Outline Digital Breast Tomosynthesis (DBT) the new standard of - - PDF document

9/25/2017 1

Digital Breast Tomosynthesis:

Update and Pearls for Implementation

Emily F. Conant, M.D.

Professor, Chief of Breast Imaging Department of Radiology Hospital of the University of Pennsylvania Philadelphia, PA

Outline

• Digital Breast Tomosynthesis (DBT) – the new

standard of care

• Breast cancer screening outcomes with DBT:
• Population level
• Patient level

– Outcomes by age – Outcomes by density

• Issues of xray dose – synthetic imaging
• Limitation of DBT

Tomosynthesis Dataset:

2D/3D (Hologic Combo Acquisition)

Arc of motion of x‐ray tube, showing individual exposures

Tomosynthesis Acquisition

Arc of motion of x‐ray tube, showing individual exposures

9/25/2017 2

“3D” Dataset

Reconstructed Tomosynthesis Slices

Projection Images

15 images/150 arc

Projections reconstructed into 1 mm thick slices

Tomosynthesis Dataset:

2D/3D (Hologic Combo Acquisition)

Projection Images 15 images/150 arc Reconstructed “Synthetic 2D” 2D “Dose” Mammogram Reconstructed Tomo Slices

Digital Breast Tomosynthesis

The new, better mammogram:

– Recall reduction (-15-37%) – Increased invasive cancer detection (up to 50%) – No increase in situ detection (? overdiagnosis)

DBT Screening Outcomes

9/25/2017 3

Author (Year) Industry

Volumes DM versus DBT

Recall Rate (%) Cancer Rate (per/1000)

Skaane (2013)

Hologic 12,631 multi reads 6.1 to 5.3% 15% reduction 6.1 to 8.0 27% increase (p<0.001)

Ciatto (2013)

Hologic 7,292 DM then DBT 4.5 to 3.5% 17.2% “conditional” reduction 5.3 to 8.1/1000 (cancers overall not by pt) 52.8% increase

Lang (2015)

Siemens – single view only 7,500 2.6 to 3.8% 46% increase (to 3.8%) 6.3 to 8.9% 43% increase

Friedewald (2014)

Hologic 454,850 (DM: 281,187 DBT: 173,663) 10.7 to 9.1 15% reduction (p<0.001) 4.2 to 5.4 24% increase (p<0.001)

Greenberg (2014)

Hologic 59,617 (DM: 38,674 DBT: 20,943) 16.2 to 13.6% 16% reduction (p<0.0001) 4.9 to 6.6 35% increase (p=0.035)

McCarthy (2014)

Hologic 26,299 (DM: 10,728 DBT: 15,571) 10.4 to 8.8% 15% reduction (p<0.001) 4.6 to 5.5 19.6% increase (p=0.32)

Lourenco (2015)

Hologic 25,299 (DM: 12,577 DBT: 12,921) 9.3 to 6.4% 31% reduction (p<0.00001) 5.4 to 4.6 15% decrease (P =0.44)

Conant (2016)

Hologic 198,881 (DM: 142,883 DBT:55,998) 10.4 to 8.7 % 16.3% reduction (p <.0001) 4.4 to 5.9 34% increase (p=0.0026)

Prospective Retrospective

Summary of DBT Screening Studies

Decrease in Recall of up to 31% Increase in Cancer Detection up to 53%

Sub‐populations reported in Friedewald, et al.

Author (Year) Industry

Volumes DM versus DBT

Recall Rate (%) Cancer Rate (per/1000)

Skaane (2013)

Hologic 12,631 multi reads 6.1 to 5.3% 15% reduction 6.1 to 8.0 27% increase (p<0.001)

Ciatto (2013)

Hologic 7,292 DM then DBT 4.5 to 3.5% 17.2% “conditional” reduction 5.3 to 8.1/1000 (cancers overall not by pt) 52.8% increase

Lang (2015)

Siemens – single view only 7,500 2.6 to 3.8% 46% increase (to 3.8%) 6.3 to 8.9% 43% increase

Friedewald (2014)

Hologic 454,850 (DM: 281,187 DBT: 173,663) 10.7 to 9.1 15% reduction (p<0.001) 4.2 to 5.4 24% increase (p<0.001)

Greenberg (2014)

Hologic 59,617 (DM: 38,674 DBT: 20,943) 16.2 to 13.6% 16% reduction (p<0.0001) 4.9 to 6.6 35% increase (p=0.035)

McCarthy (2014)

Hologic 26,299 (DM: 10,728 DBT: 15,571) 10.4 to 8.8% 15% reduction (p<0.001) 4.6 to 5.5 19.6% increase (p=0.32)

Lourenco (2015)

Hologic 25,299 (DM: 12,577 DBT: 12,921) 9.3 to 6.4% 31% reduction (p<0.00001) 5.4 to 4.6 15% decrease (P =0.44)

Conant (2016)

Hologic 198,881 (DM: 142,883 DBT:55,998) 10.4 to 8.7 % 16.3% reduction (p <.0001) 4.4 to 5.9 34% increase (p=0.0026)

Summary of DBT Screening Studies

Sub‐populations reported in Friedewald, et al.

Prospective Retrospective The improvement in outcomes achieved with DBT directly address the major concerns regarding screening for breast cancer with mammography:

• Too many false positives (low specificity)
• Too few cancers detected (low sensitivity)
• Over‐diagnosis (esp. DCIS)

Recall Reduction

DBT reduces false positive call‐backs:

47yr‐old presents for screening, focal asymmetry, left lateral on CC

Tomosynthesis imaging shows no abnormality. (Tissue superimposition present on 2D)

9/25/2017 4 DBT reduces false positive call‐backs:

43yr‐old presents for baseline screening. Architectural distortion?

On sequential DBT slices, each component of “lesion” is a separate structure (note localizer positions).

No recall needed!

Improved Cancer Detection

A case

46 year old woman presents for screening…

9/25/2017 5

6 cm Invasive Lobular Carcinoma

Teaching point: it’s not all about density…

2D DBT US MRI

Cancer detected on DBT alone:

68‐year‐old woman presenting for screening

The 2D mammogram was interpreted as negative

Findings: Architectural distortion, irregular mass in UOQ,

best on DBT. Work up was directly to US.

Pathology: Invasive lobular carcinoma

9/25/2017 6

Limitations of early trials

• Most studies “first, prevalent” round screening
• Only 3 studies were prospective (Oslo, STORM,

Malmo)

• Majority of retrospective studies had concurrent

DM screening

– potential for bias in screened cohorts

• There has been little data on false negatives….

McDonald EM et al. JAMA Oncol. 2016;2(6): 1‐7

University of Pennsylvania Data

Method:

• Three consecutive years DBT screening

– Population level analysis (each year of screening) – Patient level analysis (each round of screening) – Comparison with cancer registry data for false negatives

Results from Penn consecutive years

• f DBT screening

Metric Year 0 DM Year 1 DBT Year 2 DBT Year 3 DBT

Recall rate (%) 10.4 8.8 9.0 9.2 Cancer rate/1000 4.6 5.5 5.8 6.1 PPV1 4.4 6.2 6.5 6.7 Interval CA/1000 0.7 0.5 ‐

McDonald EM et al. JAMA Oncol. 2016;2(6): 1‐7

4.6 5.5 5.8 6.1 3.2 3.8 4.1 4.1 0.7 0.5 1.8 2 1.9 1.9 4.4 6.2 6.5 6.7

1 2 3 4 5 6 7 8 yr 0 (DM) DBT yr 1 DBT yr 2 DBT yr 3

Population‐level Cancer and Biopsy rates, PPV1 by year

Cancer/1000 Invasive cancer/1000 Interval cancer/1000 Biopsy % PPV1 McDonald EM et al. JAMA Oncol. 2016;2(6): 1‐7

9/25/2017 7 What about first round, “Prevalence Effect”?

McDonald EM et al. JAMA Oncol. 2016;2(6): 1‐7

9.1 8.6 6.9 5.9 4.6 6.3 4.2 7.3 5.1 7.4 6.1 12.4 2 4 6 8 10 12 14 DM with priors One DBT Screen with priors (n= 21395) Two DBT Screens (n= 9316) Three DBT Screens (n=3023)

Patient‐level Outcomes for Consecutive Years of Screening

Recall rate % Cancer/1000 PPV1 %

Conant EF et al. BCR&T 2016; on‐line 3/1/16

PROSPR consortium (BWH-D, UVt, UPenn)

• DM and DBT cases 2011-15 (142,883 DM and 55,998 DBT studies)

– Patient level data

• Reduction in recall (8.7% vs 10.4% p<0.0001)
• Increase in cancer detection (5.9 vs 4.4/1000, p= 0.0026),

–  34% increase in cancers overall –  27% invasive cancers

• Trend in decrease in false negatives (0.46 vs 0.6/1000)

Breast Cancer Research and Treatment – on‐line 3/1/16

2.53 3.77 5.02 4.54 2.9 3.6 5.7 7 3.8 5 7.4 8.2

1 2 3 4 5 6 7 8 9 40‐49 50‐59 60‐69 >70 Cancer Detection per 1000

BCSC DM DBT Derived from unpublished data: Friedewald, et al. JAMA 2014;311(24):2499‐2507

Cancer Detection (per 1000) by Age

Tomosynthesis Outcomes by Density Category

Author Year Mean Age Number of Dense Patients Change in Cancer Detection Rate

(per 1000)

Change in Recall Rate

(per 1000)

Prospective Trials Ciatto 2013 58 1,215 +2.5 ‐26 Lang 2016 56 3,150 +3.8 ‐ Bernardi 2016 58 2,592 +5.4 +10.5 Tagliafico 2016 51 3,231 +4 ‐ Retrospective Trials Rose 2013 54 4,006 +1.4 ‐36.8 McCarthy 2014 55 5,056 +1.8 ‐19.4 Conant 2016 57 9,265 (21,133) +2.1 ‐22.1 Rafferty 2016 ‐ 84,243 +1.4 ‐18.4

Adapted from Houssami N, Turner RM. Breast 2016: 141‐145.

9/25/2017 8

2.52 3.88 4.27 3.37 3.2 4.4 4.5 3.8 4.2 5.3 6.1 3.9

1 2 3 4 5 6 7 Fat Scattered Hetero Extreme Cancer Detection per 1000

BCSC DM DBT Derived from unpublished data: Friedewald, et al. JAMA 2014;311(24):2499‐2507

Cancer Detection (per 1000) by Density Category

What about False Negatives?

False negative DBT: Not all cancers visible with DBT.

42 yo with lump in left breast.

DM DBT Invasive Ductal CA (TNeg)

54 yo with pain, thickening of the right breast

Diagnostic mammogram (2013), ultrasound for rt pain, thickening “negative”. MRI extensive asymmetric NMLE enhancement rt breast. Invasive ductal carcinoma.

MIP Right DBT MLO Views MIP Left (Normal) 2012 2013

9/25/2017 9

What about Dose?? Comparison of DBT Dose

• Hologic DBT has dose boost at > 50mm
• Siemens DBT dose higher than DM for all thicknesses
• GE DBT dose same as DM (uses grid for both modes)

GE Siemens Hologic

Courtesy of Andrew Maidment, PhD

Synthetic 2D Imaging Overview of s2D Reconstruction

Multiple, low dose images

• btained and then

reconstruction

DBT image data used to create both 1mm slices for tomo “stack” and s2D images

Tomo Stack s2D Mammo

9/25/2017 10

Synthetic 2D is about more than just dose!

Need 2D images for interpretative tasks essential to improved performance of DM/DBT screening:

• Global assessment of breasts

– BI‐RADS density assessment – Comparison with prior images – Assessment of lesions – esp. calcs, asymmetries, etc.

• Efficient, high volume screening

Synthetic 2D DM 2D p value Non‐calc lesions* (n=73)

Observer 1 74.0% (54/73) 72.6% (53/73) 0.996 Observer 2 83.6% (61/73) 83.6% (61/73) ‐ Observer 3 84.9% (62/73) 83.6% (61/73) 0.991

Calcified lesions* (n=14)

Observer 1 92.9% (13/14) 85.7% (12/14) 0.996 Observer 2 92.9% (13/14) 100% (14/14) 0.995 Observer 3 92.9% (13/14) 92.9% (13/14) ‐ *Analysis performed on 87 cancers seen on either synthetic 2D or digital (DM) 2D

Skaane (RSNA 2014): Detection Sensitivities Synthetic 2D versus DM Gilbert et al (online Radiol 7/17/15): Accuracy of Digital

Breast Tomosynthesis for Depicting Breast Cancer Subgroups in a UK Retrospective Reading Study (TOMMY Trial)

• Multicenter/reader, retrospective study
• Dataset of 7060 cases

– 2D mammo – 2D plus DBT (DBT) – s2D plus DBT (sDBT)

• Sensitivity:

– 87% 2D – 89% 2D + DBT – 88% sDBT

• Specificity

– 57% 2D – 70% 2D + DBT – 72% sDBT “s2D + DBT and 2D + DBT had better sensitivity for depicting 11‐20 mm inv CA than 2D alone. “ “However s2D + DBT was inferior to both 2D alone and 2D + DBT in depicting microcalcs and 11‐20mm DCIS.”

DM

• Arch. Distortion only on s2D and DBT

51 yo architectural distortion seen only on s2D/DBT Pathology: invasive ductal carcinoma.

DBT s2D

9/25/2017 11

2D C‐View 3D Slice

Synthetic 2D with Tomosynthesis

Pathology: Ductal Carcinoma in situ (DCIS)

Synthetic 2D with Tomosynthesis

Pathology: Ductal Carcinoma in situ (DCIS)

s2D DM

s2D RCC

Synthetic Imaging: False Positive Calcifications

56 yo at screening. Possible calcifications? DM Magnification: No definite calcifications Magnification at recall demonstrates no calcifications. s2D reconstruction algorithm may make normal ligaments appear like calcifications.

Synthetic 2D (s2D) use at Penn

Penn: Integrated s2D over 4 mos. (9/14-12/14)

• Screening:

– Jan 7th , 2015 began screening all pts with s2D-DBT

Performance outcomes measured

– Recall rate (by lesion type, cancer detecton, PPV1) – Cancer detection – PPV1, 3

Zuckerman S, et al. Radiology 2016

9/25/2017 12

Performance metrics with s2D?

DBT screening dose reduced by 39%

• Maintained low recall rate (actually continues to decrease)
• Significant changes in calcs, asymmetries and technical recalls

Modality

Overall Recall Calcs Masses Asym. Arch dist Technical DM 10.4% 1.8% 2.4% 6.1% 0.7% 0.44% DM+DBT 8.8% 1.6% 2.7% 4.5% 1.0% 0.2% s2D+DBT 7.1% 1.1% 2.4% 3.2% 1.1% 0.1% p value

(DM/DBT vs s2D/DBT)

<0.001 0.02 0.31 <0.001 0.70 0.03

Zuckerman S, et al. Radiology, 2016

s2D Performance metrics continued…

See RSNA 2015 scientific exhibit RC 315‐09

s2D maintains benefits of DBT:

• Increased number of cancers detected per recall (PPV1)
• Sensitivity and specificity similar, thus far…
• Slight decrease in detection of in situ cancers to be monitored

Metric DM/DBT s2D/DBT p value

Biopsy rate (%) 2.0 1.3 0.001 Cancers/1000 5.45 5.03 0.732 in situ 1.48 0.9 0.301 invasive 3.85 4.10 0.840 PPV1 (%) 6.2 7.1 0.548 PPV3 (%) 27.0 38.6 0.026

Zuckerman S, et al. Radiology, 2016

4.6 5.5 5.8 6.1 5.03 3.2 3.8 4.1 4.1 4.1 0.7 0.5 1.8 2 1.9 1.9 1.3 4.4 6.2 6.5 6.7 7.1 1 2 3 4 5 6 7 8 yr 0 (DM) DBT yr 1 DBT yr 2 DBT yr 3 s2d yr 4

Population‐level Cancer and Biopsy rates, PPV1 by year

Cancer/1000 Invasive cancer/1000 Interval cancer/1000 Biopsy % PPV1

UPenn Data – with s2D as 4th consecutive year…

*s2D data is based on initial 6 months of data – not powered for cancer detection…

*Screening with s2D/DBT

Aujero et al (2017): More on Synthetic 2D/DBT

DM

n=32,076

DM/DBT

n=30,561

s2D/DBT

n=16,173

Recall rate (%) 8.7 5.8 4.3 Cancer detection/1000 5.3 6.4 6.1 Invasive In situ 3.2 1.6 3.89 2.1 4.64 1.4 Biopsy rate (%) 2.4 2.2 1.5 PPV1 (%) 6.0 10.9 14.3 PPV3 22.2 28.5 40.8 Zuckerman S, et al. Radiol 2016 Aujero MP, et al. Radiology on line Feb 23

Are we ”missing” in situ cancers? Is this ok (decreasing overdx)

• r bad (increasing interval cancers?)

9/25/2017 13

DBT is the “better mammogram”…

Additional outcome data is needed

– Modality based:

• DBT versus Ultrasound, “abbreviated MR”, Molecular Breast

Imaging, Contrast mammography (2D or DBT)

– Larger populations for subgroup analyses:

• Age, density, screening intervals, etc.

– Biology, biology, biology…

• Are additional DBT cancers biologically more aggressive?
• Surrogate for mortality reduction?

Perelman Center for Advanced Medicine University of Pennsylvania Medical Center

Thank you!