Contrast-enhanced spectral mammography for breast cancer

Study size, design and location

Diagnostic accuracy study in 100 women having CESM examinations in a symptomatic clinic setting in the UK.

Intervention and comparator(s)

Intervention: SenoBright (GE Healthcare; mammography machine not reported). BI‑RADS scores using low-energy CESM images alone, and 3 weeks later using both low-energy and combined (low- and high-energy) CESM images.

Comparators: histopathology (if having biopsy or further surgery); breast MRI at baseline (if having neoadjuvant chemotherapy).

Key outcomes

The area under the receiver operating characteristic curve (AUROC) was significantly higher when reading the combined CESM image, compared with reading the low-energy image only (0.93 versus 0.83; p<0.025). The overall and individual radiologist, sensitivity and specificity also improved when the entire CESM image was reviewed (significance not reported). The difference in measured lesion size was significantly lower between CESM and MRI, compared with between the low-energy images and MRI (p<0.001). The difference in measured lesion size was also significantly lower between CESM and histology, compared with between the low-energy images alone and histology (p<0.0001). Interpretation of the low-energy images alone led to tumour size being underestimated. Overall, radiologists categorised the CESM images to be a 'useful aid to diagnosis' in 40% of cases, and a 'significant aid to diagnosis' in 35%.

Strengths and limitations

The majority of people had clinically suspicious abnormalities and therefore larger tumours. The authors note that CESM tends to overestimate the size of larger tumours, when compared with MRI, making the findings of this study less generalisable to those with smaller tumours. Although blinded, the 5 consultant radiologists reviewing the CESM image cases knew that the people had symptoms and knew the site of concern. However, by reading the low-energy, and then low-energy and combined, CESM images in separate sessions, it was possible to determine the sensitivity and specificity of both, with minimal bias.

Study size, design and location

Diagnostic accuracy study in 199 women having CESM after being recalled from the breast cancer screening programme in the Netherlands.

Intervention and comparator(s)

Intervention: Senographe Essential with SenoBright (GE Healthcare). BI‑RADS scores assigned using low-energy CESM images alone, and then upgraded or downgraded using both low-energy and combined (low- and high-energy) CESM images.

Comparator: true disease status (histology, further imaging, or discharge according to NHS clinical guidance for breast cancer screening assessment and national guidelines).

Key outcomes

Overall, diagnostic performance in detecting breast cancer improved when using the combined CESM image, compared with using just the low-energy CESM images; mean sensitivity increased from 93.0% to 96.9%, and mean specificity increased from 35.9% to 69.7%. The AUROC increased from 0.645 to 0.833 (p<0.0001). The inter-rater variability was excellent, with a kappa value of 0.89.

Strengths and limitations

A strength of the study is the use of a panel of 10 readers (7 radiologists and 3 residents) with different skill and experience levels, to demonstrate the ease with which CESM may be adopted. Some of the included cases were assessed for a previous study, but recall bias was minimised by anonymising the images and having them read over 1 year apart. The authors presented analysis to show that recall bias did not affect the results, and acknowledged that all cases were recalls from the screening programme, where readers were not blinded to the reason for referral. However, these potential limitations reflect clinical practice and the way in which CESM would be used.

Study size, design and location

Retrospective diagnostic accuracy study in 465 women having CESM in a breast cancer clinic in Spain.

Intervention and comparator(s)

Interventions: Senographe Essential with SenoBright (GE Healthcare; CESM); Senographe Essential (GE Healthcare; FFDM); LOGIQ E9 (ultrasound). FFDM, FFDM plus ultrasound, and CESM cases were evaluated in separate sessions at least 3 weeks apart.

Comparator: histopathology, or negative imaging over 18 to 24 months.

Key outcomes

The sensitivity, specificity, positive predictive value (PPV), and accuracy of CESM were all significantly higher than those of FFDM. Sensitivity increased from 82.5% to 92.3% (p<0.001), specificity increased from 68.6% to 86.0% (p<0.001), PPV increased from 65.5% to 93.0% (p<0.0001), and accuracy increased from 74.4% to 90.2% (p<0.005). The sensitivity, specificity, PPV and accuracy of CESM were also higher than those of FFDM plus ultrasound, which had sensitivity of 89.8% (p<0.05), specificity of 82.7% (p<0.05), PPV of 88.7% (p<0.01) and accuracy of 87.0% (p<0.05). Negative predictive value (NPV) improved on CESM, compared with both FFDM and FFDM plus ultrasound, but not significantly so. The AUROC increased significantly to 0.888 (95% confidence interval [CI] 0.855 to 0.917) for CESM, from 0.755 (95% CI 0.711 to 0.799) for FFDM (p<0.001), and from 0.861 (95% CI 0.826 to 0.896) for FFDM plus ultrasound (p<0.05).

Strengths and limitations

A key strength of this study is its comparatively large sample size and use of 3 radiologists, with discrepancies resolved by consensus of 2 additional radiologists. The majority of subjects had clinically suspicious abnormalities, and therefore larger tumours. The findings may therefore be less generalisable to those with smaller tumours. Selection bias may also be present, as the study was retrospective, with no predefined recruitment criteria. The retrospective nature of the study also limited the comparison of CESM with other prospectively used imaging tools, such as magnified views, DBT and targeted ultrasound.

Study size, design and location

Diagnostic accuracy study in 138 women with palpable breast masses, who had CESM in Israel.

Intervention and comparator(s)

Interventions: Senographe Essential (GE Healthcare; CESM); Acuson S2000 (Siemens; ultrasound). BI‑RADS scores assigned using the low-energy images, and the low-energy plus recombined CESM images.

Comparator: biopsy result or negative follow‑up imaging over 12 months.

Key outcomes

The sensitivity of CESM was 100%, compared with 94.7% for the low-energy images only. The specificity of the low-energy images was 83.5%, and was higher than the recombined CESM images at 73.4% (p=0.02). This was slightly higher than that of ultrasound (67.9%); however, this difference was not statistically significant (p=0.29). Two cases had their BI‑RADS score downgraded from the low-energy images using CESM images. Of 80 cases assigned a BI‑RADS score of between 1 and 3, targeted ultrasound identified 13 positive cases, which were all benign on biopsy. Using ultrasound, 7 cases were downgraded from their BI‑RADS score on CESM. Two cases of invasive ductal carcinoma were missed on the low-energy images, but detected on both CESM and ultrasound. The NPV for the low-energy images was 97.8%, and for both CESM and ultrasound was 100%. There was a strong correlation of r=0.80 (p<0.001) between the contrast-enhanced area on CESM, and the tumour size measured by pathology, in the 22 people who had this data available.

Strengths and limitations

This was a small, retrospective, single-centre study. The authors report a high prevalence of cancer cases in a population of women with palpable lesions, and note that this may be because a large number of cases with benign imaging were excluded with a lack of appropriate follow up or histopathology. This may have influenced the NPV reported. Most of the women in the study had dense breast tissue, and the results therefore may not be generalisable to those with less dense breasts. All images were read by a single radiologist, with no blinding between images, so this may have introduced bias.

Study size, design and location

Diagnostic accuracy study in 118 female patients with unresolved or suspicious findings after FFDM plus ultrasound, having CESM and 3T MRI, in a radiology department in Italy.

Intervention and comparator(s)

Intervention: Senographe Essential with SenoBright (GE Healthcare; CESM); Discovery MR750w (GE Healthcare; MRI). BI‑RADS score assigned for FFDM plus ultrasound, and MRI images, followed by all images including CESM, 1 month later.

Comparator: Histopathology or cytology.

Key outcomes

The overall reported diagnostic accuracy of CESM was the same as that of FFDM, at 93%. Its sensitivity was 100%, the highest reported, and specificity was 50% (only MRI was lower at 47%). The PPV of CESM was second highest at 92%, with FFDM at 94%, and NPV was the highest overall at 100%. On a per-lesion basis, the authors reported a concordance between CESM and the reference standard of 63% (p<0.001). Concordance between both FFDM and CESM, and FFDM plus ultrasound and CESM was 81% (p<0.001). Concordance between MRI and CESM was 73% (p<0.001). Concordance between the 2 radiologists for CESM, MRI and ultrasound images was 100% (p<0.001), and for FFDM images alone was 53% (p=0.024).

Strengths and limitations

The 2 radiologists reading the images were experienced, but not with CESM, and received training on this beforehand. Although image readings were separated by about a month, the authors note that recall bias was still possible. Authors acknowledge that background parenchymal enhancement is known to contribute to false-positive results, but did not study it. The prospective nature of the study decreases the likelihood of selection bias, and results were reported transparently as per patient, and per lesion.

Study size, design and location

Diagnostic accuracy study in 100 consecutive patients having CESM, with breast lesions identified by clinical examination, mammography or ultrasound, in Italy.

Intervention and comparator(s)

Intervention: Selenia mammography system (Hologic; CESM, DBT, and FFDM synthesised from DBT); MAGNETOM Symphony (Siemens; MRI).

Comparator: pathology from surgical specimen or core needle biopsy.

Key outcomes
The sensitivity of CESM plus DBT was the highest of all mammography techniques, at 93.2%, followed by CESM alone at 87.4%. CESM plus DBT, and CESM alone, had the lowest specificity, at 76.5% and 80.9% respectively. The highest specificity was 83.8% for DBT alone. The AUROC for CESM plus DBT was highest, at 0.905, followed by CESM alone, at 0.883. The accuracies of CESM plus DBT and CESM alone were also highest at 86.5% and 84.8% respectively. Diagnostic accuracy was also reported including contrast-enhanced MRI, and dynamic contrast-enhanced MRI (DCE‑MRI), in a subset of people. Sensitivity and specificity were comparable across individual imaging techniques, but statistically different between DCE‑MRI and CESM (p=0.035). The highest correlation between imaging lesion size and pathologic size was obtained by DCE‑MRI (r=0.811), and was lower for CESM (r=0.722) and DBT (r=0.684). Both CESM and MRI techniques overestimated lesion size when compared with pathology.

Strengths and limitations

Images were interpreted as a consensus by 2 radiologists (from a pool of 8) who were blinded to results of other imaging techniques. However, as the FFDM images were synthesised from the DBT images, the study was limited by these 2 techniques not being independent. The authors also acknowledged that there is no dedicated BI‑RADS lexicon or other classification system for CESM images, but that the use of the existing BI‑RADS system is common. It was not clearly reported why the diagnostic performances of MRI techniques were reported in a subgroup, or how many people were in the subgroup.

Study size, design and location

Retrospective cohort study of 326 patients with biopsy proven breast cancer, having CESM for preoperative staging, in Italy.

Intervention and comparator(s)

Interventions: Selenia Dimensions (Hologic; CESM, FFDM, DBT); ESAOTE 70XVG (MyLab; ultrasound).

Comparator: histopathology.

Key outcomes

Results of CESM changed the type of surgery planned in 60 of 326 (18.4%) people: more extensive breast-conserving surgery (BCS) was planned in 19 people, mastectomy was used instead of BCS in 30 people, mastectomy was downgraded to BCS in 2 people, and 9 people had either BCS or mastectomy on the other breast. None of the conversions to mastectomy were because of false-positive CESM findings. However, there were 6 false-positive cases. In 3 cases, the CESM revealed a multifocal tumour, which histology confirmed to be a single lesion. These people had more extensive BCS than was originally planned. Three people had BCS on the other breast, for B3 lesions (for example, atypical ductal hyperplasia). There were 4 false-negative cases, with positive tumour margins, who were operated on again. Two had more extensive BCS, and 2 had mastectomy. Overall, the accuracy of CESM for preoperative staging was 97%. Sensitivity was 93%, specificity was 98%, PPV was 90% and NPV was 98%. The diagnostic accuracy of CESM was significantly better in people with palpable lesions compared with those with non-palpable lesions.

Strengths and limitations

CESM images were reviewed by 2 radiologists and all other imaging reviewed by 2 different radiologists. A fifth radiologist compared reports to determine whether CESM changed the type of surgery planned. As the study included only biopsy-proven breast cancer cases, it may not reflect true clinical use of CESM, and results may not be generalisable to other populations. The retrospective nature of the study also limits the available data. The authors also note that comparing CESM with MRI would have been useful, to understand which subgroups of people would benefit from having preoperative CESM rather than MRI.

Study size, design and location

Retrospective cohort, and diagnostic accuracy, study of 147 patients having CESM for suspicious calcifications, after being recalled from the national breast cancer screening programme in the Netherlands.

Intervention and comparator(s)

Intervention: Senographe Essential with SenoBright (GE Healthcare; CESM). BI‑RADS score and measurement of lesions assigned using low-energy CESM images, followed by using recombined CESM images to change their assessment if needed.

Comparator: histopathology.

Key outcomes

CESM had higher sensitivity, 93.8% (95% CI 85.0% to 98.3%), than the low-energy images alone, 90.8% (95% CI 81.0% to 96.6%). NPV was also higher for CESM: 88.2% (95% CI 73.6% to 95.3%), compared with low-energy images, 84.2% (95% CI 70.4% to 92.3%). The low-energy images had slightly higher specificity of 39.0% (95% CI 28.4% to 50.4%), and PPV of 54.1% (95% CI 49.4% to 58.8%). Specificity for CESM was 36.6% (95% CI 26.2% to 48.0%), and PPV was 54.0% (95% CI 49.6% to 58.3%). In 51 of 65 (78.4%) cases, the diameter of the lesion was incorrectly assessed by CESM, with a mean difference of 4.23 mm (95% limit of agreement -32 mm to 60 mm). Using only the low-energy images, and clinical information, the surgeons recommended BCS in 58 of 65 (89.2%) cases, and primary mastectomy in 7 of 65 (10.8%) cases. Using the CESM images, BCS was recommended in 55 of 65 (84.6%) cases. Decisions were discordant in 7 of 65 (10.8%) cases, and differences were not statistically significant (p=0.453). Of these discordant cases, 5 were upgraded from BCS to mastectomy, and 2 were downgraded from mastectomy to BCS.

Strengths and limitations

Two breast surgeons recommended surgery using the low-energy images and reports, then by using recombined CESM images and reports 8 weeks later. Because of the retrospective design, the surgeons planning treatment were not able to examine the patient or explore their preferences, and treatment decisions were made in a simulated environment with no follow‑up data. Using only a single radiologist, it was not possible to determine inter-observer variation.

Study size, design and location

Diagnostic accuracy study in 84 patients diagnosed with invasive carcinoma or ductal carcinoma in situ, having CESM and CE-MRI for preoperative evaluation, in South Korea.

Intervention and comparator(s)

Interventions: Selenia Dimensions with I‑View (Hologic; CESM); Lorad Selenia (Hologic; FFDM); Achieva (Philips; CE‑MRI); iU22 (Philips; ultrasound, unexpected suspicious CE‑MRI or CESM findings only).

Comparator: histological findings on biopsy.

Key outcomes

The sensitivity in detecting secondary cancer (83.9%) and detecting occult cancer (83.3%) were identical for CESM and CE‑MRI. Specificity, PPV, NPV, and accuracy were all higher for CESM than CE‑MRI, in both secondary and occult cancers, although this was not statistically significant. Six index cancers, and 6 secondary cancers were missed on CESM. Three of these were detected on CE‑MRI, 2 were detected on ultrasound, and 1 had been detected on initial FFDM. CE‑MRI missed 4 index cancers, 2 of which were not enhanced on CESM either. The third lesion was a multifocal invasive lobular carcinoma, described as background parenchymal enhancement on CE‑MRI, and an enhancing lesion on CESM. The fourth lesion was not detected on either CESM or CE‑MRI, but had been detected on ultrasound and initial FFDM.

Surgical management was changed in 26 of 84 people because of CE‑MRI findings, and in 25 of 84 people because of CESM findings (p=0.610). There was no significant difference between changes to surgical management as a result of false-positive findings, with 9 of 26 having excisional biopsy as a result of CESM findings, and 11 of 25 having excisional biopsy as a result of CE‑MRI findings (p=0.782).

Strengths and limitations

Selection bias is likely as this is a small study, in people with known cancer. This may also limit the generalisability of the results to people without known cancer, who have CESM for diagnosis. BI‑RADS-like score (on a scale of 1 to 5) was assigned to CE‑MRI images by 2 radiologists, and to CESM images by 2 different radiologists. Seven lesions were classified as benign on ultrasound, and no biopsy was taken. This may have influenced the diagnostic accuracies reported; however, the authors note that no lesions were upgraded to suspicious within 1 year of imaging follow up. Most of the false-negative results came from cases of ductal carcinoma in situ, or invasive lobular carcinoma, suggesting further investigation is needed in these subgroups.

Study size, design and location

Retrospective cohort study in 132 patients having preoperative staging using CESM, ultrasound and clinical examination, in a cancer centre in France.

Intervention and comparator(s)

Interventions: Selenia Dimensions with I‑View (Hologic; CESM); Logiq (GE Healthcare; ultrasound).

Comparator: histology (only in those having biopsy).

Key outcomes

CESM revealed additional image enhancement in 44 of 132 people, and 42 had a further biopsy (2 refused). Overall, 24 of 42 biopsies were positive, and histology results found them to be related to the same primary tumour. For 24 people (18.5%, [95% CI 12.2% to 26.2%]), planned surgery was changed because of CESM results alone. Overall, including changes not because of CESM alone, 8 people had surgery cancelled for neoadjuvant chemotherapy, 13 people (with 14 lesions) had mastectomy instead of lumpectomy, 1 person had oncoplasty instead of lumpectomy, and 2 people had a contralateral procedure. Excellent agreement was found between lesion size measured on CESM and histology, with an intraclass correlation index of 0.82 (95% CI 0.75 to 0.88). However, CESM overestimated lesion size by a mean difference of 4.25 mm, which was statistically significant (p<0.001).

Strengths and limitations

As the study was retrospective and non-comparative, the true clinical benefit of preoperative staging may not be known. The authors note that the size measurements may not be comparable, because the axis on which they were taken was not predefined. The correlation index may also have been altered because the measurements focused only on infiltrating tumours. Low-energy images were assessed as FFDM images, and an MRI BI‑RADS score was assigned to combined CESM images by 2 radiologists. Discordant results were reviewed by a breast radiology board. Diagnostic accuracy was therefore not assessed for tumours combining infiltrating, and in situ, tumours. Primary radiological investigations were conducted outside of the study centre, so quality of referrals cannot be guaranteed. However, this reflects clinical practice.

One additional study was also identified which explored patient preference and tolerance for CESM and contrast-enhanced MRI (CE‑MRI) in 49 people having both imaging techniques for breast cancer staging (Hobbs et al. 2015). CESM was preferred to CE‑MRI (p<0.001), and CE‑MRI was associated with higher rates of anxiety during the procedure than CESM (p=0.009). People preferred CE‑MRI for the contrast injection (p=0.003), and found the breast compression for CE‑MRI more tolerable (p=0.001). However, most people reported these factors as being 'comfortable' or 'neutral' during both CESM and CE‑MRI. The most common reasons for favouring CESM were the speed of the investigation, comfort (standing or lying), and the noise of the procedure.