JAMA Network Open NOVEMBER 30, 2018

Evaluation of Digital Breast Tomosynthesis as Replacement of Full-Field Digital Mammography Using an In Silico Imaging Trial

Aldo Badano, Christian G. Graff, Andreu Badal, Diksha Sharma, Rongping Zeng, Frank W. Samuelson, Stephen Glick, Kyle J. Myers

Bottom Line

The VICTRE (Virtual Imaging Clinical Trial for Regulatory Evaluation) project successfully utilized an in silico, computer-simulated trial to demonstrate that digital breast tomosynthesis (DBT) provides superior lesion detection compared to traditional full-field digital mammography (DM), aligning with clinical findings.

Key Findings

1. The study demonstrated a statistically significant improvement in lesion detection performance with DBT compared to DM, with a mean (SE) AUC of 0.9596 (0.0035) for DBT versus 0.9005 (0.0058) for DM (P < .001).
2. The differential performance favoring DBT was consistent across all breast sizes and lesion types, including masses and calcifications.
3. The improvement in AUC was more pronounced for masses (mean change 0.0903 [0.008]) than for calcifications (mean change 0.0268 [0.004]).
4. These in silico results effectively replicated the performance gains observed in traditional human clinical trials, validating the utility of computational modeling for regulatory assessment of imaging devices.

Study Design

Design
In Silico Clinical Trial
N/A
Sample
2,986
Patients
Duration
N/A
Median
Setting
Computational simulation
Population 2,986 synthetic image-based virtual patients with varying breast sizes (compressed thickness 3.5-6 cm) and Breast Imaging Reporting and Data System (BI-RADS) densities representing a screening population.
Intervention Digital breast tomosynthesis (DBT) imaging simulation using fast Monte Carlo x-ray transport.
Comparator Full-field digital mammography (DM) imaging simulation using fast Monte Carlo x-ray transport.
Outcome Difference in the area under the receiver-operating-characteristic curve (AUC) between DBT and DM for lesion detection.

Study Limitations

As an in silico (computer-simulated) study, the findings are dependent on the accuracy and representativeness of the simulated breast phantoms and imaging physics models.
The study used a computational reader to detect lesions rather than human radiologists, which may not perfectly replicate clinical real-world interpretation scenarios.
While validated against clinical trial outcomes, the results require ongoing assessment of generalizability to broader, diverse patient populations beyond the simulated cohort.

Clinical Significance

This study validates a new regulatory paradigm demonstrating that computer-simulated trials can serve as reliable, cost-effective, and efficient sources of evidence for evaluating the clinical performance of breast imaging technologies like DBT, potentially accelerating patient access to innovative diagnostic tools.

Historical Context

Traditional clinical trials for medical imaging are often expensive, time-consuming, and logistically burdensome. The VICTRE project was a landmark initiative by the FDA to establish and validate the role of high-fidelity computational modeling—in silico trials—as a complementary or alternative approach to provide scientific evidence for the regulatory evaluation and approval of new imaging systems.

Guided Discussion

High-yield insights from every perspective

Med Student
Medical Student

How does the fundamental technical difference between digital mammography (DM) and digital breast tomosynthesis (DBT) explain why DBT is more effective at detecting lesions in dense breast tissue?

Key Response

Digital mammography provides a 2D projection where overlapping glandular tissue can mask a tumor (summation artifact). DBT acquires multiple low-dose images at different angles, allowing for 3D reconstruction into thin slices. This reduces tissue overlap, making it easier to visualize the borders of masses and architectural distortions that would otherwise be obscured.

Resident
Resident

Based on the findings of the VICTRE trial and clinical correlates, how does the adoption of DBT impact the screening program's recall rate and positive predictive value (PPV)?

Key Response

DBT typically results in a lower recall rate (fewer BI-RADS 0 assessments) because the 3D slices allow radiologists to dismiss 'pseudo-lesions' caused by overlapping normal tissue. Simultaneously, it increases the cancer detection rate (CDR) and PPV because true lesions are characterized more clearly, reducing unnecessary workups while catching more invasive cancers.

Fellow
Fellow

The VICTRE study utilized 'in silico' modeling to compare DBT and DM. In the context of breast imaging subspecialty practice, how do these simulation results compare to real-world performance regarding the detection of microcalcifications versus spiculated masses?

Key Response

Clinical trials (like the Malmö Breast Tomosynthesis Screening Trial) and the VICTRE simulation both show that DBT's greatest strength is detecting architectural distortions and spiculated masses by resolving tissue overlap. While DBT is non-inferior for microcalcifications, some radiologists still prefer high-resolution 2D magnification views for fine morphology, though newer DBT algorithms are narrowing this gap.

Attending
Attending

If in silico trials like VICTRE can accurately replicate the outcomes of high-cost human clinical trials, what are the broader implications for the regulatory approval of AI-driven image reconstruction algorithms in breast imaging?

Key Response

VICTRE demonstrates that simulated trials can serve as a 'regulatory science' tool. This could allow for faster FDA clearance of iterative software updates or AI algorithms by testing them on standardized virtual populations, bypassing the years of recruitment and radiation exposure required for traditional human trials while maintaining rigorous safety and efficacy standards.

Scholarly Review

Critical appraisal through the lens of expert reviewers and guideline development

PhD
PhD

A key component of the VICTRE trial was the use of a 'model observer' for lesion detection. What are the statistical risks of using a mathematical observer versus a human observer in an in silico trial, and how can researchers ensure the model's generalizability to clinical practice?

Key Response

Mathematical observers (like the Channelized Hotelling Observer) may optimize for signal-to-noise ratios that do not mirror human visual perception or cognitive biases. Validation requires 'reader studies' where human radiologist performance is compared against the model observer on the same simulated dataset to ensure the model reflects human-like diagnostic thresholds.

Journal Editor
Journal Editor

As a reviewer, what critical validity threats should be addressed when a study relies entirely on simulated anthropomorphic phantoms and computer-generated lesions rather than actual patient data?

Key Response

A tough reviewer would flag the 'ground truth' problem: the simulation might assume a perfect model of a tumor that lacks the biological heterogeneity of real cancer. Furthermore, the absence of movement artifacts, skin folds, and clinical history in the simulation may overestimate the diagnostic performance of the technology compared to its use in a messy, real-world clinical environment.

Guideline Committee
Guideline Committee

The USPSTF currently notes that while DBT improves cancer detection, there is 'insufficient evidence' regarding its impact on long-term mortality. Does the VICTRE trial provide the type of evidence needed to move DBT from a 'Grade B' to a 'Grade A' recommendation for screening?

Key Response

No. While the VICTRE trial is landmark for demonstrating diagnostic superiority (detection), guideline committees like the USPSTF or ACR require longitudinal data showing that this increased detection leads to a reduction in breast cancer mortality or a decrease in interval cancers. In silico trials measure performance at a single point in time, not long-term clinical outcomes.

Clinical Landscape

Noteworthy Related Trials

2013

STORM Trial

n = 7,292 · JAMA

Tested

Digital breast tomosynthesis (DBT) plus digital mammography

Population

Women undergoing breast cancer screening

Comparator

Digital mammography alone

Endpoint

Cancer detection rate

Key result: The addition of DBT to digital mammography significantly increased the cancer detection rate compared to digital mammography alone.
2016

Malmö Breast Tomosynthesis Screening Trial

n = 14,800 · Lancet Oncol

Tested

Digital breast tomosynthesis (DBT)

Population

Women aged 40-74 years

Comparator

Full-field digital mammography (FFDM)

Endpoint

Cancer detection rate

Key result: DBT increased the cancer detection rate by 34% compared to FFDM alone.
2019

OSLO Trial

n = 105,745 · JAMA Oncol

Tested

Digital breast tomosynthesis (DBT)

Population

Women aged 50-69 years

Comparator

Full-field digital mammography (FFDM)

Endpoint

Detection rate of breast cancer

Key result: DBT was associated with a higher detection rate of invasive breast cancer compared to FFDM.

Tailored to your role

Want this tailored to you?

Add your specialty or training stage to get role-specific takeaways and more questions.

Personalize this analysis