Accuracy and precision of contrast enhanced mammography versus MRI for predicting breast cancer size: how “good” are they really?

Abstract

Objective:

To evaluate and compare the accuracy and precision of contrast-enhanced mammography (CEM) vs MRI to predict the size of biopsy-proven invasive breast cancer.

Methods:

Prospective study, 59 women with invasive breast cancer on needle biopsy underwent CEM and breast MRI. Two breast radiologists read each patient’s study, with access limited to one modality. CEM lesion size was measured using low-energy and recombined images and on MRI, the first post-contrast series. Extent of abnormality per quadrant was measured for multifocal lesions. Reference standards were size of largest invasive malignant lesion, invasive (PathInvasive) and whole (PathTotal). Pre-defined clinical concordance ±10 mm.

Results:

Mean patient age 56 years, 42 (71%) asymptomatic. Lesions were invasive ductal carcinoma 40 (68%) with ductal carcinoma in situ (31/40) in 78%, multifocal in 12 (20%). Median lesion size was 17 mm (invasive) and 27 mm (total), range (5–125 mm). Lin’s concordance correlation coefficients for PathTotal 0.75 (95% CI 0.6, 0.84) and 0.71 (95% CI 0.56, 0.82) for MRI and contrast-enhanced spectral mammography (CESM) respectively. Mean difference for total size, 3% underestimated and 4% overestimated, and for invasive 41% and 50% overestimate on MRI and CESM respectively. LOAs for PathTotal varied from 60% under to a 2.4 or almost threefold over estimation. MRI was concordant with PathTotal in 36 (64%) cases compared with 32 (57%) for CESM. Both modalities concordant in 26 (46%) cases respectively.

Conclusion

Neither CEM nor MRI have sufficient accuracy to direct changes in planned treatment without needle biopsy confirmation.

Advances in knowledge:

Despite small mean differences in lesion size estimates using CEM or MRI, the 95% limits of agreement do not meet clinically acceptable levels.

Introduction

Breast cancer is a heterogeneous disease that is frequently multifocal and/or multicentric. ^1–4 The American Joint Committee on Cancer (AJCC) TNM staging system is used to classify breast cancer extent and predict prognosis. Tumour size for TNM stage is defined as maximum size of the invasive lesion on surgical pathology, and for multifocality, tumour size is determined by the size of the largest invasive component alone. ⁵

For the surgeon, an accurate estimate of the local extent of disease is crucial for counselling patients regarding the type of surgery required, and whether breast conservation is possible. Estimates of lesion size on imaging should reflect total size, including surrounding in-situ disease, and multifocality ± interconnecting in-situ disease.

Standard breast imaging (full-field digital mammography (FFDM), tomosynthesis and ultrasound) has limitations in estimating lesion size; ultrasound and FFDM tending to underestimate, ⁶ and tomosynthesis to overestimate, particularly in females with dense breasts. ⁷

Functional imaging techniques that are able to demonstrate neo-angiogenesis in addition to anatomical morphology, such as contrast-enhanced breast MRI (MRI), not only improve the detection of breast cancer but are more accurate in estimating lesion size and extent. ⁸ However, MRI does have disadvantages, including high cost, relatively low specificity, ⁹ inability to visualise microcalcifications, suboptimal accessibility and limited patient tolerance. ¹⁰ An association with significant delays in treatment have also been reported with MRI. ^11,12

Contrast-enhanced mammography (CEM) is a quick and easy to perform, inexpensive new technique able to demonstrate both neovascularity and lesion morphology, including microcalcification.

CEM is performed by obtaining two view mammograms using high and low energy X-ray exposures approximately 2 min after the injection of intravenous contrast. Two images per view are available for reporting, a “ low energy” image that displays morphology (equivalent to a standard mammogram ¹³ ) and a recombined image that shows areas of iodine uptake. ¹⁴ Although lacking many of the disadvantages of MRI, a CEM examination does require breast compression and exposure to ionising radiation, whereas MRI does not. Both techniques involve intravenous contrast media injection, with renal failure or known contrast allergy contraindications for both.

The aim of this study was to evaluate the comparative and independent accuracy and precision of CEM and MRI in estimating malignant lesion size in females with core biopsy-proven invasive breast cancer, using final histopathology as the reference standard.

Methods and materials

This prospective study (Australian and New Zealand Clinical Trials Registry: ACTRN 12613000684729) was conducted at two tertiary referral hospitals, approved by our institutional Human Research and Ethics Committee, and compliant with the National Health and Medical Research Council Statement on Ethical Conduct in Human Research. Written informed consent was obtained from all participants. Aspects of this cohort have been reported in two prior publications, addressing background parenchymal enhancement ¹⁵ and patient preference. ¹⁰

Females aged over 21 years attending the Breast Clinic at two tertiary institutions in Perth, Western Australia between June 2013 and October 2015, with an invasive breast cancer (of any type) on needle biopsy and fit to undergo surgery, were invited to participate.

Exclusion criteria were: inability to give written informed consent, allergy to iodinated or gadolinium-based contrast, renal insufficiency, diabetes mellitus treated with metformin, pregnancy or lactation, breast implants, and contraindications to MRI.

Standard breast imaging included FFDM with or without further views and high resolution ultrasound performed at outside practices or in our clinic.

Each participant underwent both CEM and breast MRI according to appointment availability and not timed to menstrual cycle. The examinations were performed at one centre using previously described standardised protocols, ¹⁵ by experienced breast imaging technologists.

A Breast Imaging Fellow assessed lesion type (using the National Breast Cancer Center synoptic reporting system ¹⁶ as mandated by the national screening program) and maximum size on standard breast imaging. Breast density was independently assessed by two readers (breast imaging fellow and consultant breast radiologist), with third reader arbitration in case of disagreement, and dichotomised into non-dense (BI-RADS 1 or 2) and dense (BI-RADS 3 or 4). ¹⁷

MRI findings were reported according to the BI-RADS fifth Edition lexicon. ¹⁸ For CEM studies, low energy and recombined images were reviewed, and reported using modified MRI BI-RADS descriptors. ¹⁹ Background parenchymal enhancement (BPE) on CEM and MRI was dichotomised into minimal—mild and moderate—marked.

CEM and MRIs were independently read by pairs of subspecialist breast radiologists. A consensus report was issued, with third reader arbitration in the event of disagreement. Readers were able to view the initial standard breast imaging but were only permitted to view and report one of the two study modalities per patient. Radiologists who read the MRI and CEM studies were subspecialty trained with between 5 and 25 years of experience in breast imaging. Our centre was one of the early adopters of CEM; readers completed a series of nine training cases supplied by the CEM vendor prior to study commencement.

Reference lesions were defined as those detected on standard imaging and confirmed to be malignant on needle biopsy. The largest invasive malignant lesion per patient on surgical histopathology formed the reference lesion. Multifocal disease was defined as the presence of more than one ipsilateral malignant lesion situated more than 5 mm from but within ≤4 cm of the reference lesion, usually within the same quadrant. ²⁰ Multicentric disease was defined as presence of more than one ipsilateral malignant lesion separated >4 cm, usually within a different quadrant. ²⁰

Lesions were viewed on Agfa picture archive and communication workstations (Agfa-Gevaert NV Mortsel, Belgium) with 5MPixel Barco (NYSE Euronext Brussels, Belgium:BAR) monitors measured to the nearest millimetre in three orthogonal dimensions using electronic callipers. For MRI, measurements were made on the first post-contrast series. For CEM, low energy (LE) and recombined images were considered together: suspicious microcalcifications were included in the estimated lesion size and the size of any non-enhancing malignant lesion was recorded from the LE images (Figure 1). In cases of suspected multifocal disease, the extent of suspicious mass and/or non-mass enhancement in that quadrant was measured.

Figure 1.

Breast imaging studies of one of the study participants. CEM: LE and RC craniocaudal (a, e) and mediolateral oblique views (b, f). Left lateral magnification view (c). Breast MRI: sagittal image (d) and axial first post-contrast MIP (g). A 71-year-old patient with a calcified left breast mass on screening mammography. Clinical examination revealed a 30 mm palpable mass. LE CEM images demonstrate an irregularly shaped spiculated mass (solid arrow) with associated microcalcifications (dotted arrow) in the left UOQ. On the RC images, the left UOQ lesion shows marked heterogeneous internal enhancement measuring 16 mm. Total lesion size on CEM including the associated calcification was 28 mm. Maximum lesion size on MRI was 23 mm. Final histopathology revealed an invasive duct carcinoma not otherwise specified (25 mm), with intermediate grade micropapillary DCIS measuring 45 mm. The total tumour size was 50 mm. Lesion size was significantly underestimated by both CEM and MRI and re-excision for close margins was needed. CEM, contrast-enhanced mammography; DCIS, ductal carcinoma in situ; LE, low energy; MIP, maximum intensity projection; RC, recombined; UOQ, upper outer quadrant.

To optimise radiological–pathological correlation, the pathologist was given a diagram showing the location of lesions detected on imaging (Figure 2) prior to tissue processing. Wide local excision specimens were received fresh and following overnight fixation in 10% neutral buffered formalin, the specimens were weighed, placed in a grid and radiographed. Coordinates of imaging lesions were identified from review of the diagram and specimen radiograph and pins inserted into the specimen through the holes in the grid based on the given coordinates (Figure 3). The specimens were measured and margins inked using the surgeon’s orientation sutures.

Figure 2.

Imaging-pathology case report form for the case shown in Figure 1. The location of the lesion detected on imaging and the imaging findings are listed for the pathologist.

This form was reviewed prior to tissue processing to ensure the blocks taken were focused on the lesion site.

Figure 3.

X-ray of the fixed surgical specimen taken in a perspex grid. The ill-defined calcified lesion (dotted arrow) is seen towards the centre of the specimen at co-ordinates G-I, 8–12. A localising iodine 125 seed is present within (solid arrow). Scattered microcalcifications are seen in the tissue surrounding the lesion. Pins are inserted through the holes in the grid to mark the coordinates of the lesion so that pathology sampling can be focused on this area.

The entire specimen was serially sectioned in 3–5 mm parallel slices in the sagittal plane and consecutively laid out, maintaining orientation. The lesion and margins were blocked (Figure 4) and processed using standard processing techniques. One 4 µm section was cut from each block and stained with haematoxylin and eosin (H & E).

Figure 4.

Pathologist’s block diagram. This provides a key for the sites where blocks were taken from the surgical specimen of the case shown in Figure 1. Invasive tumour is present in blocks 8, 9, 10, 11, 12, 16, 17, 18, 19, 20 (SLICES 6–10 inclusive).The greatest dimension of invasive tumour is measured medial to lateral as five slices (with each slice 5 mm in thickness) = 25 mm. DCIS is present in blocks 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 16, 17, 18, 19 (SLICES 1–9 inclusive).The greatest dimension of DCIS is measured medial to lateral as nine slices (with each slice 5 mm in thickness) = 45 mm. The whole tumour size (invasive+DCIS) is 50 mm (SLICES 1–10 inclusive). DCIS, ductal carcinoma in situ.

Following formalin fixation, mastectomies were serially sectioned in 5–10 mm parallel slices in the sagittal plane, with slices either held together by overlying skin (akin to a book-end) or consecutively laid out, maintaining orientation. If a lesion was not apparent, or the pre-operative imaging/core biopsy suggested a predominantly in-situ process, slices were radiographed prior to blocking. Region(s) of interest (including marker clips) and margins were blocked then processed using standard techniques. One 4 µm section was cut from each block and stained with H & E.

The first author and a study pathologist reviewed each case to ensure lesion concordance. The pathologist measured ¹ size of the largest invasive component (Path Invasive) and ² total lesion size including surrounding ductal carcinoma in situ (DCIS) and/or satellite malignant lesions within the same quadrant (Path Total).

Statistical analysis

An a priori power calculation determined a sample size of 60 would have 95% power to detect a difference of ± 5 mm between CEM and MRI, and a conclusion of equivalence to be drawn if the 90% CI around the difference between the two methods was above 5 mm or below 5 mm. This a priori 10 mm margin for concordance was selected as being likely to result in clear pathological margins while avoiding excess tissue removal.

Demographics and imaging findings were reported using descriptive statistics. Lesion size on imaging and pathology (Path Invasive and Path Total) were illustrated using scatter plots of raw data around the line of perfect agreement. Outliers for lesion size estimation (defined using Tukey’s rule as those values 1.5 times the InterQuartile Range above the 75th percentile or below the 25th percentile), were evaluated for underlying common factors.

Agreement between imaging and pathology was assessed initially by generating a Bland–Altman plot of the difference between imaging and pathology vs the mean of the two measures using raw data. Clinically acceptable limits of agreement (LOAs) were set a priori to be ± 10 mm. However, these graphs demonstrated an increasing spread or fanning-out of differences with increasing size, illustrating the common association between difference and magnitude which can lead to LOA too wide for smaller sizes and too narrow for larger sizes. As per Bland and Altman, ²¹ a log transformation was applied to remove this association and the graph regenerated. The mean difference and LOAs derived from the log transformed data were exponentiated to obtain values on the original scale. Once back transformed, these values represented relative (proportional) rather than absolute differences.

To quantify the degree of agreement between imaging and pathology, Lin’s concordance correlation coefficient (CCC) and 95% CI were calculated on log transformed lesion sizes. Pearson correlation is often inappropriately reported in agreement studies, however, this test only evaluates whether points lie on any straight line, not whether data conform to the line of equality (or the 45^o line). Lin’s concordance correlation coefficient corrects Pearson correlation for how far the line of best fit sits from the line of equality. A CCC of >0.75 was considered good agreement.

A paired sample t-test was used to assess the mean difference between tests. A 90% CI around the mean difference between imaging methods was calculated to determine equivalence (within ± 5 mm). The absolute value of the differences between imaging tests and pathology were dichotomised using a cut point of 10 mm (a priori acceptable margin of error) to calculate the proportion of lesions where imaging was concordant with pathology for descriptive purposes.

All tests of significance were two-sided and the level of significance used was p < 0.05. Analyses were undertaken using SAS v. 9.4 and Stata 16 (StataCorp. 2019. Stata Statistical Software: Release 16. College Station, TX: StataCorp LL).

Results

Of the 75 women who underwent CEM, MRI or both, data were available for review in 59. The mean time interval between the two contrast-enhanced studies was 4.6 days, sd = 3.2 days, range 1–15 days. The mean patient age was 56 years, range 35–77, sd = 11. Most females (42, 71%) were asymptomatic, with lesions detected on screening mammography, 13 (22%) had palpable masses and 4 other symptoms (nipple retraction, mastitis, rash, tenderness). Two (3%) had a previous breast cancer (ipsilateral in one and contralateral in the other), four (7%) had a family history of breast and one of ovarian cancer. None had a known gene mutation.

One-third of patients had mammographically dense breasts. Minimal-mild background parenchymal enhancement was noted on CEM and MRI in 44 (75%) and 41 (69%) patients respectively, and on both modalities in 38 (64%) patients.

The commonest morphology of the reference lesion on CEM and MRI was a mass (Table 1). A measurement was available for MRI in 57 patients and CEM in 58 patients. In two cases where the reference lesion did not enhance on CEM, estimated lesion size was based on the extent of residual suspicious microcalcification. The single reference lesion not visible on CEM was a 6 mm cluster of calcifications prior to diagnostic core biopsy and no residual calcifications were visible on the LE image. This lesion was visible on MRI (5 mm mass) and final pathology revealed a 9 mm Grade 1 invasive ductal carcinoma (IDC) with low grade DCIS. The two lesions not visible on MRI were a 12 mm stellate mass and a 6 mm mass prior to core biopsy. Both were visible on CEM as masses and shown to be a 7 mm Grade 2 IDC and a 6 mm Grade 2 IDC respectively, on final pathology.

Table 1.

Imaging findings (N = 59)

Characteristics of reference lesion used for size comparison	N (%)
Side
Right	34 (58)
Left	25
Mammographic breast density
Non-dense	40 (68)
Dense	19
Background parenchymal enhancement
CEM
Minimal-mild	44 (75)
Moderate-marked	15
MRI
Minimal-mild	41 (69)
Moderate-marked	18
Mammographic findings
Not visible	4 (7)
Mass	24 (41)
Mass with calcification	10 (1)
Architectural distorsion	11 (19)
Focal asymmetry	5 (8)
Calcification a	5 (1)
Ultrasound findings
Not performed	1 (2)
Not visible	2 (3)
Hypoechoic mass	55 (93)
Multiple masses	1 (2)
CEM findings	N (%)
Mass	46 (78)
Non-mass	6 (10)
Calcification	2 (3)
Mass and non-mass	4 (7)
Clip only	1 (2)
MRI findings
Mass	48 (82)
Non-mass	6 (10)
Mass and non-mass	3 (5)
Clip only	2 (3)

CEM, contrast-enhanced mammography.

The sum of column % differ from column totals due to rounding.

^aIncludes one case of asymmetrical density with calcifications.

Most of the reference lesions (Table 2) were invasive ductal carcinoma 40/59 (68%), associated with DCIS (31/40) in 78%. DCIS in adjacent tissue was noted in 25 (42%) cases and extensive intraduct component in 14 (24%). Multifocality was noted in 12 (20%) cases.

Table 2.

Histopathological features of the reference lesions

Histopathology variables	N (%)
Invasive tumour size
≤20 mm	36 (61)
>20 and <50 mm	16 (27)
≥50 mm	7(12)
Histopathological type
IDC with DCIS	31 (52)
ILC	14 (24)
IDC	9 (15)
Other invasive (mucinous, tubular)	3 (5)
Mixed IDC and ILC	2 (3)
Focality
Unifocal	47 (80)
Multifocal	12 (20)
DCIS with largest invasive lesion
EIC	14 (24)
DCIS in adjacent tissue	25 (42)
Grade of invasive tumour
Grade 1	9 (15.3)
Grade 2	34 (57.6)
Grade 3	16 (27.1)
Subtype of largest invasive lesion
Luminal A	4 (7)
Luminal B	47 (80)
Her2 positive	1 (2)
Triple negative	7
Total	59

EIC, extensive intraduct component; IDC, invasive ductal carcinoma; ILC, invasive lobular carcinoma.

The sum of column % differ from column totals due to rounding.

The median size of the reference lesions on pathology was 17 mm (invasive) and 27 mm (Path Total), (range 5–125 mm) (Table 3). MRI and CEM lesion size outliers failed to reveal any predominant features. Scatter plots for total and invasive pathology size vs MRI and CEM (Figure 5 and Figure 6), show observations further from the line of perfect agreement as their magnitude increases, and illustrate the leftward shift of observations when only the invasive component was measured on pathology. Multifocal cases (denoted by triangles) tended to produce substantial discrepancies when only the (largest) invasive component of the lesion was measured. Lin’s CCCs were comparable between modalities but not within modality (between total and invasive) with total lesion size producing better results (CCC = 0.75, 95% CI 0.6, 0.84 and CCC = 0.71, 95% CI 0.56, 0.82 for MRI and CEM respectively, Table 4). When the lower bound of the 95% CI is considered, none of these indices of agreement reflect an acceptable level of agreement.

Table 3.

Measurements of reference lesion on pathology, CEM and MRI

Lesion variables	Mean (sd)	Median [Q1,Q3]	Min,Max
Maximum lesion size pathology (total) n = 59	33.8 (23.2)	27 [16,50]	5,125
Maximum lesion size pathology (invasive) n = 59	25.2 (22.9)	17 [12,30]	6,125
Maximum lesion size CEM (n = 58)	33.3 (23.2)	29.5 [26.9,52]	6.5,100
Maximum lesion size MRI (n = 57)	35.3 (20.3)	29.0 [19,45]	5,80
Differences: CEM vs Pathology size (total)	1.2 (20.9)	0.75 [-7,7]	−79,50
CEM vs pathology size (invasive)	9.9 (25.8)	6.5 [0,22]	−79,91
MRI vs pathology (total)	−1.5 (16.4)	0 [-6,4]	−50,40
MRI vs pathology size (invasive)	7.4 (20.5)	2 [-2,14]	−50,71

CEM, contrast-enhanced mammography.

All measurements are in mm. N=59 pathology, N=58 CEM & N=57 MRI.

Figure 5.

Scatter diagram: lesion size on pathology (a) total (invasive plus in situ) and (b) invasive vs size on CEM. CEM, contrast-enhanced mammography. Lin’s concordance correlation coefficient is based on the log transformed data.

Figure 6.

Scatter diagram: lesion size on pathology (a) total (invasive plus in situ) and (b) invasive vs size on MRI.

Table 4.

Log transformed mean differences between lesion size as estimated with MRI and CEM and path total and invasive lesion sizes, with LOA

		Log scale		Exponentiated
		Average difference	95% LOA	Average difference	95% LOA	CCC1	95% CI CCC
Path total	MRI	−0.03	−0.96, 0.89	0.97	0.38, 2.4	0.75	(0.6, 0.84)
	CEM	0.04	−1.0, 1.1	1.04	0.37, 2.9	0.71	(0.56, 0.82)
Path invasive	MRI	0.34	−0.93, 1.6	1.41	0.4, 5	0.51	(0.31, 0.66)
	CEM	0.41	−0.98, 1.8	1.5	0.37, 6	0.45	(0.25, 0.61)

CEM, contrast-enhanced mammography; LOA, limit of agreement.

Lin’s correlation coefficient, with 95% confidence intervals.

^aConcordance correlation coefficient.

The Bland–Altman LOAs for each of the modalities vs pathology are illustrated in Figure 7 and Figure 8 and detailed with back transformations in Table 4. The mean differences for total lesion size are small (3% underestimated and 4% overestimated for MRI and CEM respectively) but bigger for invasive lesion size (41 and 50% overestimated for MRI and CEM respectively). However, the graphs show undesirably wide LOA for both modalities, neither demonstrating superior performance, given the similar widths of the LOA for both total and invasive lesion size. Back transformed limits indicate differences in total lesion size vary between a 60% underestimation to almost threefold overestimation (exponentiated LOA MRI: 0.38, 2.4, CEM: 0.37, 2.9). For invasive lesion size, the error margin ranges from a 60% under to sixfold over estimation (exponentiated LOA MRI:0.4, 5, CEM:0.37, 6).

Figure 7.

Bland–Altman plots for tumour size total (invasive plus in situ) vs lesion size on MRI and CEM. CEM, contrast-enhanced mammography.

Figure 8.

Bland–Altman plots for tumour size (invasive only) vs lesion size on MRI and CEM CEM, contrast-enhanced mammography.

No significant difference was detected between mean estimated lesion size using MRI and CEM (mean difference −2.48 mm (p = 0.26)). The 90% CI extends outside the interval that would demonstrate equivalence ( ± 5 mm).

Based on a ±10 mm acceptable margin of error, total size estimation by MRI was concordant with pathology in 36 (64%) cases compared to 32 (57%) for CEM, with both modalities concordant for 26 (46%) cases (Table 5). For 14 (25%) cases, neither modality was within acceptable error limits.

Table 5.

Comparison of proportion of agreement with total and invasive tumour size on pathology (within ± 10 mm) between modalities

Total lesion size		CEM vs pathology
		Difference >10 mm	Difference ≤10 mm	Total
MRI vs Pathology	Difference >10 mm	14 (25%)	6 (11%)	20 (36%)
	Difference ≤10 mm	10 (18%)	26 (46%)	36 (64%)
	Total	24 (43%)	32 (57%)	56 (100%)
Invasive lesion size		CEM vs pathology
		Difference >10 mm	Difference ≤10 mm	Total
MRI vs Pathology	Difference >10 mm	20 (36%)	3 (5%)	23 (41%)
	Difference ≤10 mm	6 (11%)	27 (48%)	33 (59%)
	Total	26 (46%)	30 (54%)	56 (100%)

CEM, contrast-enhanced mammography.

The sum of column % differ from column totals due to rounding.

Discussion

While the differences between mean estimated total lesion size using CEM or MRI are small, not statistically different, nor equivalent, the wide LOA observed in this study indicate that neither could be considered accurate for estimating lesion size. Small differences in mean estimated lesion size between CEM and MRI, and between either modality and Path Total have been reported, with variability as to whether this represents under- or overestimation. ^22–25 In our study, CEM slightly overestimated and MRI slightly underestimated total lesion size. Youn et al ²⁴ found CEM and MRI both underestimated lesion size by −0.97 mm (95% CI −3.7 to 1.76 mm) and −3.53 (95% CI −6.19 to −0.87 mm) respectively, whereas Luczynska et al ²⁵ reported CEM and MRI both overestimated lesion size by 1.7 mm, and 1.8 mm, respectively.

The 95% CI for Lin’s concordance coefficients found in our study, all failed to exclude the benchmark of 0.75 for good agreement between lesion size estimated with either CEM or MRI and pathology. Youn et al ²⁴ reported ICCs of 0.863 (95%CI 0.752–0.924) for CEM and 0.884 (95%CI 0.791–0.935) for MRI, indicating higher agreement with pathology than in our study, however, their lesion sizes were smaller, and had less variability.

The log transformation of sizes for the Bland–Altman plot created some difficulty for comparing the calculated LOA to the pre-specified acceptable margin of error. However, the range of relative errors (of between 3 and 6 fold in estimates of lesion size using CEM or MRI compared with pathology) suggested neither CEM nor MRI perform well. Other studies have reported 95% LOAs that exceed the same pre-specified 10 mm margin of error: Patel et et al ²⁶ −15.6 to 21.4 mm, Travieso-Aja et al ²⁷ −10.3 to 16.2 mm, Blum et al ²⁸ −18.8 to 19.48, Lobbes et al ²² −18.44 to 18.4 mm (CEM), and −11.46 to 15.71 (MRI). However, despite this, half of these authors concluded that the quality of tumour size measurement with CEM is “good.” These wide LOA emphasize the need for additional needle biopsies to confirm lesion size where significant changes in treatment planning could result.

Variability exists as to the error margins used to define “concordance” between lesion size estimated with imaging vs pathology, ranging from 5 mm, ^{24,27,29–31} 10 mm ^32,33 and 15 mm. ³⁴ Our results lie in between those reported in prior studies that have used a 10 mm error margin. McGuire et al ³³ compared lesion size measured on CEM with pathology and found a 47.1% concordance rate. Lobbes et al ²² found both CEM and MRI to be concordant with pathology in 84.5% of cases and concluded that the addition of MRI to CEM did not improve the accuracy of size prediction. Our findings differ: the addition of MRI to CEM would have resulted in acceptable lesion size estimates in a further 10 patients (18%).

The small number of patients in our study precluded analysis of patient and lesion related variables that could influence the ability of CEM and/or MRI to predict tumour size such as: breast and lesion size, ²⁴ lesion type on imaging (mass vs non-mass) ^34–36 and pathology (lobular subtype, multifocality, extensive intraduct component). ^27,29,36 Technical considerations include the mammographic view or MRI sequence on which the lesion is measured, ²⁴ and pathology processing and measurement techniques. ^37–39 Further studies are needed to identify patient/lesion subgroups most likely to benefit from use of one or other modality and the optimal measurement methods for each.

The strengths of this study include the in-depth prospectively guided imaging and pathology correlation, pragmatic measurement of whole tumour size and use of robust and appropriate statistical methods. Despite a plethora of literature ^22,40–42 explaining why the use of Pearson correlation in the assessment of agreement is inappropriate, use of correlation persists. The analysis adopted in this paper follows the guidelines of Bland and Altman, ²¹ including the use of log transformations to obtain more accurate LOA.

Our study has some limitations: we fell short of the sample size determined by our power analysis to address our primary aim (whether lesion size estimation with CEM was equivalent to MRI) and the planned sample size was insufficient for subgroup analyses.

While our inclusion criteria were wide, selection bias may have occurred during recruitment (e.g. tendency to offer the study to patients where it was thought contrast imaging may be beneficial based on knowledge of situations where MRI would be offered, e.g. larger tumours, dense breasts, mammographically occult lesions). However, given the intraindividual nature of this study, this would have affected CEM and MRI equally, and in practice, these are the cases for which MRI and CEM would be considered beneficial.

Conclusion

Although the mean differences between estimated lesion size using CEM or MRI vs pathology are small, the wide LOAs observed suggest that neither CEM nor MRI has sufficient precision to direct changes in planned treatment without needle biopsy confirmation.