Penfield FindingsFind what requires attention.
Penfield Findings is AI-assisted abnormality detection and analysis software. It detects, localises, segments and quantifies candidate abnormalities, then tries to prove itself wrong.
Independent verification models challenge every finding. Confidence and uncertainty are shown explicitly, findings are compared across time, and each exists as a real object inside Penfield Fusion.
Discuss early accessHow Penfield Findings works
Detection that knows
when to say no.
Every finding carries a status. Findings only reports what survives verification, and abstains when the data cannot support a conclusion.
DETECTOR f–det 0.4.1→INDEPENDENT VERIFIER f–ver 0.2.3→CHALLENGE TESTS→ABSTENTION GATE
- Location
- Left temporal region
- Dimensions
- 18.4 × 12.1 × 9.7 mm
- Volume
- 1.36 cm³
- Detected on
- T1 · FLAIR
- Confidence
- 94%
- Verification
- 2 / 2 models agree
- Status
- REVIEW
A finding exists as an object inside Fusion, measured, verified and traceable to its evidence.
Illustrative findings on synthetic anatomy. Not diagnostic output.
- High confidence
- Review
- Uncertain
- Out of distribution
- Insufficient data
Detect. Verify. Challenge.
Abstain when uncertain.
Detect
Candidate abnormalities are located, segmented and quantified across the available modalities.
Verify
Independent models re-examine each candidate. Agreement is required before confidence is shown.
Challenge
Every finding is tested against competing explanations: artefacts, normal variation, out-of-distribution scans.
Abstain
When the evidence is insufficient, Findings says so. Uncertainty is exposed, never hidden.
No sentence without
a source.
Findings reports are generated only from verified, structured evidence. Each statement can be traced back to the scan, modality, coordinates, measurements, segmentation and model that produced it, then re-checked before a clinician ever reads it.
Report draft / Example
“A region of abnormal signal measuring 18.4 × 12.1 × 9.7 mm (1.36 cm³) was identified in the left temporal region across T1 and FLAIR sequences.”
- SCAN PF–001
- T1 + FLAIR
- −42, −18, −12 mm
- SEGMENTATION v2.3
- f–det 0.4.1
- CONFIDENCE 94%
- PIPELINE 2026.09
Every generated statement traces back to structured evidence. What cannot be traced is not written.
One system
Every Penfield product sits on the same path: measurement in, intervention out. Penfield Findings is one part of it.
Observe & Understand
Understand
Simulate