Purpose: To study the influence of annual screen-reader volume on radiologists' performance in a mammography screening program using independent double reading with consensus of digital mammograms.
Methods and Materials: We collected retrospective data from 2,373,433 digital mammograms read by 121 radiologists in BreastScreen Norway from 2006-2016. Logistic regressions with robust standard errors were used to explore how sensitivity, false-positive rate (FPR), accuracy (sensitivity/FPR) and screening-cancer detection rate (SDC) were related to annual reading volume.
Results: In the range from 100 to 10,000 annual screen readings, sensitivity increased from 87 to 89%. For higher volumes, there was a decline in sensitivity to 75% at 18,000 annual readings. The FPR declined from 6.7% at 100 annual readings to 4.5% at 4000, to 4.0% at 10 000 and to 3.0% at 18 000 annual readings. Accuracy increased with 50% from 100 to 4000 annual readings, with 15% from 4000 to 10,000 and with 12% from 10,000 to 18,000 annual readings. The SDC was 4.8/1000 at 100 annual readings and 5.1/1000 at 10,000 annual readings. For higher volumes, the SDC rate declined to 3.2 per 1000 at 18,000 annual readings.
Conclusion: Our study indicated that increasing annual reader volume had only a minor effect on sensitivity or SDC rate, except for a decline in performance for readers with extremely high annual volumes. The FPR improved with increasing annual reader volume, most markedly up to an annual volume of 4000 readings.
Purpose: To describe and model the impact of conversion from film screen mammography (FSM) with full-field digital mammography (FFDM) in the English NHS breast cancer screening programme.
Methods and Materials: Annual screening data (KC62 returns) for each of the 80 units over the seven years from 2009/10 to 2015/16 were used to examine the impact of changing from FSM to FFDM. Regression models were used to estimate percentage and absolute change in detection rates.
Results: The recall rate to assessment (approximately 3.9%) was almost unchanged by the introduction of FFDM. After conversion to FFDM the overall cancer detection rate rose from 6.95 to 7.95 per 1000 screens, 14% (11%-17%) p<0.001. High-grade DCIS detection rate rose from 0.71 to 1.00 per 1000, 39% (28%-50%) p<0.001. Invasive grade 1 and 2 detection rate rose from 4.30 to 4.93, per 1000: 14% (10%-18%) p<0.001. Invasive grade 3 detection did not change from 1.20 to 1.22, per 1000: 2% (-5%-9%) p=0.53. The magnitude of the effect for high-grade DCIS and grade 1 and 2 invasive cancers was greater for first (prevalent) screen (age 45-52 years). Grade 3 detection rates were not affected by screen type.
Conclusion: Digital mammography has increased the overall sensitivity of screening by increasing the cancer detection rate at the same recall rate as film screen and, therefore, improved the effectiveness of the screening process. However, there has been no increase in the detection of potentially life-threatening grade 3 cancers.
Purpose: To describe the mammographic features of screen-detected carcinoma in situ and atypical hyperplasia in the United Kingdom (UK) Breast Screening Programme.
Methods and Materials: The Sloane project is a prospective audit of women with screen-detected non-invasive breast cancer and atypia. Screening units in the UK were encouraged to register patients and complete forms documenting imaging features, surgical management, pathology and radiotherapy (if applicable). Data were collected from 1 April 2003 to 24 October 2018.
Results: 14,057 women are registered from 82 of the 90 UK screening units to date. 13,213 have a complete pathology record: 11,335 DCIS and 1,878 ‘atypia’. Overall, micro-calcification was the predominant radiological feature in 11,087 (83.9%) and was present in 11,638 (88.1%). The presence of micro-calcification was directly related to ‘grade’. FEA 80.6%, ADH 72.1%, low-grade DCIS 79.1%, intermediate 84% and high-grade DCIS 94.5% (trend for DCIS grade Chi2 522.9 df=2, p<0.0001). Of those women with calcification, casting/linear calcification increases with ‘grade’: FEA 8.8%, ADH 12.5%, low-grade DCIS 19.4%, intermediate 30.6%, and high-grade DCIS 50.6%. A mass was the predominant feature in 16.8%, 17.6%, 19.3%, 14.4%, and 4.1%, respectively. After calcification, parenchymal distortion was the predominant radiological feature in 14.2% (125 of 878) cases of LISN compared to 3.3% of the other pathologies combined.
Conclusion: As the ‘aggressiveness’ of non-invasive screen detected cancer/atypia increases, the proportion of women presenting with calcification (and linear casting morphology) and the proportion with a mass as the predominant radiological feature decreases.
Purpose: To model the association between cancer detection and recall rates to understand the optimal balance of harm and benefit.
Methods and Materials: Non-linear and linear regression models were used to examine and model the association between recall rate and cancer detection rate using annual screening programme information for the 80 English breast screening units (11.3 million screening tests) supplemented by previously published data from the Dutch screening programme.
Results: Low recall rates are associated with low cancer detection. As recall rates rise, the model indicates that the cancer detection rate for invasive cancers and high-/intermediate-grade DCIS reaches a near-plateau above which almost all recalls are false positive. The cancer grade predicts the point the recall rate reaches plateau. For incident screens the recall rate above which almost no additional cancers are found is 2.5% for both grade 3 and high-grade DCIS. 3.9% for grade 2 and 5.2% for grade 1. However, for low-/intermediate-grade DCIS (LIG) detection rate has no discernible plateau with detection rate increasing linearly at a rate of 0.12 (prevalent) and 0.18 (incident) per 1000 for every 1% increase in recall rate.
Conclusion: Our model predicts that there is an optimum range for recall that maximises detection of life-threatening cancers, whilst minimising harm (in England this is between 4.6% and 7% at prevalent screen and between 2.6% and 4% at incident screens).
Purpose: To determine the number and characteristics of cancers additionally detected through quality assurance sessions between screening radiographers and coordinating screening radiologists.
Methods and Materials: We included a consecutive series of 431666 biennial screening mammograms, obtained at a Dutch breast cancer screening region between January 1, 2009 and July 31, 2016. Each screen was double read by 2 certified screening radiologists and the radiographers were encouraged to classify each obtained mammogram according to BI-RADS. At regular 6 week intervals the group of screening radiographers discussed with a coordinating screening radiologist the mammograms of women that were not recalled by the screening radiologists but considered suspicious by the radiographers. The coordinating radiologist then decided for each case whether secondary recall was indicated. During 2-year follow-up, we obtained data on radiological and pathological outcome of all recalled women.
Results: Altogether, 13175 women were recalled (recall rate: 3.1%), of which 2940 were diagnosed with breast cancer (6.8 cancers detected per 1000 screens). A total of 82 women (0.6% of recalls) experienced a secondary recall after the quality assurance sessions, of which 26 (31.7%) proved malignant. These 26 cancers comprised 8 ductal carcinoma in-situ (low grade: 1; intermediate grade: 7; high grade: 2) and 18 invasive cancers (<10 mm: 1; 11-20 mm: 13; >20 mm: 4. B&R I: 9; B&R II: 7; B&R III: 2).
Conclusion: About 1% of the cancers are detected through the quality assurance sessions. A majority of these cancers is invasive and >10 mm and therefore probably not reflecting overdiagnosis.
Purpose: Multidisciplinary meetings and dedicated radiological modalities (e.g., MRI, tomosynthesis) are increasingly used for the evaluation of recalled women. We determined the trend in the incidence of delayed breast cancer diagnosis after repeated recall for the same abnormality at screening mammography.
Methods and Materials: We included all women aged 50-75 years who underwent screening mammography in a Dutch breast cancer screening region between July 1, 1996 - June 30, 2006 (cohort I) and between July 1, 2006 and June 30, 2016 (cohort II). Data were collected on radiologic procedures and histopathology for all recalled women, with a follow-up period of at least 2 years. For each woman with a repeated recall and histological confirmation of breast cancer at the latest recall, two radiologists independently determined whether the woman had been recalled for this lesion at a previous screen.
Results: A total of 1411 recalled women were diagnosed with breast cancer among the 280184 screens of cohort I, resulting in a cancer detection rate (CDR) of 5.0 per 1000 screens. In 28 of these women (2.0%), a repeated recall for the same mammographic abnormality proved malignant. Among the 507828 screens of cohort II, 3504 recalled women had breast cancer, with a CDR of 6.9. In 71 (2.0%) women, a repeated recall for the same screening abnormality showed malignancy.
Conclusion: No decline in the proportion of women with a >2 year delay in breast cancer diagnosis was observed. The workup of recalled women in the Dutch breast cancer screening programme needs improvement.
Purpose: To determine the frequency of bilateral recall at screening mammography and compare outcome with that of unilateral recall.
Methods and Materials: We included a consecutive series of 197,566 screening mammograms obtained between January 1, 2014 and January 1, 2017. During 2-year follow-up, clinical data were collected of all recalls. Screening outcome parameters were determined for women with unilateral and bilateral recall.
Results: A total of 5,629 women were recalled (recall rate 3.0%), of which 153 (2.6% of recalls) comprised a bilateral recall. Biopsy was more frequently performed in women with bilateral recall compared to unilateral recall (P<0.001). The proportion of DCIS among screen detected index cancers (lesion with highest BI-RADS) was comparable for unilateral and bilateral recall (P=0.3). Invasive index cancers after bilateral recall showed a worse tumour grading than those after unilateral recall (P=0.04). The proportion of lymph node positive invasive cancers was comparable for both groups (P=0.7), as well as hormone receptor characteristics of bilateral breast cancer after unilateral and bilateral recall. There was no difference in the proportion of true positives after unilateral versus bilateral recall (P=0.8). Unilateral recall showed a better PPV for biopsy (P=0.01).
Conclusion: Bilateral recalls comprise a small proportion of all recalls. After bilateral recall, biopsy is more frequently performed compared to unilateral recall, with a better PPV of biopsy after unilateral recall. Tumour characteristics of index cancers and bilateral breast cancers are comparable for unilateral and bilateral recall, except for a worse tumour grading of invasive index cancers after bilateral recall.
Purpose: To determinate the incidence, pathological significance and risk factors associated with the presence of solitary dilated duct visualised at mammography.
Methods and Materials: Prospectively evaluation of consecutive mammography was performed in a breast cancer control center, in accordance with local ethical approval. Patients with solitary dilated duct (SDD) at mammography were referred to additional second-look ultrasonography (US). SDD with intraductal components were submitted to percutaneous biopsy, following anatomopathological correlation. Exclusion criteria considered patients previously submitted to breast surgery.
Results: In the period from March 17, 2016 to March 10, 2017, 9,035 mammographic exams were included, 8,125 (90%) screening and 910 (10%) diagnostic exams. 135 SDD (1.49%) were identified at mammography and 94 (1.04%) second-look US were performed. Of these, 22 revealed intraductal components and had percutaneous biopsy performed. No cancer was found at biopsy results. The most prevalent histological findings were: 8 papillomas and 8 fibrocystic changes. Risk factors for mammographic SDD with statistical significance (p<0.05), using T test were: breast density pattern “A” or “B”, breastfeeding, pregnancy, hormone replacement therapy and papillary discharge. Main mammographic and US findings associated with second-look US papilloma diagnostic with statistical significance (p<0.05), using Chi Quadrado test: suspicious calcification, duct diameter > 3.0mm at US, intraductal mass and hard lesion at elastography.
Conclusion: SDD at mammography benefits from second-look US for analysis of intraductal content’s, which often presents benign findings. When intraductal mass was found, the main result was papilloma. We propose that SDD at mammography should be classified as BI-RADS category 0.
Purpose: To compare the outcome of microcalcifications classified as BI-RADS3, 4a and 4b at mammography in patients with a history of breast cancer to that of women undergoing mammography screening.
Methods and Materials: 176 patients(mean age 61 years) with microcalcifications classified as BI-RADS3, 4a and 4b, without sonographic correlation, whom had undergone vacuum-assisted biopsy (VAB) between Oct 2016-Oct 2017 were retrospectively included. The agreement imaging/pathologic result from VAB was verified and for BI-RADS4b lesions and B2 pathologic result excisional biopsy was performed. For each BI-RADS category, pathologic results from VAB in case of B2 diagnosis and from surgical specimens in all the other cases were compared between patients with a history of breast cancer(group A, n=45) and women undergoing screening(group B, n=131). Positive predictive values (PPVs) were compared using Fisher’s exact test.
Results: A total of 73 lesions were classified as BI-RADS3(41.5%), 57 BI-RADS4a(32.4%) and 46 BI-RADS4b(26.1%). The overall PPV for BI-RADS 3, 4a and 4b were 5.5%(4/73), 10.5%(6/57) and 58.7%(27/46). The PPV for BI-RADS3 lesions in group A and B were 13.0%(3/23)and 2.0%(1/50), respectively(p=0.089). The PPV for BI-RADS4a lesions in group A and B were 15.4%(2/13)and 9.1%(4/44), respectively(p=0.611). The PPV for BI-RADS4b lesions in group A and B were 77.8%(7/9) and 54.0%(20/37), respectively(p=0.270).
Conclusion: The likelihood of malignancy in cases of microcalcifications classified as BI-RADS 3, 4a and 4b at mammography tends to be higher in women with a history of breast cancer than in women undergoing screening, the difference was however not statistically significant.
Purpose: Breast borderline lesions (B3) demonstrates variable (9%-35%) malignancy upgrade rate on surgical excision (SE). Our purpose was to evaluate the existence of a mammographic (Mx) finding able to predict the likelihood of malignancy on SE of Mx only B3 lesions, without US correlate, that underwent tomobiopsy-guided vacuum-assisted biopsy (VAB).
Methods and Materials: Between March 2015-March 2018 we have retrospectively analyzed 88 B3 lesions (10 papillary lesions [PL], 44 radial sclerosing lesions [RSL], 13 atypical ductal hyperplasia [ADH], 21 lobular neoplasia [NL]). Mx findings were divided into 4 categories: microcalcifications (MI), architectural distorsions (DA), masses (MA) alone-associated to other Mx findings, and architectural distorsions with microcalcifications (DA+MI). The relation between Mx findings and malignancy upgrade rate on SE was evaluated with exact Fisher test.
Results: An overall upgrade rate of 16% (14/88) was observed for presence of malignancy at SE (9 intraductal carcinomas G1-G2, 2 intraductal carcinomas G3, 1 invasive ductal-lobular carcinoma G2 and 2 invasive lobular carcinomas G2). Mx findings were MI on 32 cases (36%), DA on 31 cases (35%), MA on 14 cases (16%) and DA+MI on 11 cases (13%). A statistically significative association was found between malignancy upgrade rate and lesions having DA+MI as Mx finding (p=0.01). None significant association was found in the other Mx findings. Positive predict values were 10% for PL (1/10), 14% for RSL (6/44), 31% for ADH (4/13) and 14% for NL (3/21).
Conclusion: A high statistically significant likelihood of upgrade was found for B3 lesions that Mx present as DA+MI.
Purpose: to review the most common feature in false positive recall in order to suggest a strategy for assessment of challenging screening mammogram
Methods and Materials: We reviewed a retrospective cohort of 7230 patients involved in Mammogram Screening program between November 2015 and September 2018. We included the false positive recall which at second level exam have been classified as BIRADS 1 or 2. We evaluated two different groups according to agreement recall (AR) or disagreement recall(DR).We compared the differences among tumoral imaging features between the two groups and considering also the overall findings prevalence in screening recall (SR)
Results: We obtained a final cohort of 1157 suspicious findings whose 416(35,9%) have been in AR group and 741(64,1%) in DR group. In the AR group we found, as single or one of multiple features in suspicious lesion, 287(69%) mass, 26 (6,25%) calcifications, 78(18,75%) architectural distortion and 25(6%) asymmetric density; in the DR group we detected 578(78%) mass, 89(12%) architectural distortion, 43(6%) asymmetric density, 31(4%) calcifications. Differences in tumor imaging features between the two groups were statistically significant (p<0.05)
Conclusion: Our results show how mass is the most common feature in false positive recall with no changes between one or two readers. Calcification, when classified as BIRADS 2, is a rare challenging feature, mostly due to the breast radiologist's expertise. Our study proves how Screening Program is a good tool to avoid false positive recall for architectural distortion and asymmetric density, considering the higher prevalence of these findings in recall group