Abstract
Objectives:
To assess whether NHS breast screening programme (NHSBSP) mammogram readers could effectively interpret first post-contrast acquisition subtracted (FAST) MRI, for intended use in screening for breast cancer.
Methods:
Eight NHSBSP mammogram readers from a single centre (four who also read breast MRI (Group 1) and four who do not (Group 2)) were given structured FAST MRI reader training (median 4 h: 32 min). They then prospectively interpreted 125 FAST MRIs (250 breasts: 194 normal and 56 cancer) comprising a consecutive series of screening MRIs enriched with additional cancer cases from 2015, providing 2000 interpretations. Readers were blinded to other readers’ opinions and to clinical information. Categorisation followed the NHSBSP MRI reporting categorisation, with categories 4 and 5 considered indicative of cancer. Diagnostic accuracy (reference standard: histology or 2 years’ follow-up) and agreement between readers were determined.
Results:
The accuracy achieved by Group 2 (847/1000 (85%; 95% confidence interval (CI) 82–87%)) was 5% less than that of Group 1 (898/1000 (90%; 95% CI 88–92)). Good inter-reader agreement was seen between both Group 1 readers (κ = 0.66; 95% CI 0.61–0.71) and Group 2 readers (κ = 0.63; 95% CI 0.58–0.68). The median time taken to interpret each FAST MRI was Group 1: 34 s (range 3–351) and Group 2: 77 s (range 11–321).
Conclusion:
Brief structured training enabled multiprofessional mammogram readers to achieve similar accuracy at FAST MRI interpretation to consultant radiologists experienced at breast MRI interpretation.
Advances in knowledge:
FAST MRI could be feasible from a training-the-workforce perspective for screening within NHSBSP.
Introduction
Mammographic screening programmes result in both over diagnosis and under diagnosis of breast cancer.1–3 Under diagnosis leads to cancers presenting symptomatically between screening visits (interval cancers), and to continued presentation of Stage 2 or greater breast cancers.4 Although MRI is the most sensitive method to detect breast cancer, currently only women classified as high risk (>30% lifetime risk) are offered screening MRI in the UK.5 However, in the future, personalised screening could enable larger numbers of women to be offered different screening regimes, each incorporating different imaging modalities, according to their level of risk.6,7
Finding breast cancer early saves lives,8,9 and there is therefore a need to develop cost-effective imaging tests that will benefit women at risk of breast cancer by finding significant disease early.10 First post-contrast acquisition subtracted (FAST) MRI is a type of abbreviated breast MRI and has been suggested as such a screening test since proof of concept studies suggest it could offer accuracy of breast cancer detection almost equivalent to full protocol breast MRI with speed of acquisition and reporting that approaches that of mammography.11–13 This technique might especially benefit women with dense breasts since cancers obscured by dense tissue on mammograms are often visible on MRI.14
Mammographic population screening for breast cancer necessitates a high volume of throughput of images for interpretation, and in many countries radiologists who interpret mammograms do not necessarily also interpret breast MRI, a less frequently performed and more complex modality. In the UK, skills-mix has enabled professionals other than radiologists, including advanced practitioner radiographers, to learn to interpret screening mammograms and studies demonstrate their adequacy of performance at this task.15,16 The present study is the first to look at the capability of mammogram readers in the context of FAST MRI interpretation. FAST MRI was the new technology chosen for this study because it is the MR sequence common to most reported types of abbreviated protocol, with the shortest reported interpretation time and simplest format.11 In the design of this study, the authors postulated that its simplicity and the similarities of display between it and mammographic modalities would be likely to enable mammogram readers to easily and quickly learn FAST MRI interpretation. In the current study, we chose to further simplify the display protocol, and unlike in Kuhl’s original description of FAST MRI,11 the unsubtracted images were not made available for interpretation by our readers. This simplification of the display protocol was intended to ease training of mammogram readers, unfamiliar with multiple sequences.
The aim of this study was to explore whether NHS breast screening programme (NHSBSP) mammogram readers can learn to effectively interpret FAST MRI with less than one day’s additional training and to match the capabilities of expert breast MRI readers at this task in terms of accuracy and speed of interpretation.
Methods and materials
Research Ethics Committee and Health Research Authority approvals were obtained (references: REC 17/SW/0142, IRAS 219332). Informed consent was obtained from all study participants.
Study design
Prospective, blinded interpretation by multiple readers of an enriched dataset with known outcome.
Test set of images
A set of FAST MRI examinations was created by copying and reducing (post-processing) breast MRI scans acquired at a single centre during 2015. MRIs were acquired exclusively during a single year to standardise scan quality. A consecutive sample of all breast MRIs performed as screening for females at high risk of breast cancer according to NICE guidelines (>30% lifetime risk from the age of 20 years)5 were included (72 MRIs), enriched with 54 symptomatic MRIs including one with bilateral cancer and 53 with unilateral cancer (Figure 1). Two of the screening MRIs also showed a unilateral cancer. The clinical indications given for the 125 MRI scans included in the dataset are shown in Table 1. To increase task difficulty, only cancers smaller than 25 mm were included, as measured on the original full MRI report. All cancer cases were confirmed by histological analysis of breast tissue. The histology of the 56 cancers in the dataset is shown in Table 2. Breasts were classified as not having cancer through either negative interpretation of full MRI by at least one fully trained radiologist or uncertain interpretation of full MRI and negative ultrasound (± biopsy), and at least 2 years’ follow up data without cancer.
Figure 1. .
Flow diagram to illustrate FAST MRI dataset.
Table 1.
Clinical indication for MRI scans of the dataset
| Clinical indication | Number of scans |
|---|---|
| High-risk screening | 72a |
| Lobular carcinoma | 25 |
| Cancer occult on mammogram | 18 |
| Size discrepant: clinical/imaging modalities | 7 |
| New symptomatic cancer diagnosis in patient found to be a gene carrier | 1 |
| Other | 2b |
| Total | 125 |
includes one patient with unilateral invasive cancer and one with unilateral high-grade (HG) DCIS and 70 cases without cancer.
includes one patient with a small invasive cancer having MRI as part of a research study protocol and one with a new cancer diagnosis, performed for problem-solving in the contralateral breast.
Table 2.
Histology of cancers in dataset
| Characteristic | Number | % |
| Total number of cancers | 56 | |
| Tumour type | ||
| No Specific Type (Ductal) | 26 | 46 |
| Lobular | 25 | 44 |
| Mixed | 2 | 4 |
| Mucinous | 1 | 2 |
| Ductal carcinoma in situ | 2 | 4 |
| Grade and receptors (54 invasive cancers) | ||
| Histological grade | ||
| 1 | 5 | 9 |
| 2 | 39 | 72 |
| 3 | 10 | 19 |
| Receptor status | ||
| ER +and/or PR+, HER2 negative, Ki67 ≤10% | 32 | 59 |
| ER +and/or PR+, HER2 negative, Ki67 >10% | 12 | 22 |
| Triple positive | 3 | 6 |
| ER negative, HER2+ | 0 | 0 |
| Triple negative | 5 | 9 |
| ER+, PR negative, HER2+ | 2 | 4 |
| Ki67 | ||
| Median (range) | 10 (0–60) |
The median age of the women imaged was 40 years (range 28–61) in the high-risk screening population and 58 (35–83) for the women with a new cancer diagnosis.
Breast MRI protocol
All MRI examinations in the dataset were originally acquired on either a Philips (Amsterdam, Netherlands) Ingenia 1.5T or a Philips Ingenia 3T scanner. The breast coils used were dStream Breast seven-channel coils. The paramagnetic contrast agent used was gadobutrol 1.0 mmol ml−1 and the dose administered was 0.1 ml gadobutrol per kg body weight. The dynamic sequence used (from which the dataset’s FAST MRI images were obtained through post-processing) was dyn_eTHRIVE (Axial 3D T1 fast field echo (FFE), TR/TE 5.1/2.8 with 10 degree flip angle and SPAIR Power two fat suppression). Post-contrast scan commenced contemporaneous with the commencement of contrast injection (average duration 1.08 minutes)). Since the images used in the current study were originally acquired in 2015 and then later reprocessed and anonymised for the study, the acquisition protocol conformed to our own centre’s standard. This differed from Kuhl’s description of FAST MRI11 as follows:
The breasts were not compressed during MR acquisition.
The T1 images of the dynamic study that were used to form the subtracted images were fat-suppressed (dyn_eTHRIVE).
The MRI scans performed for a screening indication were performed during day 6–16 of the woman’s menstrual cycle, but those performed post cancer diagnosis were performed promptly without reference to the woman’s menstrual cycle.
The MRI studies were copied, anonymised and allocated study identifiers chronologically for the date they were acquired, and as a consequence normal and abnormal scans were presented to the readers in an unpredictable order. They were then reduced to comprise simply those MR sequences that would have been obtained if they had originally been acquired as a FAST MRI, displayed as an axial maximum intensity projection image (MIP), and also as a stack of axial slices (slice stack) of the first post-contrast-subtracted images from the dynamic series of the breast MRI examination.11 This process was performed by two of the research team who were not subsequently part of either of the two reading groups. These subtracted images alone comprised the FAST MRI scans interpreted by the readers.
Study participants
Eight radiology practitioners from a single centre were recruited as image readers; all practised as NHSBSP mammogram readers. These eight practitioners comprised four readers who also practised as NHSBSP breast MRI readers (Group 1) and four readers who did not read breast MRI in their normal clinical practice (Group 2). Group 1 were all consultant radiologists between 4 and 11 years’ experience of reading both mammograms and breast MRIs. Individual members of Group 1 read between 5000 and 6500 mammograms and between 100 and 225 breast MRIs each year. Group 2 comprised a consultant radiologist, a consultant radiographer and two film reading radiographers, with individual experience of reading mammograms ranging from 2 to 28 years and each member of Group 2 read between 5000 and 18,000 mammograms each year.
Standardised training
All eight readers were trained to read FAST MRI using a structured training package17 including one-to-one training and interactive small group presentation components. All readers were then offered an additional one-to-one teaching session if they felt they would benefit. The examples of FAST MRI shown to the readers during the training were not from the subsequently interpreted dataset.
The structured training delivered during the present study to all-but-one reader took half a day to deliver. One reader (Reader Identifier 2.2) requested further training, and this was delivered as a second one-to-one session, lasting 1 h, when the reading task was four-fifths completed.
Once trained, each reader completed a data-form for each FAST MRI case within the dataset, blinded to the identity of the patient, clinical history, original full MRI, all other imaging including mammograms, ultrasounds and previous breast MRI examinations, the outcome (cancer or no cancer) and to the opinions and completed data forms of the other readers. The FAST MRI case studies were read in batches of up to 25 because the workstation that displayed the images had a limited capacity for data storage and was also required for routine clinical work.
Classification system
The readers were instructed to classify each breast of each FAST MRI examination in accordance with a modified version of the MRI screening reporting categories of classification outlined in the 2012 NHSBSP guidelines for screening higher risk women, where MRI1 and MRI2 indicate normal and benign, MRI3 indicates an indeterminate classification, and MRI4 and MRI5 indicate suspicious and definitely malignant appearances, respectively.18 The recommended MRI screening reporting categories were modified because FAST MRI differs from the full diagnostic protocol in providing limited morphological information and only a single time point from the dynamic scan.
Statistical analysis
A per breast analysis of the frequency of MRI classifications against the true outcome was obtained overall and for each reader. The overall accuracy, sensitivity, specificity, false-positive and false-negative rates (with total reads as denominator) and the positive and negative predictive values of the readers’ MRI classification with the true outcome were calculated. Differences in accuracy, sensitivity and specificity across reader groups were analysed using a multilevel generalised mixed model to account for multiple readers per case. The inter-reader variability and the agreement between readers and the true outcome were assessed using Cohen’s κ coefficient to account for the probability of the agreement occurring by chance. A κ-statistic value of >0.60 would represent good agreement. MRI classifications of 4 and 5 were considered indicative of cancer, and classifications of 1–3 considered a normal result. A sensitivity analysis was performed whereby those with an indeterminate classification MRI3 were classified as cancer. The FAST MRI interpretation times were compared across reader groups using a Wilcoxon signed-rank test and the reader training times compared using a Wilcoxon rank-sum test.
Results
All eight readers completed the reading task of 125 cases (250 breasts). Per breast analysis comparing the readers’ MRI classification with the true outcome (cancer or normal) showed an overall concordance with the true outcome of 87% (95% CI 86–89%; 1745/2000 reads), with 393 (88%) cancers correctly identified and 1352 (87%) normal results correctly identified (Table 3) (Figure 2). The overall sensitivity was 88% (95% CI 84–91%) and specificity 87% (95% CI 85–89%) (Table 4). The agreement between all readers and the true outcome demonstrated good concordance with a κ of 0.69 (95% CI 0.65–0.72).
Table 3.
Comparison of readers’ MRI classification against true outcome
| True result | |||
|---|---|---|---|
| Cancer | Normal | Total | |
| All Readers’ MRI classification | |||
| MRI four or 5 | 393 (88%) | 200 (13%) | 593 (30%) |
| MRI 3 | 26 (6%) | 221 (14%) | 247 (12%) |
| MRI one or 2 | 29 (6%) | 1131 (73%) | 1160 (58%) |
| Total | 448 (22%) | 1552 (78%) | 2000 |
| All Group 1 Readers’ MRI classification | |||
| MRI four or 5 | 193 (86%) | 71 (9%) | 264 (26%) |
| MRI 3 | 17 (8%) | 138 (18%) | 155 (16%) |
| MRI one or 2 | 14 (6%) | 567 (73%) | 581 (58%) |
| Total | 224 (22%) | 776 (78%) | 1000 |
| All Group 2 Readers’ MRI classification | |||
| MRI four or 5 | 200 (89%) | 129 (17%) | 329 (33%) |
| MRI 3 | 9 (4%) | 83 (11%) | 92 (9%) |
| MRI one or 2 | 15 (7%) | 564 (73%) | 579 (58%) |
| Total | 224 (22%) | 776 (78%) | 1000 |
Figure 2. .
MRI classifications of each of the 250 breasts by the eight image readers. (a) 56 breasts with cancer. (b) 194 normal breasts.
Table 4.
Accuracy of the first post-contrast acquisition subtracted (FAST) MRI readers against the true outcome
| Total | Group 1 | Group 2 | |
|---|---|---|---|
| Measure | |||
| MRI classifications 4 and 5 considered as cancer | |||
| Concordance | 1745/2000 (87%) | 898/1000 (90%) | 847/1000 (85%) |
| True-positive rate (Sensitivity) | 393/448 (88%) | 193/224 (86%) | 200/224 (89%) |
| True-negative rate (Specificity) | 1352/1552 (87%) | 705/776 (91%) | 647/776 (83%) |
| False-positive rate | 200/1552 (13%) | 71/776 (9%) | 129/776 (17%) |
| False-negative rate | 55/448 (12%) | 31/224 (14%) | 24/224 (11%) |
| Positive predictive value | 393/593 (66%) | 193/264 (73%) | 200/329 (61%) |
| Negative predictive value | 1352/1407 (96%) | 705/736 (96%) | 647/671 (96%) |
| MRI Classifications 3, 4 and 5 considered as cancer | |||
| Concordance | 1550/2000 (78%) | 777/1000 (78%) | 773/1000 (77%) |
| True-positive rate (Sensitivity) | 419/448 (94%) | 210/224 (94%) | 209/224 (93%) |
| True-negative rate (Specificity) | 1131/1552 (73%) | 567/776 (73%) | 564/776 (73%) |
| False-positive rate | 421/1552 (27%) | 209/776 (27%) | 212/776 (27%) |
| False-negative rate | 29/448 (6%) | 14/224 (6%) | 15/224 (7%) |
| Positive predictive value | 419/840 (50%) | 210/419 (50%) | 209/421 (50%) |
| Negative predictive value | 1131/1160 (98%) | 567/581 (98%) | 564/579 (97%) |
The concordance with the true outcome achieved by Group 2 (847/1000 (85%; 95% CI 82–87)) was 5% less than that achieved by Group 1 (898/1000 (90%; CI 88–92)) (Table 5), a small but significant difference (p < 0.0001). Results for readers in Group 2 showed a non-significant trend towards higher sensitivity (89%; 95% CI 85–93) but significantly lower specificity (83%; 95% CI 81–86) than for readers in Group 1 (sensitivity 86%; 95% CI 81–90, p = 0.23; specificity 91%; 95% CI 89–93, p < 0.0001). Good inter-reader agreement was observed between the Group 1 readers (κ = 0.66; 95% CI 0.61–0.71) and also between the Group 2 readers (κ = 0.63; 95% CI 0.58–0.68) (Table 5).
Table 5.
Individual readers’ results when MRI classifications 4 and 5 are considered as cancers
| True outcome | ||||
|---|---|---|---|---|
| Number correctly identified (%) |
Cancer (N = 56) |
Normal (N = 194) |
Total concordance | κ (95% CI) |
| Group 1 Individual Reader | ||||
| 1.1 | 44 (79%) | 180 (93%) | 224 (90%) | |
| 1.2 | 47 (84%) | 186 (96%) | 233 (93%) | |
| 1.3 | 49 (88%) | 166 (86%) | 215 (86%) | |
| 1.4 | 53 (95%) | 173 (89%) | 226 (90%) | |
| Total Group 1 | 193 (86%) | 705 (91%) | 898 (90%) | 0.66 (0.61–0.71) |
| Group 2 Individual Readers | ||||
| 2.1 | 51 (91%) | 140 (72%) | 191 (76%) | |
| 2.2 | 49 (88%) | 175 (90%) | 224 (90%) | |
| 2.3 | 48 (86%) | 165 (85%) | 213 (85%) | |
| 2.4 | 52 (93%) | 167 (78%) | 219 (88%) | |
| Total Group 2 | 200 (89%) | 647 (83%) | 847 (85%) | 0.63 (0.58–0.68) |
A sensitivity analysis considering classifications MRI3–MRI5 to indicate cancer gave an overall accuracy of 78% (1550/2000; 95% CI 76–79; Table 4).
Time taken to report
The time taken for the individual readers to interpret each FAST MRI examination was significantly less for Group 1 (median 34 s, range 3–351 s) than for Group 2 (77 s, 11–321 s, p < 0.0001).
Time taken to train
It took less than 1 day of structured training to train the readers in this study. The training time between groups was not significantly different (median training time 4:01 (hours:minutes) and range 3:25–4:42 for Group 1 and 4:55 (4:25–6:04) for Group 2, p = 0.11).
Discussion
Following less than 1 day’s structured training, the NHSBSP mammogram readers in this study achieved a good agreement between their interpretation of FAST MRI and the clinically proven outcome in this enriched dataset. The group with no previous experience of breast MRI performed just 5% less well in terms of overall accuracy in comparison with the group of expert breast MRI readers. Looking at the individual accuracy for the readers within each group (Table 4), there is some overlap between groups: the lowest accuracy in Group 1 of 86% is lower than the accuracy of the two best performing readers in Group 2.
The median time taken to interpret FAST MRI by individual readers in Group 1 ranged from 27 to 44 s with a median for the whole group of 34 s. This is similar to the 28 s taken by expert readers in Kuhl’s original proof of concept publication.11 For Group 2, the range was 57–144 with a median for the whole group of 77 s, double that of Group 1. This significant difference between groups may indicate that following training, new readers of FAST MRI may take time to achieve the same reading speed as experienced MRI readers.
Interestingly, the readers achieved a better match in terms of overall accuracy when MRI4 and MRI5 were taken to indicate cancer and MRI1–MRI3 to indicate normal than if MRI3 was to be included with MRI4 and MRI5 to indicate cancer. This is unsurprising because the use of the “3” category by a reader indicates uncertainty about whether or not a cancer is present. It has previously been reported that in mammogram reporting 10–30% of breasts classified as “3” demonstrate a true cancer,19,20 and our study’s results fall within that range; of the 247 “3” classifications given by our readers, only 26 (11%) were cancers. Mammographic screening programmes recall “3” classifications for further imaging. This is necessary because mammographic techniques selectively pick up low-grade, biologically indolent cancers21,22 so that choosing to ignore indeterminate abnormalities in mammographic screening is risky because small, high-grade, biologically aggressive cancers can have a subtle or indeterminate appearance on mammogram.23 In contrast, FAST MRI was originally designed to preferentially pick up biologically aggressive cancers,11,24 therefore choosing not to recall indeterminate findings from screening with FAST MRI might be less risky because aggressive cancers would be clearly seen as cancer. Whilst choosing to screen with FAST MRI rather than mammograms has the primary objective of reducing under diagnosis, choosing not to recall indeterminate (“3”) FAST MRI findings might have additional benefits of reducing both overdiagnosis and false-positive recall rate.24 Reducing the number of screening assessments by reducing the number of women with normal breasts incorrectly recalled from screening would also reduce cost to the NHS. Examples of FAST MRI images of breasts with cancer from the dataset are shown in Figure 3.
Figure 3. .
Examples of FAST MRI images of breasts with cancer from the dataset. (a) FAST MRI axial MIP image showing a Grade 3 invasive breast cancer (arrow) as an enhancing mass correctly classified as MRI 4 or 5 by 8/8 image readers. The original clinical indication for breast MRI was ‘palpable cancer not visible(occult) on mammogram’. (b) FAST MRI axial MIP image showing high gradeductal carcinoma in situ (DCIS) as asymmetric non-mass enhancement (narrow arrows) correctly classified as MRI 4 or 5 by 7/8 image readers. This image also illustrates that the heart (paired arrows) and blood vessels exhibit enhancement on FAST MRI. (c) FAST MRI axial slice from slice stack showing a lobular carcinoma (single arrow) that showed only on this single slice from the slice stack and not on the MIP. This case was the only cancer from the dataset to be ‘missed’ by 8/8 readers. This figure also demonstrates the appearance of movement artefact on subtracted images like FAST MRI. Movement artefact appears as adjacent black and white lines, together producing a ‘ghost-like’ appearance (paired arrows). Movement artefact is likely to have contributed to the readers’ failure to perceive the lobular cancer.
Limitations of the study include that it uses an enriched dataset and not a real-life data series of screening cases. Therefore, comparison of our results with those of screening studies is spurious. The creators of any enriched dataset determine its degree of difficulty, and therefore comparisons between readers of differently created datasets can be meaningless. As a consequence of the selection of cancers in our dataset being dependent on the indications for breast MRI at our institution, there were a very high proportion of lobular cancers included in our dataset (25/56 = 45%). This bias in our selection of cancer cases and our choice to limit the size of the included cancers increased the difficulty of the test we set our readers. In opposition to this effect, we chose not to include MRIs performed following a known cancer diagnosis that had originally been reported as MRI 1, 2 or 3. This excluded cases that were occult on full protocol MRI, but since only five scans were excluded for this reason (Figure 1), the effect on reader results is not likely to be marked. In the UK, both MRI and mammograms performed for breast screening are double-read, but in this study, the dataset of FAST MRIs was single-read, without access to previous examinations.
Overall, we believe that the enriched dataset of FAST MRI images developed during this study is an effective and challenging test of performance at FAST MRI interpretation, and that the performance of readers when reading this dataset is likely to be an underestimate of their potential performance in screening practice and may also be an underestimate of the difference between groups.
It may be considered as a further limitation that the FAST MRI dataset was acquired on a mixture of 1.5T and 3T scanners, and a strength of this study is that both this mixture of scanners used and the mixture of multiprofessional readers who participated in the study represent the full range of scanners and of mammogram readers within NHSBSP at our centre.
The results of this study suggest that training mammogram readers to interpret FAST MRI may not take long, but allowance for longer interpretation times should be factored into workforce planning whilst they adapt to the new technology. Questions for future research include whether readers new to the technology can, with experience or additional training, speed up to the level achieved by expert MRI readers and how much additional experience or training is required to achieve adequate parity of performance.
Conclusion
Overall, the results of this study, of a small sample of image readers, suggest that brief training of the whole NHSBSP image-reading workforce is likely to be sufficient to enable effective interpretation of FAST MRI in terms of accuracy and speed of interpretation.
Further studies, increasing the number of study participants (readers) and lengthening the training to a whole day of interactive teaching are necessary to decide whether the difference between the groups can be reduced and the overall accuracy and interpretation speed further improved prior to subsequent prospective studies of FAST MRI in a real-world screening setting.
A prospective study in real-life screening practice would be needed to determine whether choosing only to recall suspicious or malignant-appearing cases (MRI4 and MRI5) on FAST MRI could be an effective strategy, rather than the current standard of additionally recalling indeterminate classification breasts (MRI3).18
Footnotes
Acknowledgment: This study was performed on behalf of the FAST MRI Study Group which at the time of this study, in addition to the authors, comprised Christiane Kuhl, Alexandra Valencia, Elisabeth Kutt, Alice Pocklington, Anjum Mahatma, Helen Massey, Gillian Clark, Clare McLachlan, Gemini Beckett, Joanne Robson and Anna Mankelow. The authors wish to thank the Breast Unit Support Trust (BUST) and Independent Cancer Patients’ Voice (ICPV) charities and the National Institute for Health Research (NIHR) Research and Design Service for their invaluable contribution to the study design of this research.
Funding: This research has been carried out with the support of North Bristol NHS Trust Research Capability Funding.
Consent for publication: Consent for publication of the MRI images included in this article has been obtained from the patients imaged.
Availability of data and material: The datasets generated during the current study are not publicly available because the dataset has not yet been converted into an electronic form. The development of the dataset into an electronic form is the subject of a successful grant funding bid to NIHR. Once this has been achieved, and the dataset is in a readily shareable form, it will be available from the corresponding author on reasonable request.
Ethics approval and consent to participate: The study was performed in accordance with the Declaration of Helsinki. Ethics approval was given following review by the South West–Central Bristol Research Ethics Committee, with the reference number REC:17/SW/0142. The documents reviewed and approved by the ethics committee included the participant information sheets and consent forms for the image readers in the study. All eight image readers gave informed consent to their participation in the study.
Author contributions: Guarantor of integrity of the entire study: Lyn I Jones
Study concepts and design: Lyn I Jones, Rebecca Geach, Sam A Harding, Christopher Foy, Victoria Taylor, Andrea Marshall, Sian Taylor-Phillips and Janet A Dunn
Literature research: Lyn I Jones, Rebecca Geach and Sam A Harding
Clinical studies: N/A
Data analysis: Lyn I Jones, Rebecca Geach, Sam A Harding, Christopher Foy, Victoria Taylor, Andrea Marshall and Janet A Dunn
Statistical analysis: Andrea Marshall and Janet A Dunn
Manuscript preparation: Lyn I Jones
Manuscript editing: Lyn I Jones, Rebecca Geach, Sam A Harding, Christopher Foy, Andrea Marshall, Sian Taylor-Phillips and Janet A Dunn
Contributor Information
Lyn I Jones, Email: lyn.jones@nbt.nhs.uk, lyn@coppock.uk.com.
Rebecca Geach, Email: rebecca.geach@nbt.nhs.uk.
Sam A Harding, Email: samantha.harding@nbt.nhs.uk.
Christopher Foy, Email: c.foy@nhs.net.
Victoria Taylor, Email: victoria.taylor4@uhbristol.nhs.net.
Andrea Marshall, Email: andrea.marshall@warwick.ac.uk.
Sian Taylor-Phillips, Email: s.taylor-phillips@warwick.ac.uk.
Janet A Dunn, Email: j.a.dunn@warwick.ac.uk.
REFERENCES
- 1.Marmot M. Review the benefits and harms of breast cancer screening: an independent review independent UK panel on breast cancer Screening*. Lancet 2012; 380: 1778–86. [DOI] [PubMed] [Google Scholar]
- 2.Autier P, Boniol M, Koechlin A, Pizot C, Boniol M. Effectiveness of and overdiagnosis from mammography screening in the Netherlands: population based study. BMJ 2017; 359: j5224. doi: 10.1136/bmj.j5224 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Jacklyn G, Glasziou P, Macaskill P, Barratt A. Meta-Analysis of breast cancer mortality benefit and overdiagnosis adjusted for adherence: improving information on the effects of attending screening mammography. Br J Cancer 2016; 114: 1269–76. doi: 10.1038/bjc.2016.90 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Breast Cancer Care. Breast Cancer (C50) Proportion of Cases Diagnosed at Each Stage, All Ages [Internet]. 2017. p. 2014. Available from: http://www.cancerresearchuk.org/sites/default/files/cstream-node/inc_by_stage_country_breast.pdf.
- 5.NICE Guidelines Familial breast cancer : classification, care and managing breast cancer and related risks in people with a family history of breast cancer. Natl institute Heal Care Excell UK [Internet]. 2017. Available from: https://www.nice.org.uk/guidance/cg164/chapter/Recommendations#surveillance-and-strategies-for-early-detection-of-breast-cancer. [PubMed]
- 6.Evans DG, Astley S, Stavrinos P, Harkness E, Donnelly LS, Dawe S, et al. Improvement in risk prediction, early detection and prevention of breast cancer in the NHS breast screening programme and family history clinics: a dual cohort study. Programme Grants Appl Res 2016; 4: 1–210. doi: 10.3310/pgfar04110 [DOI] [PubMed] [Google Scholar]
- 7.Evans DGR, Donnelly LS, Harkness EF, Astley SM, Stavrinos P, Dawe S, et al. Breast cancer risk feedback to women in the UK NHS breast screening population. Br J Cancer 2016; 114: 1045–52. doi: 10.1038/bjc.2016.56 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Saadatmand S, Bretveld R, Siesling S, Tilanus-Linthorst MMA. Influence of tumour stage at breast cancer detection on survival in modern times: population based study in 173,797 patients. BMJ 2015; 351: h4901. doi: 10.1136/bmj.h4901 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Saadatmand S, Obdeijn I-M, Rutgers EJ, Oosterwijk JC, Tollenaar RA, Woldringh GH, et al. Survival benefit in women with BRCA1 mutation or familial risk in the MRI screening study (MRISC). Int J Cancer 2015; 137: 1729–38. doi: 10.1002/ijc.29534 [DOI] [PubMed] [Google Scholar]
- 10.Jones LI, Dunn JA, Marshall A, Kuhl CK. rapid-responses @ www.bmj.com Mapping the drivers of overdiagnosis to potential solutions: Is the UK ready for an Imaging Biomarker solution to the Breast Screening Debate? [Internet]. BMJ (Online). 2017. Available from: https://www.bmj.com/content/358/bmj.j3879/rr-3.
- 11.Kuhl CK, Schrading S, Strobel K, Schild HH, Hilgers R-D, Bieling HB. Abbreviated breast magnetic resonance imaging (MRI): first postcontrast subtracted images and Maximum-Intensity Projection—A novel approach to breast cancer screening with MRI. JCO 2014; 32: 2304–10. doi: 10.1200/JCO.2013.52.5386 [DOI] [PubMed] [Google Scholar]
- 12.Chhor CM, Mercado CL. Abbreviated MRI protocols: wave of the future for breast cancer screening. AJR Am J Roentgenol 2017; 208: 284–9. doi: 10.2214/AJR.16.17205 [DOI] [PubMed] [Google Scholar]
- 13.Chen S-Q, Huang M, Shen Y-Y, Liu C-L, Xu C-X. Application of abbreviated protocol of magnetic resonance imaging for breast cancer screening in dense breast tissue. Acad Radiol 2017; 24: 316–20. doi: 10.1016/j.acra.2016.10.003 [DOI] [PubMed] [Google Scholar]
- 14.Melnikow J, Fenton JJ, Whitlock EP, Miglioretti DL, Weyrich MS, Thompson JH, et al. Supplemental screening for breast cancer in women with dense breasts: a systematic review for the U.S. preventive services Task force. Ann Intern Med 2016; 164: 268–78. doi: 10.7326/M15-1789 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Pauli R, Hammond S, Cooke J, Ansell J. Radiographers as film readers in screening mammography: an assessment of competence under test and screening conditions. Br J Radiol 1996; 69: 10–14. doi: 10.1259/0007-1285-69-817-10 [DOI] [PubMed] [Google Scholar]
- 16.Bennett RL, Sellars SJ, Blanks RG, Moss SM. An observational study to evaluate the performance of units using two radiographers to read screening mammograms. Clin Radiol 2012; 67: 114–21. doi: 10.1016/j.crad.2011.06.015 [DOI] [PubMed] [Google Scholar]
- 17.Harding S, Geach R, Jones L, .FAST MRI Study Group . The use of 'Think-Out-Loud' methodology in the development of teaching materials for abbreviated breast Magnetic Resonance Imaging scan (FAST MRI) interpretation, and a comparison of the learning experience of two reader cohorts. Eur J Radiol Open 2019; 6(May): 220–4. doi: 10.1016/j.ejro.2019.06.002 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.England PH. Technical guidelines for magnetic resonance imaging (MRI) for the surveillance of women at higher risk of developing breast cancer (NHSBSP Publication No 68) [Internet].Gov.Uk. 2012. Available from: https://www.gov.uk/government/publications/nhs-breast-screening-using-mri-with-higher-risk-women.
- 19.Taylor K, Britton P, O'Keeffe S, Wallis MG. Quantification of the UK 5-point breast imaging classification and mapping to BI-RADS to facilitate comparison with international literature. Br J Radiol 2011; 84: 1005–10. doi: 10.1259/bjr/48490964 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Britton P, Warwick J, Wallis MG, O'Keeffe S, Taylor K, Sinnatamby R, et al. Measuring the accuracy of diagnostic imaging in symptomatic breast patients: team and individual performance. Br J Radiol 2012; 85: 415–22. doi: 10.1259/bjr/32906819 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Kuhl CK, Strobel K, Bieling H, Leutner C, Schild HH, Schrading S. Supplemental breast MR imaging screening of women with average risk of breast cancer. Radiology 2017; 283: 361–70. doi: 10.1148/radiol.2016161444 [DOI] [PubMed] [Google Scholar]
- 22.Sung JS, Stamler S, Brooks J, Kaplan J, Huang T, Dershaw DD, et al. Breast cancers detected at screening MR imaging and mammography in patients at high risk: method of detection reflects tumor histopathologic results. Radiology 2016; 280: 716–22. doi: 10.1148/radiol.2016151419 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Boisserie-Lacroix M, Hurtevent-Labrot G, Ferron S, Lippa N, Bonnefoi H, Mac Grogan G. Correlation between imaging and molecular classification of breast cancers. Diagn Interv Imaging 2013; 94: 1069–80. doi: 10.1016/j.diii.2013.04.010 [DOI] [PubMed] [Google Scholar]
- 24.Heacock L, Lewin AA, Gao Y, Babb JS, Heller SL, Melsaether AN, et al. Feasibility analysis of early temporal kinetics as a surrogate marker for breast tumor type, grade, and aggressiveness. J Magn Reson Imaging 2018; 47: 1692–700. doi: 10.1002/jmri.25897 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
Availability of data and material: The datasets generated during the current study are not publicly available because the dataset has not yet been converted into an electronic form. The development of the dataset into an electronic form is the subject of a successful grant funding bid to NIHR. Once this has been achieved, and the dataset is in a readily shareable form, it will be available from the corresponding author on reasonable request.