Curated outputs
Four journal articles, five conference papers, one scientific exhibit, two theses and two patent families.
Review the catalogueEvidence
Overall evidence across research and clinical pilots
A machine-learning and clinical evidence view of BreastScreening-AI across model development, human-AI design, controlled evaluation and exploratory clinical integration. Measured findings are separated from scale-up estimates.
Research portfolio
Project-Related Evidence
The portfolio combines scholarly outputs, protected intellectual property, human-AI studies and hospital engagement. Study participations may overlap across publications and are not presented as unique clinicians.
Scholarly outputs
Peer-reviewed papers, scientific exhibit and academic theses connected to the project's research lineage.
Explore publicationsStudy participations
Reported participation across human-centred, adoption, clinician-AI and exploratory pilot studies. Cohorts may overlap.
Trace the research lineageHospitals engaged
Internal business-development traction across hospital conversations, research engagement and collaboration activity; not 25 clinical validation sites.
Review documented pilot activityEvidence layers
One programme, several forms of validation
The Evidence page now focuses on how the research components fit together. Detailed performance figures remain on the Platform, Workflow, Publications and Voucher pages where their study context is easier to preserve.
Model evidence
Multimodal fusion, weak supervision, lesion detection and external validation establish the technical research foundation.
Explore platform evidenceHuman-AI evidence
Interaction, adoption, explanation and personalization studies examine how clinicians understand and use AI assistance.
Explore publicationsClinical integration evidence
Exploratory activities investigate workflow integration, usability, structured reporting and evidence-generation readiness.
Startup Voucher ReportML validation journey
Model and design evidence developed together
The assumed timeline maps published work to five validation tracks: interface design, human factors, clinical evaluation, model validation and clinical integration. It is a synthesis of the research record, not a formal regulatory development chronology.
Foundational multimodality research
Early work established the human-centred, multimodal and intelligent-agent foundations for breast imaging decision support.
Explore publicationsControlled clinician-AI evaluation
A peer-reviewed study involved 45 clinicians from nine institutions and reported differences in errors, task time and clinician response.
Read the studyPersonalized AI communication
A CHI study evaluated assertiveness-based communication and reported faster mean task completion with a statistically significant result.
Read the studyClinical integration activities
Hospital da Luz and CHTMAD activities examined integration, usability, triage support and structured clinical reporting in relevant environments.
Open Voucher reportDefine
Intended users, modalities, decision points, endpoints and failure modes.
Develop
Model, fusion, interaction and explanation components with versioned data.
Verify
Technical performance, calibration, robustness, subgroup and error analysis.
Validate
Clinician comparison, usability, workflow and prospective clinical endpoints.
Monitor
Drift, overrides, safety signals, equity and post-deployment performance.
Validation priorities
Turn the research portfolio into decision-grade evidence
The next evidence phase should focus less on extrapolation and more on prospective, reproducible and institution-specific validation.
Lock intended use
Define target users, modalities, patient populations, outputs and clinical decision points.
Freeze evaluation plans
Prespecify endpoints, denominators, subgroup analyses, missing-data rules and statistical tests.
Validate externally
Test across independent hospitals, scanner vendors, populations and workflow conditions.
Measure workflow
Capture reading time, overrides, recalls, follow-up, workload and integration reliability.
Build economic evidence
Use observed local outcomes, costs and implementation resources rather than headline extrapolations.
Monitor continuously
Plan drift, safety, equity, cybersecurity and post-deployment performance surveillance.
Startup Voucher clinical activities
The Voucher report documents the operation, technology-readiness context, Hospital da Luz exploratory results, CHTMAD fieldwork and specialist support. It provides the project-level context behind the newest evidence shown here.
Open the complete Voucher reportTRL 5Project-level position
Progressing toward TRL 6 based on integration and usability activities. This is not an external certification, regulatory authorization or clinical deployment approval.
Evidence boundaries
What remains to be demonstrated
A credible evidence strategy must make the gaps as visible as the positive results.
Prospective performance
Larger, prespecified and prospective multicentre studies are needed to test diagnostic performance in intended-use populations.
Patient outcomes
The available studies do not yet demonstrate reduced interval cancer, morbidity, mortality or unnecessary procedures.
Generalizability
Performance requires validation across sites, modalities, scanner vendors, demographics and clinically relevant subgroups.
Workflow value
Reading time, recall, throughput and workload effects need local baseline measurement and prospective comparison.
Health economics
Cost-effectiveness, budget impact and realized return on investment have not yet been established.
Regulatory readiness
Intended use, risk management, quality systems, cybersecurity and post-market monitoring require formal development.
Review the newest project evidence
See the Startup Voucher operation for detailed pilot context, activities, limitations and supporting results.
Startup Voucher Report Explore publications