Support for clinicians beyond the initial diagnosis: enhancing breast imaging with AI
Artificial intelligence is moving from "nice-to-have" to a practical support tool across the breast cancer pathway. At the European Society of Breast Imaging (EUSOBI) annual scientific meeting in Aberdeen, Scotland, Professor Gerald Lip outlined where AI adds value after the initial diagnosis-especially in ultrasound and MRI-as clinicians plan and deliver treatment. European Society of Breast Imaging (EUSOBI)
Ultrasound: sharper triage, fewer unnecessary procedures
AI is improving lesion detection, characterisation, and classification in ultrasound. It can also support reader education and improve specificity-if it integrates cleanly into clinical workflow. That last point matters; if adoption adds friction, it won't stick.
Research use cases now include tumour type prediction, forecasting response to neoadjuvant therapy, and predicting axillary nodal metastases. In the UK, teams are also assessing whether AI can help avoid biopsy of clearly benign lesions in women under 35, reducing over-investigation, over-treatment, and the anxiety that comes with both.
- Clinical takeaways: focus pilots on specificity gains, biopsy reduction in benign lesions, and decision support for nodal assessment.
- Integration tip: prioritise AI that plugs into your current reporting and PACS with minimal clicks.
Contrast-enhanced mammography: promising, but data needed
There's potential for AI in contrast-enhanced mammography, particularly with synthetic contrast approaches. The signal is encouraging, but broader datasets and stronger validation are still needed before wide deployment.
MRI: stronger predictions, leaner workflows
MRI is where AI may pay off fastest: prediction of tumour response, risk stratification, breast cancer characterisation, and segmentation based on radiomics. Done well, these tools can streamline workflow, letting radiologists spend more time on complex cases. They may also enable use of less contrast-or even synthetic alternatives-without sacrificing diagnostic value.
Colleagues in Aberdeen are advancing Field Cycling MRI to better characterise breast cancer, with promising results in DCIS. AI algorithms are being used in post-processing to refine characterisation and support reporting.
Pathology: the next growth area
Digital breast pathology is building momentum and could be the next major investment frontier. AI can support tumour segmentation, mitotic counting, and response prediction-key inputs for treatment planning and trial stratification. The caveat: Whole Slide Imaging requires far more storage and compute than typical radiology images, so plan infrastructure early.
Workflow tools that actually save time
Beyond imaging, voice technology, dictation tools, auto-scribing, and chatbots can trim minutes off every case. That adds up across a service. If you're exploring these tools, see how Speech-To-Text fits into your reporting stack.
- Look for tools that auto-populate key fields and structure reports without fighting your style.
- Keep a human in the loop; GenAI and LLMs help, but oversight protects quality.
Governance: move fast, stay safe
As you scale AI, build guardrails: post-market surveillance, plans for new hardware and software, procurement standards, bias and diversity checks, and clear ethics review. Keep performance dashboards visible-specificity, PPV, recall rate, time-to-report, and false-positive biopsy rate should be tracked and shared.
Toward connected, patient-centered care
"By linking pathology, MRI and radiology together and putting in AI adaptors and tools, we are going to start talking about prediction and personalisation of treatments." As Professor Lip noted, the goal is simple: "If we can enhance our technology with AI and not work in silos, we can work together to build a more patient centred and sophisticated future."
How to get started (practical checklist)
- Define the clinical questions first: detection, specificity, response prediction, or biopsy reduction?
- Map workflow integration points and measure baseline metrics before any rollout.
- Run focused pilots with clear endpoints (e.g., benign biopsy rate in women under 35, time-to-report).
- Stand up data governance and bias testing across age, ethnicity, and scanner vendors.
- Plan infrastructure for WSI and model hosting; budget for storage, GPUs, and network bandwidth.
- Invest in reader education and feedback loops; use cases that teach as they assist stick better.
- Write a post-market surveillance plan with trigger thresholds for retraining or rollback.
- Coordinate patient communication to reduce anxiety when AI influences care pathways.
Further learning
For teams building skills around clinical AI, see AI for Healthcare for practical resources, courses, and case studies that focus on workflow, safety, and outcomes.
Profile: Professor Gerald Lip
Professor Gerald Lip is Clinical Director of the North East of Scotland Breast Screening Service and Principal Investigator of the GEMINI prospective evaluation of mammography AI. He is President of the British Society of Breast Radiology and the newly appointed Head of AI in Medicine at the University of Aberdeen Medical School, reflecting his leadership in integrating AI into healthcare.
Your membership also unlocks:
