only (4%) or iodine images only (12%). Among lesions present on both low-energy and iodine images, a true-positive result was more likely (p < .001) when the type of mammographic finding was an asymmetry (46%) or calcification (80%) than a mass (11%) or distortion (0%). A true-positive result was more likely (p = .01) among those with, versus those without, an ultrasound correlate (36% vs 9%) and also was more likely (p = .02) among those with, versus those without, an MRI correlate (18% vs 2%). Of 25 false-positive calcifications, 24 had no associated mammographic enhancement; of five true-positive calcifications, four had mammographic enhancement.

CONCLUSION.

A low-energy mammographic finding with associated enhancement or a finding with a sonographic or MRI correlate predicts a true-positive result. Calcifications with associated enhancement had a high malignancy rate. Nonetheless, half of true-positive lesions enhanced on iodine images without a mammographic finding on low-energy images.

CLINICAL IMPACT.

These observations inform radiologists’ management of abnormalities detected on screening CEDM.

Breast cancer is the most common cancer worldwide, with an estimated 2.3 million new cases diagnosed in 2020 [1]. Mammography is the standard of care in breast cancer screening and has been shown to significantly reduce mortality from breast cancer [2–4]. However, despite its proven benefits, mammography has limitations. The sensitivity of full-field digital mammography (FFDM) ranges from 75% to 85% [5], but the sensitivity of FFDM is only 38–50% in women with dense breasts [6, 7].

Contrast-enhanced digital mammography (CEDM) is a modified digital mammographic examination performed after the addition of an iodine-based contrast agent. Studies have shown that CEDM, compared with FFDM, has a higher sensitivity, specificity, and accuracy [8–11]. Studies have also shown that CEDM has comparable performance to dynamic contrast-enhanced MRI, with some studies reporting similar sensitivity [12] and specificity [12, 13] for both imaging modalities and others reporting slightly lower sensitivity but higher specificity for CEDM compared with MRI [14, 15].

Prior studies have investigated the role of CEDM for the evaluation of calcifications [16], distortions [17], and asymmetries [18]. Prior studies have also evaluated the false-positive and true-positive rates for FFDM [19–23] and breast MRI [24–26]. However, to our knowledge, no study has performed an overall comparative analysis of true-positive findings versus false-positive findings on CEDM. The aim of this study was to compare the imaging characteristics of false-positive and true-positive findings on CEDM.

Methods

Patients

The institutional review board approved this retrospective study and waived the requirement for written informed consent. The study was conducted in compliance with HIPAA. The study was conducted at a single tertiary care center.

The radiology department database was searched to identify patients who underwent a baseline (initial) screening CEDM examination between January 2013 and December 2018. Of these patients, those who received an American College of Radiology (ACR) benign assessment (BI-RADS category 1 or 2) on the baseline screening CEDM were excluded from the analysis (aside from the calculation of the true-negative rate of CEDM, as described later). Of the remaining patients, those without a histopathologic diagnosis or without 2-year cancer-free imaging follow-up (i.e., 2-year negative MRI, mammographic, or CEDM follow-up) were considered lost to follow-up and were excluded. These exclusions resulted in a study sample of patients with a BI-RADS assessment of incomplete, probably benign, suspicious, or highly suspicious (BI-RADS categories of 0, 3, 4, and 5, respectively) on the baseline screening CEDM examination and with available follow-up for the mammographic abnormality on CEDM. Medical records were reviewed for patients in the final study sample to record patient age, prior breast pathology (including history of breast cancer or high-risk lesion), history of thoracic radiation, and family history of breast cancer.

A total of 68 patients in the study sample were included in prior studies that evaluated the screening performance of supplemental CEDM in comparison with conventional mammography [13] and MRI [12, 13], evaluated the role of CEDM in patients with intermediate breast cancer risk [27], and evaluated the utility of targeted ultrasound in predicting malignancy among lesions detected on CEDM [28]. The patient sample in this study differs from those in the previous studies, as this study included patients with both high and intermediate breast cancer risk and evaluated only baseline screening CEDM examinations. This study also has a distinct aim from the earlier studies of identifying the clinical and imaging characteristics that may differentiate true-positive findings from false-positive findings on CEDM.

Baseline Screening CEDM Examinations

At our institution, screening CEDM is performed in asymptomatic patients with high or intermediate breast cancer risk, which includes patients referred from the institution’s high-risk clinic (patients with genetic mutations or variants, personal history of high-risk breast lesion, history of chest wall radiation, or strong family history of breast cancer) as well as patients with a history of breast cancer referred by breast surgeons or medical oncologists. In these patients, screening CEDM is performed in lieu of annual screening mammography and includes the acquisition of standard low-energy 2D images, which serve as the equivalent of routine FFDM [29].

During the study period, screening CEDM examinations were performed using a dual-energy mammography system (Senographe Essential, GE Healthcare; or 3Dimensions, Hologic). Iohexol (350 mg I/mL; Omnipaque 350, GE Healthcare) was administered IV at a dose of 1.5 mL/kg, up to a maximum dose of 150 mL, using a power injector at a rate of 3 mL/s. Imaging was performed 2.0–2.5 minutes after injection. Craniocaudal and mediolateral oblique views were obtained of each breast. In patients who underwent surgery within 5 years before screening CEDM, two-view spot images were acquired over lumpectomy sites. Two energy exposures straddling the K-edge of iodine were performed nearly simultaneously for each exposure: one low-energy (26–30 kVp) and one high-energy (45–49 kVp) exposure. Recombination low- and high-energy images (hereafter, described as iodine images) were generated using a vendor-specific proprietary algorithm that highlights areas of contrast enhancement.

Patients who experienced a contrast agent reaction during the CEDM examination were evaluated by a nurse and a radiologist. Each contrast agent reaction was graded as mild, moderate, or severe, as outlined in the ACR Manual on Contrast Media [30], and documented in the medical record.

Additional Imaging

Abnormalities seen on low-energy or iodine images were further evaluated at the time of screening CEDM with additional mammographic views and/or ultrasound. In patients with abnormal enhancement on CEDM but without a mammographic or sonographic correlate, either contrast-enhanced breast MRI or 6-month follow-up CEDM was recommended, depending on the interpreting radiologist’s level of suspicion. The decision to perform further ultrasound evaluation versus to recommend MRI without ultrasound evaluation was at the discretion of the interpreting radiologist. When recommended, MRI served to further characterize enhancement on CEDM and to provide a modality for biopsy, as necessary. At the time of stereotactic-, ultrasound-, and MRI-guided biopsies, postbiopsy clip markers were placed; clip placement on the postprocedure mammogram was correlated to the original targeted mammographic finding.

Imaging Interpretation and BI-RADS Assignment

Screening CEDM examinations performed at our institution are routinely interpreted by dedicated breast radiologists. The baseline screening CEDM examinations performed in the study sample were interpreted clinically by 30 dedicated breast radiologists with a range of 3–40 years of experience in breast imaging and 0–10 years of experience interpreting CEDM. Imaging interpretation was based on the 5th edition of the BI-RADS mammography lexicon for findings on the low-energy images and the MRI BI-RADS lexicon for findings on the iodine images, aside from contrast enhancement kinetics [31]. The interpreting radiologist assigned a single overall BI-RADS category for each screening CEDM examination, which was based collectively on all images (low-energy and iodine). BI-RADS categories of 0, 3, 4, or 5 could be assigned given that the necessary diagnostic mammographic or sonographic workup was performed at the time of the baseline screening CEDM. BI-RADS categories of 0 or 3 were assigned for patients who were recommended for further evaluation with MRI after this additional workup. A separate dedicated BI-RADS assessment was assigned for any subsequent MRI examination.

A single breast radiologist (T.A.) with 3 years of experience (who was among the radiologists who had performed the clinical CEDM interpretations) reviewed the clinical reports of the baseline screening CEDM examinations to determine breast density (categorized as almost entirely fatty, scattered fibroglandular densities, heterogeneously dense, or extremely dense); background parenchymal enhancement (categorized as minimal, mild, moderate, or marked); the laterality of the finding relative to the laterality of prior high-risk pathology (categorized as ipsilateral or contralateral); the images on which the findings were visualized (categorized as low-energy images only, iodine images only, or both); the type of mammographic finding on low-energy images (categorized as asymmetry, mass, calcification, or distortion); the type of contrast finding on iodine images (categorized as focus, mass, or on-mass enhancement); the number of views on which the finding was present (categorized as one or two views); and, for findings present on one view, the mammographic projection showing the finding (categorized as craniocaudal only or mediolateral oblique only). The investigator also recorded the presence versus absence of an ultrasound correlate and of an MRI correlate for the abnormality on CEDM, among examinations for which this additional workup was performed.

Reference Standard

Each baseline CEDM examination and each lesion were categorized as a true-positive or a false-positive on the basis of histopathologic results (classified as benign or malignant) or follow-up imaging. A true-positive finding was defined as an examination or a lesion in which biopsy yielded a diagnosis of malignancy. A false-positive finding was defined as an examination or a lesion in which biopsy yielded benign or benign high-risk pathology or, if biopsy results were unavailable, in which 2-year follow-up MRI, mammography, or ultrasound was negative. Examinations with both true-positive and false-positive lesions were classified as a true-positive at the examination level in concordance with the BI-RADS Atlas [31].

Determination of Performance Metrics of Screening CEDM

To determine performance metrics of screening CEDM, the negative examinations that were excluded from the study cohort (i.e., screening CEDM examinations with a BI-RADS category of 1 or 2) were reviewed to identify those with at least 1 year of cancer-free imaging follow-up; these cases represented true-negatives. The false-positive rate of screening CEDM was calculated as the number of false-positive examinations divided by the sum of the false-positives and true-negatives. The specificity of screening CEDM (i.e., 100% minus the false-positive rate as a percentage) was calculated as the number of true-negative examinations divided by the sum of false-positives and true-negatives. The PPV1 of screening CEDM was calculated as the number of true-positive examinations divided by the number of positive examinations. The number of lesions that underwent biopsy was recorded, and the PPV of biopsies performed (PPV3) was calculated on a per-lesion basis as the number of lesions that were positive on biopsy divided by the number of lesions that underwent biopsy. The true-positive rate (i.e., the sensitivity) was calculated as the number of true-positive examinations divided by the sum of true-positives and false-negatives.

Statistical Analysis

Findings were summarized using descriptive statistics. Patient- and examination-level characteristics were compared between patients with false-positive and true-positive baseline CEDM examinations using the Wilcoxon rank sum test for continuous variables and Fisher exact test for categoric variables. Lesion-level imaging characteristics were compared between true-positive and false-positive lesions using the generalized estimating equation framework to account for intrapatient correlation. Univariable logistic regression models were used to calculate odds ratios for predicting a true-positive result for the presence of the finding on both low-energy and iodine images, the presence of an ultrasound correlate, and the presence of an MRI correlate. A subanalysis was performed to tabulate false-positives and true-positives among suspicious calcifications stratified by whether the calcification exhibited associated enhancement observed on iodine images. All p values were obtained using robust z-scores. Multiple-comparison adjustments were performed using the Benjamini-Hochberg procedure, which controls the false discovery rate. The type I adjusted error rate (α) was set to 0.05. Multivariable analyses were not performed because of the limited sample size. Statistical analyses were conducted using R software (version 3.6.3, The R Foundation).

Results

Patient Characteristics

A total of 1805 patients underwent baseline screening CEDM during the study period. Of these 1805 patients, 184 (10%) had a positive screening CEDM examination (i.e., examination assigned a final BI-RADS category of 0, 3, 4, or 5). Of these 184 patients, 27 (15%) were lost to follow-up. The remaining 157 patients (all women; median age, 52 years; age range, 27–74 years) formed the study sample. Table 1 summarizes the characteristics of these patients. Among the 157 patients, 64 (41%) had a history of breast cancer, and 50 (32%) had a history of a high-risk lesion. Among the 157 patients, none had breasts that were almost entirely fatty, 14 (9%) had scattered fibroglandular densities, 126 (80%) had heterogeneously dense breasts, and 17 (11%) had extremely dense breasts.

TABLE 1:

Comparison of Patient-Level Characteristics Between False-Positive and True-Positive CEDM Examinations

Characteristic	Totala (n = 157)	False-Positivea (n = 133)	True-Positivea (n = 24)	p b	Percentage True-Positivec
Age (y), median (IQR)	52 (46–59)	52 (45–58)	52 (50–63)	.07
Prior breast pathology				.30
Cancer	64 (41)	55 (41)	9 (38)		14
High-risk lesion or atypia	50 (32)	39 (29)	11 (46)		22
None	43 (27)	39 (29)	4 (17)		8
History of thoracic radiation				.08
Yes	19 (12)	19 (14)	0 (0)		0
No	138 (88)	114 (86)	24 (100)		17
Family history of breast cancer				.03
Yes	24 (16)	24 (18)	0 (0)		0
No	133 (84)	109 (82)	24 (100)		18
Breast density				.91
Almost entirely fatty	0 (0)	0 (0)	0 (0)		NA
Scattered fibroglandular densities	14 (9)	12 (9)	2 (8)		14
Heterogeneously dense	126 (80)	107 (80)	19 (79)		15
Extremely dense	17 (11)	14 (11)	3 (13)		18
Background parenchymal enhancement				.12
Minimal	61 (39)	47 (35)	14 (58)		23
Mild	52 (33)	46 (35)	6 (25)		12
Moderate	40 (25)	37 (28)	3 (13)		8
Marked	4 (3)	3 (2)	1 (4)		25

Note—CEDM = contrast-enhanced digital mammography, IQR = interquartile range, NA = not applicable (i.e., no patients exhibit given characteristic).

^aUnless otherwise indicated, values represent number of patients with percentages in parentheses. Some percentages do not add up to 100 because of rounding.

^bComparison of characteristic between false-positives and true-positives; values are presented in boldface type when statistically significant after multiple-comparisons adjustment.

^cPercentage of true-positives among total number of patients with given characteristic.

Performance Characteristics of Baseline Screening CEDM

Of the 1805 patients who underwent baseline screening CEDM, 1621 had a negative screening examination (i.e., examination assigned a final BI-RADS category of 1 or 2); of these, 1503 had at least 1 year of negative imaging follow-up and were considered true-negatives, and five were diagnosed with interval cancers and were considered false-negatives. Of the 157 patients with positive CEDM examinations, 24 (15%) had true-positive examinations and 133 (85%) had false-positive examinations. Thus, the false-positive rate was 8%: [133 false-positives / (133 false-positives + 1503 true-negatives)]. The specificity was 92%: [1503 true-negatives / (133 false-positives + 1503 true-negatives)]. The sensitivity was 83%: [24 true-positives / (24 true-positives + 5 false-negatives)]. The PPV1 was 15% (24 true-positives / 157 positives).

Flow of Lesion Diagnostic Evaluation

The 157 patients had a total of 173 lesions on the baseline screening CEDM examinations with an available reference standard (histopathologic results from biopsy in 77 lesions; 2-year negative imaging follow-up in 96 lesions). Of the 77 lesions that underwent biopsy, 51 were benign, and 26 were malignant. Thus, there were a total of 26 true-positive lesions and 147 false-positive lesions (51 with benign biopsy + 96 with 2-year negative imaging follow-up). The PPV3 was 34% (26 true-positive lesions / 77 lesions biopsied). On all but one examination with multiple lesions, either all lesions were true-positives or all lesions were false-positives. One examination showed both a true-positive lesion and a false-positive lesion and was classified as a true-positive at the examination level.

Of the 173 lesions identified on CEDM, 28 were identified on low-energy images only, 106 on iodine images only, and 39 on both low-energy and iodine images (Fig. 1A). Of the 26 true-positive lesions, one (4%) exhibited only a mammographic finding on low-iodine images, 13 (50%) exhibited only a contrast finding on iodine images, and 46% (12) exhibited both a mammographic finding on low-energy images and a contrast finding on iodine images. Of the 147 false-positive lesions, 27 (18%) exhibited only a mammographic finding on low-iodine images, 93 (63%) exhibited only a contrast finding on iodine images, and 27 (18%) exhibited both a mammographic finding on low-energy images and a contrast finding on iodine images.

Fig. 1 —

Flowcharts of study results. CEDM = contrast-enhanced digital mammography, LE = low-energy, US = ultrasound, FU = follow-up, IDC = intraductal carcinoma, ILC = invasive lobular carcinoma, DCIS = ductal carcinoma in situ.

A, Flowchart shows results of evaluation for 173 lesions identified on CEDM.

B, Flowchart shows results of evaluation for 106 lesions identified only on iodine images.

C, Flowchart shows results of evaluation for 39 lesions identified on both iodine and low-energy images.

D, Flowchart shows results of evaluation for 28 lesions identified only on low-energy images.

Figure 1B shows the flow of diagnostic evaluation for the 106 lesions identified on iodine images only. Of these 106 lesions, 13 (12%) were cancer (four mass enhancement, nine nonmass enhancement). Of these 13 cancers, 12 (92%) underwent same-day ultrasound evaluation, of which six (50%) had a sonographic correlate for the enhancement on CEDM, allowing ultrasound-guided biopsy. The other six of 12 (50%) cancers lacked a sonographic correlate for the enhancement on CEDM, but subsequent MRI revealed a correlate, enabling MRI-guided biopsy. One cancer (nonmass enhancement on CEDM) underwent same-day MRI rather than same-day ultrasound evaluation; MRI did not reveal a correlate for the enhancement on CEDM. Screening ultrasound follow-up at 6 months in this patient identified an ultrasound correlate, and subsequent ultrasound-guided core biopsy yielded invasive lobular carcinoma.

Figure 1C shows the flow of diagnostic evaluation for the 39 lesions identified on both low-energy and iodine images. Of these 39 lesions, 12 (31%) were cancer (two masses with enhancement, four calcifications with enhancement, and six asymmetries with enhancement). Of the 12 cancers, 10 (83%) underwent same-day ultrasound, of which eight (80%) had a sonographic correlate, allowing ultrasound-guided core biopsy. The other two of 10 (20%) cancers (one calcification with enhancement and one asymmetry with enhancement) lacked an ultrasound correlate and were biopsied by stereotactic guidance. Of the two of 12 (17%) cancers that did not undergo same-day ultrasound, one (8%) cancer (calcification with enhancement) underwent stereotactic biopsy, and one (8%) cancer (asymmetry with enhancement) underwent MRI and subsequent MRI-guided biopsy of correlative enhancement.

Figure 1D shows the flow of diagnostic evaluation for the 28 lesions identified on low-energy images only. Only one cancer was identified, which presented as calcifications and was biopsied using stereotactic guidance.

Figure 2 shows a false-positive finding at CEDM. Figures 3 and 4 show true-positive findings at CEDM.

Fig. 2—

45-year-old woman with history of right breast fibroepithelial lesion.

A, Contrast-enhanced digital mammography shows mass (arrows) in upper outer left breast on mediolateral oblique (MLO) (left) and craniocaudal (CC) (right) low-energy images.

B, Associated mass enhancement (arrows) is present on MLO (left) and CC (right) iodine images.

C and D, Gray-scale (C) and color Doppler (D) ultrasound images of left breast show correlative oval circumscribed parallel homogeneously hypoechoic mass at 2-o’clock position, 8 cm from nipple. Ultrasound-guided biopsy yielded benign phyllodes tumor, which was confirmed at surgical excision.

Fig. 3—

70-year-old woman with history of left breast atypical ductal hyperplasia.

A, Contrast-enhanced digital mammography shows focal nonmass enhancement (arrows) at postsurgical scar site on craniocaudal (left) and mediolateral oblique (right) iodine images. No correlate was identified on low-energy images or ultrasound evaluation.

B, Sagittal subtraction T1-weighted postcontrast MR image shows oval enhancing mass (arrow) that correlates with finding on iodine images (A). MRI-guided biopsy yielded invasive lobular carcinoma.

Fig. 4—

42-year-old woman with history of lumpectomy for right breast ductal carcinoma in situ.

A, Contrast-enhanced digital mammography of right breast upper outer quadrant shows grouped amorphous calcifications (arrows) separate from postsurgical scar site on low-energy (magnification, insets) craniocaudal (CC) (left) and mediolateral (right) images.

B, Iodine images show associated enhancement (arrows) on CC (left) and mediolateral oblique (right) views. Stereotactic biopsy of calcifications yielded ductal carcinoma in situ.

Comparison of Imaging Characteristics of False-Positive Lesions Versus True-Positive Lesions

Table 1 compares patient-level characteristics between patients with false-positive examinations and those with true-positive examinations. Table 2 compares lesion-level characteristics between false-positive lesions and true-positive lesions. A true-positive result, compared with a false-positive result, was significantly less likely in patients with, versus those without, a family history of breast cancer (0% vs 18%, p = .03). A true-positive result was significantly more likely (p = .02) for lesions present on both low-energy and iodine images (31%) than on low-energy images only (4%) or iodine images only (12%). Among lesions present on both low-energy and iodine images, a true-positive result was significantly more likely (p < .001) when the type of mammographic finding was asymmetry (46%) or calcification (80%) than mass (11%) or distortion (0%). Among lesions undergoing ultrasound workup, a true-positive result was significantly more likely (p = .01) among those with, versus those without, an ultrasound correlate (36% vs 9%). Among lesions undergoing MRI workup, a true-positive result was significantly more likely (p = .02) among those with, versus those without, an MRI correlate (18% vs 2%).

TABLE 2:

Comparison of Lesion-Level Characteristics Between False-Positive and True-Positive Lesions on CEDM

Characteristic	Totala	False-Positivea	True-Positivea	p b	Percentage True-Positivec
Laterality of CEDM finding relative to laterality of prior high risk of cancer pathology	126	103	23	.90
Ipsilateral	67 (53)	54 (52)	13 (57)
Contralateral	59 (47)	49 (48)	10 (43)
CEDM images showing finding	173	147	26	.02
Low-energy image only	28 (16)	27 (18)	1 (4)		4
Iodine image only	106 (61)	93 (63)	13 (50)		12
Both low-energy and iodine images	39 (23)	27 (18)	12 (46)		31
Type of mammographic finding for lesions seen only on low-energy images	28	27	1	.06
Asymmetry	2 (7)	2 (7)	0 (0)		0
Mass	1 (4)	1 (4)	0 (0)		0
Calcification	25 (89)	24 (89)	1 (100)		4
Distortion	0 (0)	0 (0)	0 (0)		NA
Type of mammographic finding for lesions seen on both low-energy and iodine images	39	27	12	< .001
Asymmetry	13 (33)	7 (26)	6 (50)		46
Mass	19 (49)	17 (63)	2 (17)		11
Calcification	5 (13)	1 (4)	4 (33)		80
Distortion	2 (5)	2 (7)	0 (0)		0
Type of contrast finding for lesions seen only on iodine images	106	93	13	.20
Focus	12 (11)	12 (13)	0 (0)		0
Mass	19 (18)	15 (16)	4 (31)		21
Nonmass enhancement	75 (71)	66 (71)	9 (69)		12
Type of contrast finding for lesions seen on both low-energy and iodine images	39	27	12	> .99
Focus	1 (3)	1 (4)	0 (0)		0
Mass	17 (44)	12 (44)	5 (42)		29
Nonmass enhancement	21 (54)	14 (52)	7 (58)		33
No. of views on which a contrast lesion was detected				.49
One	39 (27)	33 (28)	6 (24)
Two	106 (73)	87 (73)	19 (76)
View showing lesion for lesions seen on one view				.60
Craniocaudal	22 (56)	19 (58)	3 (50)
Mediolateral oblique	17 (44)	14 (42)	3 (50)
Presence of US correlate for lesions that underwent US workup	130	107	23	.01
Yes	42 (32)	27 (25)	15 (65)		36
No	88 (68)	80 (75)	8 (35)		9
Presence of MRI correlate for lesions that underwent MRI workup	89	81	8	.02
Yes	40 (45)	33 (41)	7 (88)		18
No	49 (55)	48 (59)	1 (13)		2

Note—CEDM = contrast-enhanced digital mammography, NA = not applicable (no lesions exhibit given characteristic), US = ultrasound.

^aUnless otherwise indicated, values represent number of lesions with percentages in parentheses. Some percentages do not add up to 100 because of rounding.

^bComparison of characteristic between false-positives and true-positives; values are presented in boldface type when statistically significant following multiple-comparisons adjustment.

^cPercentage of true-positives among total number of lesions with given characteristic.

Patient age, prior cancer or high-risk pathology, history of thoracic radiation, breast density, degree of background parenchymal enhancement, laterality of the finding relative to the laterality of prior high-risk or cancer pathology, the type of mammographic finding for lesions seen only on low-energy images, the type of contrast finding for lesions seen only on iodine images, the type of contrast finding for lesions seen on both low-energy and iodine images, the number of views on which a lesion was detected, and the view showing the lesion for lesions seen on one view were not significantly different between false-positive cases and true-positive cases (all, p > .05).

Table 3 provides the results of the univariable regression analyses for predicting a true-positive result. A true-positive result was significantly associated with lesion detection on both low-energy and iodine images (OR = 12.50; 95% CI, 2.17–238.00) relative to detection on low-energy images only; the presence of an ultrasound correlate (OR = 4.87; 95% CI, 1.86–13.50) relative to absence of an ultrasound correlate; and presence of an MRI correlate (OR = 10.50; 95% CI, 1.81–200.00) relative to absence of an MRI correlate.

TABLE 3:

Results of Univariable Logistic Regression Analyses for Predicting a True-Positive Result for Lesions Detected on CEDM

Factor	Characteristic	Reference	OR (95% CI)	p
Images showing finding	Both low-energy and iodine images	Low-energy image only	12.50 (2.17–238.00)	.02
US correlate for lesions that underwent US workup	Present	Absent	4.87 (1.86–13.50)	.01
MRI correlate for lesions that underwent MRI workup	Present	Absent	10.50 (1.81–200.00)	.02

Note—CEDM = contrast-enhanced digital mammography, OR = odds ratio, US = ultrasound.

In the subgroup analysis of calcifications, 30 suspicious calcifications were observed in 29 patients. Of the 30 calcifications, 25 were false-positives, and five were true-positives. Of the 25 false-positive calcifications, 24 (96%) had no associated enhancement on iodine images, and one (4%) had associated nonmass enhancement (pathology yielded fibroadenomatoid change, dense stromal fibrosis, and foreign body giant cell reaction). Of the five true-positive calcifications, four (80%) were associated with enhancement on iodine imaging (nonmass enhancement in all four), and one (20%) had no associated enhancement.

Contrast Agent Reactions

Two of the 157 (1%) patients experienced a contrast agent reaction. Both patients had a mild reaction that involved hives; one patient required administration of diphenhydramine. Both patients were evaluated by a radiologist and observed for resolution of symptoms before leaving the department in stable condition.

Discussion

In this study of women with an intermediate-to-high risk of breast cancer who underwent baseline screening CEDM, lesions with both a mammographic and contrast finding as well as lesions with an ultrasound or MRI correlate were more likely to represent a true-positive for cancer. Among lesions present on both low-energy and iodine images, those exhibiting a mammographic abnormality of asymmetry or calcification, rather than mass or distortion, were also more likely to represent cancer. Nonetheless, 50% of true-positive abnormalities exhibited a contrast finding but not a mammographic finding.

An increasing number of studies have reported on the clinical utility and performance of CEDM [12, 13, 27, 28, 32, 33], and the results to date support the technique’s utility as an alternative to contrast-enhanced MRI [12–14, 34–36]. A recent meta-analysis reported a pooled sensitivity of 89% and specificity of 84% for CEDM [37]. The PPV of biopsies performed based on abnormal findings on screening CEDM (PPV3) has been reported as 29.4% [13], which is comparable to the PPV3 of both mammography (20–40%) and breast MRI (20–50%) [31, 38] and is greater than that reported for breast ultrasound (11–21%) [39]. In this study, our data yielded a specificity of 92% and a PPV3 of 34%, which are consistent with the published literature [13, 38]. The observed false-positive rate of CEDM of 8% is also consistent with false-positive rates reported for noncontrast mammography [40].

We aimed to determine the imaging characteristics that could differentiate false-positive lesions from true-positive lesions at baseline screening CEDM. Although prior studies have specifically assessed CEDM in the evaluation of calcifications, architectural distortions, and asymmetries [16–18], to our knowledge, no data have been published providing a comparative analysis of true-positive lesions and false-positive lesions on CEDM.

At baseline screening CEDM, a lesion seen on both low-energy and iodine images was 12.50 times more likely to be a true-positive finding than a lesion seen on only low-energy images. Indeed, 46% of true-positive findings, compared with only 18% of false-positive findings, were identified on both low-energy and iodine images. For lesions present on only low-energy or iodine images, the types of mammographic and contrast findings, respectively, were not associated with a true-positive result. However, for lesions present on both low-energy and iodine images (e.g., mammographic abnormalities with associated enhancement), asymmetries and calcifications were significantly associated with a true-positive result.

Our data suggest that the presence of enhancement associated with a mammographic abnormality on CEDM may help reinforce the radiologist’s decision to pursue further workup and biopsy. Nonetheless, a finding identified only on iodine images (i.e., an enhancing abnormality without a mammographic correlate), although less likely to be cancer than a finding detected on both images, should not be dismissed, as 50% of detected cancers exhibited this pattern. Importantly, 80% (4/5) of calcifications with associated enhancement represented a true-positive finding. In comparison, 96% (24/25) of calcifications that lacked associated enhancement represented a false-positive finding. These observations are consistent with those of Cheung et al. [16], who reported that calcifications with associated enhancement are likely to represent malignancy. However, as with MRI, the lack of enhancement in patients with suspicious calcifications should not preclude stereotactic biopsy [41].

Lesions on CEDM should also be evaluated in conjunction with correlates on other modalities. If a finding on CEDM had a sonographic correlate, it was 4.87 times more likely to represent a true-positive finding. Likewise, Coffey et al. [28] reported that the likelihood of malignancy was higher among enhancing lesions on CEDM that had a sonographic correlate than those without a sonographic correlate. Further, a finding on CEDM with an MRI correlate was 10.50 times more likely to be associated with a true-positive finding. One true-positive lesion did not have an MRI correlate. However, ultrasound performed 6 months after MRI revealed a sonographic correlate, and subsequent ultra-sound-guided biopsy yielded invasive lobular carcinoma.

In contrast to breast MRI, in which nonmass enhancement is a major cause of false-positive lesions [25], our data showed no significant difference in the type of contrast finding (mass vs nonmass enhancement) between true-positive lesions and false-positive lesions on CEDM. Detection of a lesion on one versus two views and the projection for lesions seen on one view were also not statistically significant. Thus, these characteristics should not be primary considerations in radiologists’ assessment of lesions on CEDM.

This study has limitations. First, it was a single-institution retrospective study. Second, the sample predominantly included women with a history of cancer or high-risk pathology. False-positive and true-positive characteristics may differ in this intermediate- to high-risk population compared with an average-risk population, possibly due to increased pretest probability. Third, some patients with initial BI-RADS 3, 4, or 5 assessments were excluded from analysis as they were lost to follow-up; however, in general, false-positive and true-positive cases should be randomly distributed among those lost to follow-up. Fourth, given the small sample sizes of the subgroups, multivariable analysis was not performed. Finally, our study did not account for evolution in examination interpretation as the radiologists gained more experience over time. Our data include baseline studies between January 2013 and December 2018, and performance metrics may have improved with time.

In conclusion, when women are screened by CEDM, the presence of a low-energy mammographic finding with associated enhancement or the presence of a CEDM finding with a sonographic or MRI correlate helps predict a true-positive examination. Calcifications with associated enhancement had a particularly high malignancy rate. Despite these associations, as half of the true-positive lesions exhibited enhancement on iodine images without a mammographic finding on low-energy images, these lesions should be considered suspicious for malignancy. These observations should help inform radiologists’ management of abnormalities detected on screening CEDM examinations.

HIGHLIGHTS.

Key Finding

• On screening CEDM examinations, a true-positive result was significantly more likely for findings present on both low-energy and iodine images (31%) than present on only low-energy (4%) or iodine (12%) images, as well as for findings with, versus those without, an ultrasound (36% vs 9%) or MRI (18% vs 2%) correlate.

Importance

• When interpreting CEDM, consideration of associated mammographic enhancement or of ultrasound or MRI correlates may reinforce decisions to perform biopsy or help avoid unnecessary biopsies.