Human-centered artificial intelligence for safer, clearer, and more efficient breast cancer screening.
Website · Our story · Evidence and reports · YouTube
BreastScreening-AI is a multidisciplinary initiative developed by SensiPerception, Lda. We bring together artificial intelligence, medical imaging, human-computer interaction, and clinical practice to investigate how intelligent decision support can help clinicians interpret breast examinations with greater confidence and efficiency.
Our work is built around a simple principle: AI should strengthen clinical judgment, not replace it. We therefore study the complete interaction between clinician, patient data, imaging modalities, workflow, and algorithm rather than treating model performance as the only measure of success.
Clinical notice: BreastScreening-AI is a research and decision-support initiative. The software and information presented in our public repositories do not provide medical advice and must not be used as a substitute for qualified clinical judgment.
Our story began in 2015, not with an algorithm in isolation, but with clinicians and their working environment. On 20 November 2015, the earliest documented fieldwork took place at Hospital Amadora-Sintra. The team observed a radiologist's workflow and discussed how to annotate lesion contours and BI-RADS findings across mammography, ultrasound, and MRI, compare both breasts, navigate image slices, preserve screen space, and follow findings over time.
Those early observations established principles that still guide us: begin with clinical practice, design with users, connect multiple imaging modalities, make system state understandable, and treat AI as support for professional judgment. The work became the Medical Imaging Multimodality Breast Cancer Diagnosis User Interface (MIMBCD-UI), a collaboration involving ISR-Lisboa/LARSyS, ITI/LARSyS, and INESC-ID at Instituto Superior Técnico and the University of Lisbon.
During 2016, the project moved from field notes into a structured research programme. The team coordinated its work through GitHub, completed state-of-the-art reviews, developed the first prototypes and master's research, and prepared clinical evaluation methods grounded in Human-Computer Interaction and User-Centered Design.
MIMBCD-UI subsequently became the research precursor to two complementary initiatives. MIDA explored intelligent agents, explainable assistance, clinician trust, workload, and human-AI interaction. BreastScreening connected this work with automated multimodal medical-image analysis and collaborations involving Instituto Superior Técnico, the University of Adelaide, and the University of Queensland.
The lineage continued through national research projects, peer-reviewed studies, prototypes, datasets, evaluation instruments, and a doctoral programme on human-centered intelligent agents. In 2022, the FCT-funded MIA-BREAST project advanced multiple-instance attention learning for multimodal breast cancer analysis. In 2025, the FCT-funded AI-Radiologist project extended the work into structured reporting and clinical research with CHTMAD / ULSTMAD.
BreastScreening-AI, developed by SensiPerception, Lda., is the translational continuation of that decade of work. It brings the earlier scientific, clinical, interaction-design, and technical streams together around a path from research prototypes toward responsibly validated clinical decision support.
| Period | Evolution |
|---|---|
| 2015 | Initial clinical fieldwork at Hospital Amadora-Sintra defined multimodal annotation, BI-RADS, lesion follow-up, PACS, and user-centered workflow requirements. |
| 2016 | MIMBCD-UI became a structured research and GitHub collaboration, with early reports, literature review, prototypes, and master's research. |
| 2017-2019 | MIDA expanded the work toward AI-assisted diagnosis and human-AI interaction; BreastScreening connected it with multimodal deep-learning research and international collaboration. |
| 2020-2022 | Peer-reviewed and multi-institution studies evaluated multimodality, workflow, adoption, diagnostic support, cognitive workload, and clinician-AI collaboration. MIA-BREAST began with FCT support in 2022. |
| 2023-2024 | Research expanded into multimodal fusion, weakly supervised learning, personalized intelligent agents, assertive communication, and intellectual property. |
| 2025-2026 | BreastScreening-AI and AI-Radiologist advanced clinical validation, structured reporting, hospital integration studies, regulatory planning, public funding delivery, and European innovation programmes. |
A controlled study involving 45 clinicians from nine institutions reported fewer false-positive and false-negative decisions, shorter diagnosis time, and strong clinician acceptance when using the AI-assisted workflow. These results are part of a broader evidence programme; they do not by themselves establish clinical effectiveness in production.
Selected publication: Artificial Intelligence in Medicine, 2022.
Our work connects five complementary areas:
- Multimodal decision support: bringing mammography, ultrasound, MRI, clinical history, and structured findings into coherent clinical workflows.
- Human-centered AI: designing understandable, inspectable, and appropriately assertive assistance for healthcare professionals.
- Clinical workflow research: evaluating accuracy, workload, time, decision stability, reporting, and human-machine readability in realistic settings.
- Responsible translation: progressing through validation, interoperability, regulatory planning, privacy, cybersecurity, and quality-management activities.
- Scientific and public communication: sharing the initiative's development, evidence, limitations, and funding context clearly.
The core programme combines the research interface, AI-assisted clinical reasoning, analytics, and the pathway toward an integrated clinical decision-support platform. MIMBCD-UI, MIDA, and BreastScreening established its multimodal, intelligent-agent, machine-learning, and interaction-design foundations.
The FCT-funded Multiple Instance Attention Learning for Multimodal Breast Cancer (MIA-BREAST) project, reference 2022.04485.PTDC, advanced multimodal learning research with contributions from the MIMBCD-UI lineage.
Exploratory work with Hospital da Luz examines AI-assisted triage, clinician decision stability, and workflow integration. Early findings are encouraging, but the sample is limited and the results remain subject to further validation. BI-RADS and clinician assessment remain the primary clinical references.
The FCT-supported AI-Radiologist project studies structured reporting, human-machine readability, and responsible AI support using ethics-approved and anonymized clinical research processes. Quantitative findings will only be reported after the relevant analyses are consolidated.
Project reference: 2024.07344.IACDC.
Startup Voucher support helps move the initiative from research toward organizational, technical, and market readiness. Public reporting, project visibility, and dissemination are developed in accordance with Portugal's Recovery and Resilience Plan (PRR) and European Union funding requirements.
Learn more at Recuperar Portugal.
Our European innovation work connects product development, clinical validation, regulatory strategy, intellectual property, and commercialization planning. It has included activities related to the EIC Accelerator Challenges, EIC Pre-Accelerator, EIC Pathfinder, and Horizon Europe.
These initiatives are related rather than isolated: clinical evidence informs regulatory planning; regulatory and IP work shape product development; public and European programmes support the transition from research maturity toward wider validation and deployment readiness.
We work with clinical, scientific, legal, regulatory, intellectual-property, funding, and innovation specialists. Their roles support different parts of the same translation pathway:
| Organization | Contribution |
|---|---|
| SNAP | European proposal development, EIC project management, and Grant Agreement Preparation support. |
| Leyton | Startup Voucher and PRR reporting, cost eligibility, and communications-compliance support. |
| SAVEAS | Intellectual-property strategy, freedom-to-operate work, and related consultancy. |
| KGSA | Legal provider supporting corporate, contractual, and broader legal matters. |
| Complear | Regulatory-strategy and independent-validation discussions for health-technology development. |
| AAVANZ | Preparation support for EIC Pathfinder and Horizon Europe proposals. |
The nature and scope of each collaboration may differ by project. References here describe their contribution to our development journey and should not be interpreted as clinical endorsement.
Our current work reflects a transition from TRL 5 toward TRL 6: validating integrated technology in relevant clinical environments while strengthening evidence, interoperability, quality, regulatory, and deployment processes.
This is a project-level maturity assessment, not a regulatory certification or authorization for clinical use.
- BreastScreening: On the Use of Multi-Modality in Medical Imaging Diagnosis — AVI 2020
- Human-centric artificial intelligence for breast cancer screening — IJHCS 2021
- Artificial intelligence in breast screening: A study of intelligent agents for improved diagnosis — Artificial Intelligence in Medicine 2022
- Technology adoption in breast cancer screening — IJHCS 2022
- Assertiveness-based agent communication for a personalized medicine on medical imaging diagnosis — CHI 2023
See our story and evidence reports for additional context.
BreastScreening-AI is connected to a wider open research history. The organizations below capture different stages and questions within the same programme:
| Organization | Role in the story |
|---|---|
| MIMBCD-UI | The original 2015 initiative and shared foundation for multimodal interfaces, annotation, clinical workflow research, datasets, manuals, and evaluation tools. |
| mida-project | Medical Imaging Diagnosis Assistant research on intelligent agents, explainable AI, clinician experience, trust, workload, and adoption. |
| BreastScreening | Automated multimodal breast-image analysis, deep learning, datasets, and international research collaboration. |
| BreastScreeningAI | The translational programme connecting the research lineage with clinical validation, product development, reporting, and deployment readiness. |
- MIMBCD-UI/meta — history, scope, and shared project context.
- MIMBCD-UI/prototype-multi-modality — multimodal lesion annotation and segmentation prototype associated with AVI 2020.
- MIMBCD-UI/prototype-cornerstone — medical-image viewer experimentation based on CornerstoneJS.
- MIMBCD-UI/data-pipeline — tooling for medical-imaging data processing.
- MIMBCD-UI/user-manuals and technical-manuals — user-facing and technical documentation.
- MIMBCD-UI/nasa-tlx and sus — reusable workload and usability evaluation instruments.
- mida-project/prototype-multi-modality-assistant — multimodal AI-assistance prototype.
- mida-project/prototype-heatmap — explainable-AI heatmap research for clinical workflows.
- mida-project/phd_thesis_mida — doctoral research materials on human-centered intelligent agents.
- mida-project/dots — Dimensions of Trust Scale materials.
- BreastScreening/meta and BreastScreening.github.io — project context and public research site.
| Repository | Purpose |
|---|---|
| breastscreeningai.github.io | Public website, story, reports, and project communication. |
| meta | Organization overview, shared context, and community entry point. |
| prototype-assertive-proactive | Research prototype for assertive and proactive intelligent-agent interaction. |
| redirect-breastscreeningai-pt | Portuguese-domain redirect infrastructure. |
| redirect-breastscreeningai-com | International-domain redirect infrastructure. |
We welcome responsible collaboration with clinicians, researchers, hospitals, patient representatives, engineers, designers, and specialists in medical-device regulation and health-technology evaluation.
- Explore the organization repositories
- Watch our project video
- Contact us at info@breastscreeningai.com
Please do not submit patient records, medical images, protected health information, credentials, or other sensitive data through public issues, discussions, or email.
Parts of this journey have received or pursued support through Portugal's Startup Voucher and PRR, the European Union, Fundação para a Ciência e a Tecnologia (FCT), and European Innovation Council and Horizon Europe programmes. Each funded activity remains subject to its specific agreement, eligibility rules, reporting obligations, and acknowledgement requirements.
Unless a repository states otherwise, consult its own license before using its contents. This repository is available under the MIT License.