ClarusC64/ffr-physiological-plausibility-decay-detection-v0.1
Goal Detect when an AI-derived FFR valuebecomes physiologically implausiblegiven other modalities. The warning signal is coherence loss.Not a single bad threshold. Inputs ai_ffr_prediction myocardial_perfusion_index wall_motion_score stress_test_result vital signs (heart rate, blood pressure) physiological_coherence_score expected plausibility band Required outputs plausibility_decay_flag decay_type modality_conflict_label plausibility_drop_score Decay types Examples:… See the full description on the dataset page: https://huggingface.co/datasets/ClarusC64/ffr-physiological-plausibility-decay-detection-v0.1.
Goal
Detect when an AI-derived FFR value becomes physiologically implausible given other modalities.
The warning signal is coherence loss. Not a single bad threshold.
Inputs
- aiffrprediction
- myocardialperfusionindex
- wallmotionscore
- stresstestresult
- vital signs (heart rate, blood pressure)
- physiologicalcoherencescore
- expected plausibility band
Required outputs
- plausibilitydecayflag
- decay_type
- modalityconflictlabel
- plausibilitydropscore
Decay types
Examples:
- perfusion-discordant perfusion looks normal but FFR implies severe ischemia
- wall-motion-discordant wall motion abnormal but FFR looks normal
- multi-modality-conflict multiple modalities disagree with FFR
- borderline-mismatch small but clinically relevant mismatch
Why cardiologists care
A stable-looking number can still be wrong for the patient.
This dataset flags when:
- physiology and prediction stop matching
- the reading needs verification or escalation
Evaluation
The scorer checks that the response includes:
- a binary decay flag
- a named decay type
- a conflict label
- a 0 to 1 plausibility drop score