Prepare the case
Route the patient, indication, prior imaging and available metadata into a controlled review environment so the clinician begins with a structured case context.
Workflow
A measurable operating model for multimodal breast imaging
BreastScreening-AI is designed to help providers move from isolated image review toward a coordinated workflow across mammography, ultrasound and MRI. The business goal is to release specialist capacity, reduce avoidable variation and make AI adoption measurable without transferring clinical authority to software.
Operating pathway
From case intake to accountable decision
This pathway describes the product workflow being designed and evaluated. It is a commercial workflow model, not a statement that every component is deployed or authorized in routine care.
Review MG
Mammography anchors screening and first-line detection. AI support is positioned around reading efficiency, suspicious-region attention and consistency checks.
Add US when indicated
Ultrasound supports targeted characterization, dense-breast assessment, palpable findings and biopsy guidance, creating a richer picture than mammography alone.
Use MRI selectively
MRI adds high-sensitivity assessment for selected high-risk screening, disease extent, treatment planning and problem-solving when other modalities are insufficient.
Reconcile AI and clinician judgement
The clinician compares the initial assessment, AI outputs and multimodal evidence, then accepts, rejects or disregards assistance before final interpretation.
Document and measure
The final decision remains clinician-led, while time, decision changes, disagreement, trust and usability can be measured for local value assessment.
Why multimodality
Radiology needs MG, US and MRI because each answers a different business and clinical question
No single modality covers population screening, lesion characterization, dense-tissue assessment, disease extent and treatment planning equally well. A scalable breast-imaging platform must therefore connect the pathway while respecting indication-specific use.
Mammography: scalable screening foundation
Mammography is the population-screening anchor and can reveal early findings such as calcifications. Its scale makes small reading-time gains commercially meaningful across high-volume programs.
Workflow value: standardize first-line review, prioritize attention and measure reader-AI disagreement.
Ultrasound: targeted characterization layer
Ultrasound helps distinguish solid, cystic and mixed findings, supports assessment of palpable abnormalities and can guide biopsy. It is especially useful as a complementary modality rather than a replacement for mammography.
Workflow value: clarify next steps, reduce fragmented handoffs and connect diagnostic workup to the initial review.
MRI: high-sensitivity problem solving
MRI adds detailed contrast-enhanced information for selected high-risk screening, disease extent, treatment planning and unresolved diagnostic questions.
Workflow value: coordinate advanced imaging decisions, support complex-case review and make escalation criteria more traceable.
Combined value: fewer isolated decisions
A multimodal workflow lets providers evaluate the full imaging journey rather than separate tools for separate exams. This supports procurement, governance and performance monitoring across the breast-care pathway.
Business value: one adoption program, one measurement model and a clearer route to budget-impact analysis.
Clinical context: WHO breast cancer fact sheet, mammography, breast ultrasound and breast MRI. Modality choice remains patient-, indication- and protocol-specific.
Workflow value
Six operational levers for providers and imaging networks
The workflow is designed to make clinical AI commercially useful by connecting performance to staffing, throughput, governance and quality measurement.
Capacity release
Published experimental task-time reduction creates a basis for local capacity modeling and staffing scenarios.
Quality support
Lower false-positive and false-negative proportions in controlled research point to quality levers that need prospective validation before routine claims.
Pathway coordination
MG, US and MRI can be managed as connected workflow decisions instead of isolated imaging events.
Team adoption
Personalized interaction evidence supports a rollout strategy for mixed-experience teams, not only expert users.
Measurable procurement
Time, disagreement, decision change, trust and usability can be captured as buyer-relevant outcomes.
Controlled governance
Human oversight, traceability and deployment gates help buyers evaluate AI without weakening clinical accountability.
Modeled workflow impact
How measured workflow gains can scale operationally
The following scenarios apply the published 69-second mean task-time difference to annual review volumes. They show potential capacity value, not guaranteed savings, ROI, reimbursement value or clinical performance in routine care.
Per 100,000 annual reviews
A 69-second reduction per review equals about 1.2 full-time-equivalent years of potential specialist capacity, assuming 1,600 productive hours per FTE-year.
Per 1 million annual reviews
At screening-program scale, the same effect equals approximately 12 FTE-years of gross released capacity before implementation costs.
Illustrative value per 1 million reviews
Using an assumed fully loaded specialist-hour value of €75–€150, one million reviews implies this gross capacity value.
At 10 million annual reviews
A large network or national-scale scenario equals approximately 120 FTE-years of potential capacity.
Illustrative value at 10 million reviews
This applies the same hourly assumption to 10 million reviews. It is capacity value, not cash released to the budget.
Global diagnoses in 2022
WHO estimated 2.3 million women diagnosed with breast cancer and 670,000 deaths globally in 2022, reinforcing the need for scalable early-detection workflows.
Review WHO contextTransparent model
Capacity hours = annual reviews × 69 seconds ÷ 3,600. FTE-years = capacity hours ÷ 1,600. Gross capacity value = capacity hours × assumed €75–€150 specialist hour.
False-positive and false-negative improvements are not converted into avoided biopsies, recalls, treatment costs or mortality gains here because the controlled research figures require prospective validation and local pathway data before downstream economic claims are responsible.
Integration context
Designed to meet the workflow where it operates
Clinical AI depends on more than a model. Useful evaluation requires imaging access, identity and permission controls, interface fit, secure infrastructure, traceable outputs and institution-specific governance.
Clinical systems
Research integrations may involve PACS-adjacent environments or approved application platforms, depending on the site.
Case preparation
Study datasets and identifiers require protocol-led selection, de-identification and controlled handling.
Access controls
Permissions, network configuration and user access must be agreed with the participating institution.
Interface validation
Display, navigation and interaction changes are tested because they can affect study validity and clinical use.
Traceability
Research activities should distinguish the clinician's input, AI output, final decision and relevant context.
Local governance
Ethics, privacy, cybersecurity and clinical responsibilities remain specific to each deployment or study.
Evidence context
Workflow claims grounded in current quantitative evidence
The current evidence supports evaluation of capacity, quality support and adoption. It does not yet prove real-world clinical efficacy, customer ROI or health-system savings.
Clinicians across nine institutions
The principal clinician-AI study compared clinician-only and clinician-AI scenarios across a multi-institution reader sample.
Read the studyLower mean task time
Mean task time decreased from 377 to 308 seconds, a 69-second difference that underpins the workflow capacity scenarios above.
Inspect the publicationFalse-positive proportion
False-positive classifications decreased in the controlled study, supporting a quality hypothesis for future prospective validation.
Review evidence contextFalse-negative proportion
False-negative classifications decreased under the experimental clinician-AI condition, but this should not be generalized as routine-care detection benefit.
Review evidence contextPositive clinician response
Positive expectations and perceptive satisfaction support adoption planning, training design and customer discovery.
Read the studyFaster personalized interaction
Mean time decreased from 166.12 to 124.02 seconds with assertiveness-based communication; p = 0.005 and r = 0.49.
Read the CHI studyEvidence boundary: these figures support validation partnerships and workflow modeling. They should not be presented as regulatory clearance, clinical efficacy, fewer biopsies, reduced mortality, improved cancer detection or realized savings.
Human oversight and adoption
Commercial deployment depends on controlled validation
The distinction between assistance and authority is essential for safe adoption and credible business development.
The workflow may support
- Organizing multimodal case information.
- Providing an additional AI-assisted perspective.
- Exploring second-reading and triage workflows.
- Measuring time, interaction, trust and decision changes.
- Building customer-specific budget-impact evidence.
The workflow does not replace
- A qualified clinician's interpretation.
- Institutional protocols or multidisciplinary review.
- Formal BI-RADS assessment and clinical judgement.
- Regulatory, safety or quality-management obligations.
- Emergency services, diagnosis or medical advice.
Explore the platform behind the workflow
Review its capabilities, research scope, design principles and current readiness.
Explore platform