Full-field digital mammography: a retrograde step for small screen-detected cancers?

Our breast surgery service receives many breast cancer referrals from neighbouring breast screening centres with analogue mammography systems. As hospital protocol we perform repeat full-field digital mammography (FFDM) in these women in an attempt to better stage the primary tumours. The aim of this study is to assess whether FFDM detects more disease than analogue mammography in patients with screen-detected cancer with pathological correlation.

Introduction It is widely accepted that there is an association between mammographic density and breast cancer risk [1]. Various studies have examined relationships between ethnicity and breast density patterns using the Wolfe classifi cation system [2], with a view to investigating potential breast cancer risk. Quantra™ is a volumetric assessment tool, which allows reproducible objective measurement of mammographic breast density, eliminating inter-observer variability. Our study sought to investigate the correlation between ethnicity and breast density using Quantra™ measurements. Methods The Quantra™ value was recorded from the mammograms of symptomatic breast patients at St George's Hospital, London over a 6-month period. We compared three distinct ethnic groups; Black (African and Afro-Caribbean), Asian (Indian subcontinent) and White (Caucasian). Mean Quantra™ values were calculated for each group and the Kruskal-Wallis test was applied. Results A total of 428 patients were included in the study. Mean breast density values for the three ethnic groups were as follows: Black, 24.31%; Asian, 21.94%; White, 24.74%. P = 0.0046 (Kruskal-Wallis). Conclusion There is a statistically signifi cant diff erence between the objectively measured breast densities of these three ethnic groups. This is of relevance to the assessment of breast cancer risk. References Introduction Supplementary views (SV) are routinely used in the diagnostic workup of mammographic lesions. Previous studies have demonstrated that digital breast tomosynthesis (DBT) increases the accuracy of classifi cation of mammographic lesions [1]. In this study the diagnostic accuracy of SV is compared with DBT. Methods The study included cases from screening assessment and symptomatic clinics requiring further mammographic workup. The cases were read by seven specialist breast radiologists. In the fi rst session, readers read bilateral two-view full-fi eld digital mammography (FFDM) and SV of the lesion. In the second session, at least 1 week later, readers read bilateral two-view FFDM plus one-view DBT in the same projection as the SV. Mammographic scores (1 to 5), lesion type and fi nal outcome were recorded. Results A total of 356 cases were assessed using receiver operative characteristic analysis to evaluate the diff erence between the two modes. For FFDM plus SV, the area under the curve (AUC) was 0.8807 (95% CI = 0.84374 to 0.91766) and for FFDM plus DBT the AUC was 0.9186 (95% CI = 0.88821 to 0.94895); diff erence in AUCs was 0.0379 with P value 0.055. Conclusion These results demonstrate that the accuracy of DBT has borderline superiority to SV in the diagnostic workup of mammographic lesions. Reference 1 Introduction In 2009, the South West of Scotland Breast Screening Unit (SWSBSU) had to repeat 236 examinations due to mammographic image unsharpness. Such recalls were undesirable because of radiation safety issues, anxiety caused and cost. The aim of this study was to model predictors of recall that could be used in clinical practise to help reduce the numbers of repeat examinations required due to blurring. Methods This retrospective study compared two sample groups (n = 118), randomly selected from a cohort of women who attended the SWSBSU in 2009 (n = 16,194). Tests of signifi cance were used to compare a range of variables in each group and logistic regression was employed to produce four predictor models for recall. Statistical analysis was carried out by the authors using SPSS version 18. Results A 12 mm increase in compressed breast thickness (CBT), a 50 mAs increase in exposure or a 350 ms increase in exposure time will double the odds of recall due to unsharpness. Also, an increase of 2 kVp, when imaging a breast with a CBT of 60 mm, will reduce the odds of recall by approximately 39%. Conclusion CBT is a major determinant of recall due to unsharpness. Within the context of the reported increase in average CBT [1] and rising national obesity, reconsideration of kVp selection criteria may be necessary to minimise the incidence of recall due to unsharpness.
Introduction Objective critique of mammographic image quality (IQ) is vital to assess effi cacy of services provided by mammography units. The European Guidelines for Screening Mammography require an inadequate image rate of 3%. The subjectivity within current IQ criteria allows symptomatic mammograms with surgically altered breasts to appear to achieve these requirements; however, as evidenced in the analysis of symptomatic breast images within a national optimisation study, current methods of classifi cation for symptomatic mammograms are largely subjective, resulting in poor inter/intra-departmental agreement on IQ.  Introduction Numerous publications suggest the compression force applied in mammography must be reduced to encourage initial/ continued attendance by women. These studies (many with limited patient populations) have not used suffi cient statistical correlation of compression force data to image quality to reinforce these contentions. The growth of digital mammography has enabled a comparison of compression force and image quality in analogue and digital images.
Methods A quantitative and qualitative study of mammography units within Ireland collected comprehensive image quality, compression (force and depth) and radiation dose data from 4,790 patient images. The data were analysed using univariate analysis of variance and SPSS statistics including ANOVA to re-examine the connections between all parameters for optimisation of mammographic imaging. Data presented as mean ± SD.
Breast Cancer Research 2011, Volume 13 Suppl 1 http://breast-cancer-research.com/supplements/13/S1 Results The amount of compression force consistently showed signifi cant eff ects on the image quality; perfect and good images consistently required signifi cantly more compression force than moderate and inadequate images. This was especially apparent in digital images. The mean compression force (Newtons) required to produce a perfect image was: 121.34 N for digital craniocaudal; 134.23 N for digital mediolateral oblique; 112.23 N for analogue craniocaudal; and 129.66 N for analogue mediolateral oblique. Only 2% of patients expressed dissatisfaction with the higher compression forces that were applied. Conclusion Compression forces are too low, aff ecting image quality; greater compression force by 11 to 15 N is needed to achieve a perfect image especially in full-fi eld digital mammography. Greater training of radiographers performing mammography is required to standardise the undertaking of the mammographic projections with regard to achievable compression depth and the application of compression force delivered to the breasts of Irish women attending the symptomatic breast services. Introduction Third-reader screen-reading is undertaken in our unit to reconcile discordant reader results. An audit has been conducted to analyse the nature of cancers diagnosed at assessment for arbitration recalls. The objective was to identify and describe discordant cancer types and appearances.

Methods
The study period covered two screening rounds. Arbitration records were reviewed and mammographic shape, size, position, cancer type and grade, and histological size were recorded. Both fi lm and digital cases were included and compared. Descriptive statistics were produced comparing discordant and concordant cancers. Results A total of 128 arbitration cancers were analysed (3.6% of total cancers, n = 3,516). There were 5,635 total arbitrations of which 27% (n = 1,519) were assessed. A total of 8.4% were cancer. There were a higher number of smaller sized (1 to 15 mm) cancers in the arbitrated group, 61% compared with 48% in the concordant group. There was no diff erence between fi lm and digital cancer size. More cancers appeared as lobular, tubular and DCIS in the arbitration group. There was an equal spread of calcifi cation and mass type mammographic appearances. Conclusion No previous study had specifi cally analysed arbitration (third-reader) cancers. This audit showed that lesion size is smaller, all cancer types are present and both calcifi cations and masses are equally represented. Digital cases did not show any smaller cancers for the discordant group. A future audit will be to look at the arbitration interval cancers; that is, when two readers have not recalled a subsequent false positive case, to assess the features that have been ignored. Methods Information from the last round of the PERFORMS scheme was reanalysed for which the low performance threshold value was known. Data for randomly selected small groups of participants were repeatedly bootstrapped with the aim of artifi cially determining a threshold of low performance and comparing this with the known actual threshold. Using a varying number from four to 50 participants, a sample of 1,000 randomly selected small groups was constructed for each number of participants. After bootstrapping each small group, a distribution of 1,000 thresholds of low performance was constructed and median values and standard errors of this distribution calculated to determine how the number of participants aff ected the estimation accuracy.

P13
Results The standard error of the estimated threshold reduced as group size increased, indicating better estimation accuracy. Using data from as few as 10 people the artifi cial threshold approached the known actual threshold of poor performance. Conclusion Individuals who are performing less than their peers on the scheme can be identifi ed early without all screeners having fi rst taken part. Whilst not an absolute outlier measure, this information can be fed back in a timely manner, so enabling the individual to improve their cancer identifi cation performance. Introduction TARGIT-A is a recent prospective randomised multicentre controlled trial comparing intraoperative radiotherapy (IORT) and external beam whole breast irradiation (EB). The aim of this study was to compare localised and generalised fi ndings at follow-up mammography and the need for interventions such as ultrasound and biopsy between women in the two treatment arms from a single recruiting centre.

Methods
We have compared the 61 women who received IORT alone with the 63 women who received EB alone. All mammograms were reviewed by radiologists blinded to the treatment received. The focal soft tissue appearance was classifi ed. The presence of generalised skin thickening and increased density was recorded. The performance of ultrasound and/or biopsy was recorded.
Results The number of follow-up mammograms and length of followup was similar in both groups (2.46 IORT vs. 2.09 EB and 3.27 years IORT vs. 3.0 years EB). There was no diff erence in the mammographic appearance of the postoperative site between the two groups. However, generalised skin thickening and increase in density were more common in the EB group compared with the IORT group (20 of 63 (37%) vs. 10 of 61 (16%), P = 0.04 and 20 of 63 (37%) vs. 5 of 61 (8%), P = 0.001, respectively). Ultrasound at follow-up was more frequent in the IORT group compared with the EB group (15 of 61 (25%) vs. 7 of 63 (11%), P = 0.049).
Conclusion Generalised reactions on mammography are more common following EB compared IORT. However, follow-up ultrasounds were more frequent in the IORT group. Introduction Vacuum-assisted core biopsy (VACB) is a highly eff ective method of sampling breast tissue. Breast care nurses, at the Queen Elizabeth Breast Unit in Gateshead, raised concerns about pain experienced by patients during this procedure. The aim of this study was to assess how patients perceive VACB.
Methods Prospective data collection for 6 months from November 2010 of consecutive patients undergoing VACB. Data were collected using a questionnaire about pain experienced immediately and 4 weeks post procedure, using the Numerical Rating Scale and Short Form McGill validated pain scores [1]. Results Fifty questionnaires were completed. Immediately post procedure, 88% felt no or mild pain. Of those reporting pain, 100% had a sensory dimension. Four weeks later the memory of the pain experienced during the procedure was worse in 46%. In this group, pain 4 weeks post procedure had an aff ective dimension in 50% of cases. Four weeks post procedure, 91% felt no or mild pain. One hundred per cent would reassure a friend about the procedure. Conclusion Minimal pain was experienced immediately or 4 weeks post procedure. There is a discrepancy between pain experienced during the procedure compared with the memory of it. The eff ect of the biopsy result could be a contributory factor. Further work linking responses to histology may be revealing. Reference In 14 women the NPI group was worse than the PNPI, and in two it was better because the tumours were downgraded at postoperative pathology.
Conclusion Preoperative estimation of the NPI approximates to the defi nitive NPI in the majority of women studied, and could therefore be used to guide systemic treatment decisions preoperatively.
Introduction In 2009 the Royal College of Radiologists Breast Group set out a classifi cation system for breast imaging scoring [1]. A score of 4 'suspicious of malignancy' or 5 'highly suspicious of malignancy' should be used 'in most cases' of proven breast cancer. However, there is no universally accepted target for this in the UK. The American BIRADS system suggests a 98% target [2]; however, their scoring system is slightly diff erent. Methods We examined imaging reports (mammography, ultrasound and MRI) of patients diagnosed with a breast malignancy attending the symptomatic breast clinic in 2009. Our aim was to determine an achievable target for correctly scoring breast imaging. Results A total of 203 patients' imaging was examined. One hundred per cent of reports included a score. Ninety per cent of the cancers were scored 4 or 5. In three underscored cases, radiologist opinion was clouded by the fact the patient had had proceeding FNA. Cases of cancer recurrence also proved diffi cult to correctly score. Conclusion We suggest a target of 90 to 95% of all breast cancers being scored 4 or 5. A 95% target is achievable especially if FNA/biopsy is left until after imaging has been performed.

Introduction
In the NHS Breast Screening Programme (NHSBSP) there is a transition to digital mammography following recommendations made by the Cancer Reform Strategy [1]. A number of US studies have demonstrated that the time taken to interpret digital mammography is longer than that for analogue [2,3]. There are no published data about this from the NHSBSP.
Methods Over a 2-month period, 11 readers were timed in their interpretation of batched analogue or digital mammograms. These were either hung on a multiviewer or preloaded onto Sectra PACS.
Previous images were not digitised. A total of 396 batches were included in the analysis (unpaired t test), 330 digital and 66 analogue.
Results It takes more time to report a digital mammogram compared with analogue (40 ± 1 vs. 35 ± 2 seconds, P <0.05). There is no diff erence in the time taken to report prevalent screens between the analogue and digital groups (34 ± 7 vs. 39 ± 2 seconds). The incident screens were quicker to interpret as analogue.
Conclusion Our data support the hypothesis that digital interpretation is slower than analogue (albeit by 5 seconds) but in the absence of needing to compare with previous images there is no diff erence between the two modalities.

References
Introduction In our Trust, in line with NICE guidance, all breast cancers undergo preoperative axillary ultrasound and, where indicated, needle testing. The current overall sensitivity of this process in our Trust is 57%. This study investigates tumour size, type and location as possible factors for patient triage to repeat preoperative axillary staging, aiming to increase our preoperative axillary staging sensitivity. Methods This prospective study included all patients diagnosed and operated on for unifocal breast cancer in our Trust from September to December 2010. Descriptive statistics, chi-squared and logistic regression were performed upon data collated at MDT meetings. Results Logistic regression of 101 females implies that as tumour size increases by 1 cm, nodal disease risk is 1.75 times larger, with 95% confi dence limits. Comparing tumours <20 mm with those >20 mm, nodal disease risk is 5.818 times larger in the >20 mm group (P <0.0005). No signifi cant diff erence was found in nodal disease risk between the histological tumour types, although numbers of lobular and tubular carcinomas were small (P = 0.633). Data suggested a diff erence (0.27× smaller risk in UIQ versus UOQ) in probability of axillary node metastasis with tumour location with 95% confi dence limits. Conclusion This study demonstrates a clear and statistically signifi cant association between tumour size and nodal disease. The data also suggest a diff erence in probability of nodal disease with diff erent tumour location, although increased patient numbers are needed to confi rm this. A larger trial for stratifying patients for single or double preoperative staging of the axilla is recommended. We used a 14G Achieve needle to make four passes into the node. Results Out of 41 CBs, on histology, eight were negative and 33 were positive. All positives were proven to be true positive at ANC. Seven out of eight negatives were found to be true negative (87.5% true negative) at SNB and ANS. One out of eight negative CBs was found to be false negative (12.5% false negative) requiring axillary clearance. Conclusion We found that our results were comparable with published recent studies from the UK and abroad. Aggressive and more focused sampling could be suggested to further reduce false negatives.

Can radiation dose in mammography be further reduced by
appearance of images from a CR imaging system at the same dose and on both systems at half this dose. Seven experienced observers marked the location of suspicious regions, assigning a fi ve-point score for confi dence that the suspicious region was a cluster. The data were analysed using the area under the alternative free-response receiver operating characteristic (AFROC) and the area under the receiver operating characteristic (ROC) as fi gures of merit. Results There was a signifi cant reduction in detection using CR compared with DR; the AFROC area decreased from 0.83 to 0.63 and the ROC area decreased from 0.91 to 0.79 (P <0.0001). A signifi cant reduction in detection was also evident at half the original dose for both DR and CR. Conclusion The detection of subtle clusters was reduced signifi cantly with CR compared with DR and it is possible that CR will miss cancers manifesting as microcalcifi cation that would be found by DR. Calcifi cation detection was sensitive to the dose used, which should be refl ected in image quality standards to ensure adequate image quality is achieved even at the cost of a higher dose.

Conclusion
The study shows that MRI is superior to DBT in predicting the histological size of lobular carcinoma although there is disparity with both techniques. With the advent of contrast-enhanced 2D mammography, we hope for the future development of contrastenhanced DBT to give us a technique that will be comparable with MRI.

Introduction
The objective was to compare the image quality obtained from a dedicated specimen modality with the image quality from a standard direct digital mammography unit. Methods All wide local excision (WLE), vacuum-assisted 10G core biopsy and 14G core biopsy samples were imaged with a Hologic Dimensions mammography unit and a Bioptics Biovision digital specimen radiography system. WLE specimens were imaged without magnifi cation on both systems. Biopsy specimens were imaged with magnifi cation. Two readers assessed each set of images. The number of microcalcifi cations was recorded and visibility of each lesion was assessed on a four-point scale.
Results The total number of specimens was 97. Of these specimens, 67 contained microcalcifi cation, 23 masses, four distortions and three masses with calcifi cation. In 44/67 specimens, the Biovision system demonstrated >20 microcalcifi cations as opposed to only 24/67 with the Hologic system; this is shown to be signifi cant with P = 0.001. In two of the specimens no calcifi cation was demonstrated on the Hologic system, whereas the Biovision system demonstrated <5 in these cases. A signifi cant diff erence was also shown in the conspicuity of the lesions between the two systems, with the lesions having greater conspicuity on the Biovision system (P = 0.027). Conclusion Signifi cantly more microcalcifi cation is demonstrated by the Biovision system and conspicuity is signifi cantly better. This provides increased confi dence that a representative sample has been obtained at biopsy, and therefore increased diagnostic confi dence.
screening radiologists from both countries examined the same FFDM case set, albeit using diff erent resolution displays. Methods Sixteen experienced American breast screening radiologists interpreted 40 diffi cult FFDM cases containing various mammographic features, excluding small calcifi cations, using dual 202 DICOM calibrated monitors. For comparison purposes, the anonymous data were used of 16 experienced UK breast radiologists who had read the same cases as part of the PERFORMS scheme using clinical mammographic workstations.

Results
The 16 American radiologists were split into two groups of low volume (<5,000 cases p.a.) and high volume (≥5,000 cases p.a.) and performances were compared. There was no signifi cant diff erences (t = 0.23, P >0.05). Consequently their performance data were combined and compared with those of the 16 UK radiologists. There was no signifi cant diff erence between the two groups in correct recall decisions (UK, 97.1%; USA, 92.9%; t = 0.042, P >0.05) although there were signifi cant diff erences in correct return to screening decisions (UK, 88.9%; USA, 80%; t = 0.089, P <0.05) and the number of malignancies detected (UK, 98.7%; USA, 93%; t = 0.049, P <0.05).
Conclusion The use of lower resolution monitors (approximately half that of a mammographic workstation) by the American group was off set by their experience (>15 years) such that even very experienced but low-volume readers performed well. Whilst the UK group overall performed better on these cases, the American group still recalled more, refl ecting their real-life screening criteria. Introduction Digital breast tomosynthesis (DBT) has the potential to improve the accuracy of standard digital mammography (DM) [1]. The TOMMY Trial is a multicentre, multireader, retrospective matched comparison of the diagnostic performance of DBT and DM. Methods Study population Women (47 to 73 years old) recalled for further assessment after routine breast screening and women <50 years with a family history of breast cancer, attending annual mammographic screening. Intervention Women who consent to participate in the trial undergo standard two-view DM and DBT imaging of both breasts. Images are acquired in a single examination under the same degree of breast compression on a commercially available (Hologic) digital mammography system. Outcome measures The primary outcome measure is the relative sensitivity and specifi city of DM and DBT in the detection of early-stage cancers and subtle lesions, particularly in women with dense breasts. This will be evaluated in a retrospective reading study where readers at each centre conduct blinded independent reviews of anonymised DM, or DBT, or DM and DBT images of cases from other centres.

Results
The trial set-up has involved lengthy and complex legal negotiation with collaborating sites, the equipment manufacturer and the grant-awarding body. Designated readers from each centre have completed tomosynthesis training and recruitment has commenced. Conclusion It should be noted that the negotiation of contracts and commercial agreements adds a considerable time burden to the set-up phase of multicentre trials.  Introduction Axillary ultrasound staging with core biopsy (CB) or fi ne needle aspiration (FNA) in primary breast cancer is well established. Negative patients will have a sentinel lymph node biopsy (SLNB). This study compares the initial ultrasound fi nding versus fi nal axillary histology in patients undergoing SLNB. Methods A total of 249 breast carcinoma patients, who underwent SLNB between August 2007 and January 2011, were included. Axillary ultrasound and histology results were reviewed. Ultrasound fi ndings and any subsequent biopsies were recorded in positive and negative SLNB groups. The axillary lymph node biopsy histology slides were reviewed in the false negative axillary ultrasound biopsy/FNA group.

Results
Of 249 patients, 191 (76.7%) were SLNB-negative and 58 (23.3%) were SLNB-positive. Thirty out of 191 (16%) patients without axillary metastases had ultrasound-guided sampling. Two out of 191 did not have ultrasound-guided sampling as the procedure was deemed unsafe. Twelve out of 58 (21%) SLNB-positive patients had abnormal ultrasound appearances; 8/12(67%) had CB and 4/12 (33%) had FNA, not signifi cantly diff erent to the CB/FNA rates in SLNBnegative patients (24/32 (75%) and 6/32 (25%), respectively; P = 0.2). See Figure 1. Review of 12 false negative histology slides revealed 1/12 (8%) had micrometastasis and 11/12 (92%) were benign. Conclusion There are no signifi cant diff erences in the rate of axillary ultrasound abnormality between SLNB-positive and SLNB-negative patients. This suggests that ultrasound nodal abnormalities due to malignancy are probably being diagnosed by ultrasound-guided sampling and do not progress to SLNB. It also supports SLNB for sonographically abnormal lymph nodes as opposed to axillary nodal dissection, as many of these patients will not have metastases, if USS sampling is negative. Our small study has not shown signifi cant benefi t of CB over FNA in axillary staging. We found a low upgrade rate on reviewing original ultrasound-guided histology slides, supporting current pathology techniques. Introduction It is accepted that a proportion of patients treated by breast-conservation surgery for unifocal malignancy will undergo further surgery for residual disease. Preoperative planning for impalpable lesions involves a radiological prediction of the extent of disease. The optimal outcome is disease clearance achieved by one operation only. We have revisited our cases which required a subsequent theatre visit in order to facilitate an understanding of the factors involved in accurate lesion sizing and thereby reduce reexcision rates. Methods All women screened through the Wrexham centre with a positive diagnosis of in situ or invasive disease going on to have operative treatment over a 1-year period have been included. The lesions were double-read by experienced fi lm readers to obtain a maximal dimension on two-view mammography and ultrasound where applicable. Each case was attributed a grading, 1 to 3, for ease of measurement. This was compared with the postoperative histology, reported by a specialist breast histopathologist. The size, type, grade, receptor status and nodal status were recorded for each case. Results A total of 61 cases were selected. The preoperative prediction of size was closest to the mammographic measurement in 74% of cases, and to the ultrasound dimension in 21%. The remainder showed equal measurements on both.
Conclusion The mammogram provides a more accurate prediction of lesion size when compared with ultrasound images. It should be borne in mind that a certain proportion of women will choose mastectomy over conservation regardless of the available option of conservation.
Introduction Symptomatic mammogram machines in trusts that do not run screening services are rarely used by the National Breast Screening Service. This is a potential untapped site for screening. This study surveyed whether women attending current screening sites would go to hospital-based symptomatic units instead. Methods All women who attended NHS breast screening in fi ve diff erent sites in South East England were surveyed over 1 day. One site was at the screening centre. The others were mobile, two being rural and two urban. Home postcodes were used to calculate how far women had travelled and the distance to the nearest symptomatic unit. This was correlated with a questionnaire about mode of transport and whether the women would be ready to attend screening at a symptomatic unit. Results Women at four of the sites said they would just as likely to attend their local hospital for screening. This included one of the rural sites, where the average women would have had to travel no further to her local hospital. The fi fth site was also rural but here most women would have to travel more than 15 miles to the hospital, and most had travelled less than 5 miles to the mobile unit. One woman in three would be less likely to attend screening at the local hospital than this site. Conclusion Women going for screening would be happy to have this in their local hospital symptomatic unit as long as this did not entail extra travel. Breast Cancer Research 2011, Volume 13 Suppl 1 http://breast-cancer-research.com/supplements/13/S1 S11 most appropriate diagnostic pathway, the prognosis and the best form of treatment may be uncertain. To answer these questions, we have examined the outcomes of women presenting in this way over a number of years. Methods Thirteen women were identifi ed prospectively over a 12year period, presenting with suspicious lymphadenopathy but no identifi able breast tumour on initial mammography or ultrasound. Biopsy of the abnormal nodes was consistent with a breast primary in all cases. All women had further imaging with breast MRI (11 cases), breast scintigraphy (one case) and CT scanning of the chest and abdomen. Second-look ultrasound was targeted to suspicious areas identifi ed on second-line imaging. The type of treatment, presence of distant metastases and survival were recorded. Results Further imaging revealed a primary breast lesion in seven cases, six remained truly occult. Follow-up ranged from 3 to 144 months (mean 38 months). Three patients died, one is alive with distant metastases and nine remain disease free. Those with no identifi able primary were treated with chemotherapy usually in combination with radiotherapy.
Conclusion The use of MRI and targeted ultrasound-guided biopsy revealed primary tumours in approximately half our cases presenting with lymphadenopathy and negative conventional imaging.
Chemotherapy with radiotherapy appears to be an eff ective treatment for occult breast cancer. The initial staging tests are crucial and if clear the prognosis appears similar to patients with breast cancer and positive axillary nodes.