Patient-level separation
Our evaluation plan separates patients across development and evaluation sets to help limit data leakage.
RESEARCH & VALIDATION
Evidence builds trust.
We’re starting there.
A useful medical AI system needs more than promising demonstrations. We’re developing a research framework to characterize performance, understand limitations and evaluate its role in clinical review.
HUMAN EXPERTISE. INTELLIGENT ASSISTANCE.LUNG CT / 01
Model development underwayA framework for
A framework for
careful evaluation.
Clinical validation is planned. No performance results, clinical benefit or regulatory authorization are claimed on this website.
EVIDENCE FRAMEWORKValidation planned
Sensitivity—
Pending validation
Precision—
Pending validation
False positives / scan—
Pending validation
Subgroup performance—
Pending validation
Performance results will be published following validation.
EVALUATION METHODOLOGY
Ask the right questions.
The following are evaluation priorities, rather than completed studies or published findings.
Detection performance
Sensitivity and precision will be assessed alongside the candidate burden presented for review.
False-positive analysis
We plan to characterize unnecessary findings and the contribution of candidate verification.
False-negative analysis
Missed reference findings will be reviewed to understand limitations and guide further development.
Nodule characteristics
Planned subgroup analysis includes size, appearance and location, subject to available reference data.
Independent validation
Evaluation on independent data and clinical workflow assessment are planned before clinical use.
DATASET METHODOLOGY
Independent patients. Transparent evaluation.
Our methodology is being designed to keep development decisions separate from final evaluation, and to make the basis of results understandable.
Define the reference
Document the source, annotation process, inclusion criteria and limitations of the evaluation data.
Separate at patient level
Keep patients in distinct development and evaluation sets. Account for related studies and repeat examinations.
Report the full picture
Describe detection performance alongside false positives, missed findings and variation across relevant subgroups.
Assess clinical fit
Plan clinician-led review of candidate presentation, interpretability and the practical burden of using the system.
PUBLICATIONS & RESULTS
Evidence, when it is ready.
Results will be published following validation. Study methods, evaluation conditions and limitations will accompany the findings.
PUBLICATION STATUS
No publications listed yet.
This space will host validated results and research publications as they become available.
Discuss the evaluation frameworkLET’S BUILD THE NEXT CHAPTER
A conversation today.
A clearer view tomorrow.
We’re looking to collaborate with radiologists, imaging centers
and clinical research partners.