EU guideline group backs AI-supported double reading of mammograms

In July 2023, the ECIBC guideline group suggested AI-supported double reading of screening mammograms. It found that AI support may detect one more breast cancer for every 1,000 women screened.

  • 2 for an organization

    The group found that double reading with AI support may detect one more breast cancer for every 1,000 women screened, an estimate that rests on a simulation with archived examinations.

Depth runs from 1, a convenience or time saved, to 5, a life or a livelihood preserved.

Status
Partly reported benefit.
Evidence

foundstudiedreportedclaimed

Two independent assessment reports support the benefit.

Compared with

Double reading without AI support, pooled from six studies that the guideline group reviewed. Measured.

The system’s part
The group compared double reading with and without AI support. Its figures with AI came from a simulation on archived examinations, and its figures without AI were pooled from six studies. The group advises programs to check vendor data such as the number of images used to train the algorithm, which means that the systems learn from data.
Not yet checked
LimitsCosts and harmsLearns from dataLater outcome

Download data Cite Suggest a correction

Benefit type
Improved Task Accuracy
System
AI support for double reading of screening mammograms: three commercial algorithms kept anonymous in the study behind the estimate, with Transpara, Lunit INSIGHT MMG, Mia and a DeepHealth model in the other studies reviewed
Made by
Three vendors that asked to remain anonymous in the study behind the estimate, with ScreenPoint Medical (Transpara), Lunit (INSIGHT MMG), Kheiron Medical Technologies (Mia) and DeepHealth in the other studies reviewed
Used by
Screening services in the studies that the ECIBC guideline group reviewed, with Karolinska University Hospital's archived examinations behind the estimate and organized screening programs in the EU as intended users
Where
European Union (guideline), with the study behind the estimate in Stockholm, Sweden
Stage
Tested in a simulation
Date
, when the source was published
Kind
Aggregate result
Broad benefit
Identifying Objects, Actions, and Events
Collections
Scrutiny and findings
Last revised
1 November 2026

Sources

  1. Peer-reviewed study, by authors with no tie to the system. Context or contrary evidence.

    Salim M, Wåhlin E, Dembrower K, Azavedo E, Foukakis T, Liu Y, Smith K, Eklund M, Strand F. External Evaluation of 3 Commercial Artificial Intelligence Algorithms for Independent Assessment of Screening Mammograms. JAMA Oncology 2020;6(10):1581-1588, published online 27 August 2020. doi:10.1001/jamaoncol.2020.3321. Methods and Table 3. Funded by Stockholm County Council. The three developers provided their algorithms with no exchange of funds.

    The vendors asked to remain anonymous, with the possibility for each vendor to later decide to waive anonymity.

  2. Assessment report, by authors with no tie to the system.

    European Commission Initiative on Breast Cancer (ECIBC), Guidelines Development Group, European Commission Joint Research Centre. European guidelines on breast cancer screening and diagnosis, printed overview of the recommendations. Breast cancer screening, 4. Use of artificial intelligence, recommendation 4.2. Issued July 2023.

    The ECIBC's Guidelines Development Group (GDG) suggests to use double reading (with consensus or arbitration for discordant readings) supported by artificial intelligence (AI) over double reading (with consensus or arbitration for discordant readings) without AI support

  3. Assessment report, by authors with no tie to the system.

    European Commission Initiative on Breast Cancer (ECIBC), Guidelines Development Group. European guidelines on breast cancer screening and diagnosis, 4. Use of artificial intelligence, recommendations 4.1 and 4.2 with their justifications. Issued July 2023. Supporting material: GRADE Evidence-to-Decision Framework for recommendation 4.2 (https://dbep.ec.gdt.evidenceprime.com/profile/PNED9kXB-h0), which cites Salim et al. 2020 for double reading with AI support and pools that study with Dahlblom et al. 2022, Dembrower et al. 2023, Larsen et al. 2022, Marinovich et al. 2023 and Sharma et al. 2023 for double reading without it.

    However, there may be one more breast cancer detected for every 1 000 women screened when double reading is supported by AI, which outweighed the false positives.

Details

AI system

Workflow

Commercial algorithms scored archived screening mammograms from Karolinska University Hospital for the suspicion of cancer. The study counted an examination as abnormal when an algorithm or either radiologist flagged it.

Follow-up

Conditions for replication

Programs check the legal context and run quality control of the AI system before they adopt it. Programs can sustain the added cost of AI support.

Entry details

Sector

Healthcare.

Purpose

Commercial vendors, including ScreenPoint Medical, Lunit, Kheiron Medical Technologies and DeepHealth, built AI software to score screening mammograms for the suspicion of cancer.

30 more questions not yet checked

model, release, provider, role, configuration, inputs and outputs, autonomy, human oversight, beneficiaries, reach, distribution and access, duration, every measurement question, every money question, ownership and control, dependencies, cross-border connections, geopolitical implications, replication attempts, availability, institutional decision, grounds, action and result, success criterion, every evidence question, observed or intended, alternative assessments, related entries, other identifiers, category mappings.

Cite this record

Peter H. Vartanian. (2026). EU guideline group backs AI-supported double reading of mammograms (AIBD-1140). Artificial Intelligence Benefits Database, edition of 1 November 2026. https://petervartanian.xyz/aibd/records/AIBD-1140.html (archived at https://doi.org/10.7910/DVN/AADSII, version 7.0)

BibTeX Data Field definitions