Metrics

The result field encapsulates the comprehensive results of the diagnostic process. Three subfields describe the classifier's behaviour for this case: metrics, preliminaryFindings, and iaSeconds.

• result.id: UUID of the result record. Stable for a given diagnostic report; useful for client-side caching and deduplication.

Metrics

metrics reports classifier performance for the case. All values are percentages from 0 to 100, and any of them can be null if the metric is not applicable to this body site or pathology.

• sensitivity and specificity: classifier accuracy expressed as percentages.
• entropy: how confident the AI is in its own response. Entropy is computed over the probability distribution of the diagnostic hypotheses; lower entropy means higher confidence, and higher entropy means lower confidence. Triage flows often use entropy as an uncertainty gate.

"metrics": {
  "sensitivity": 83.31,
  "specificity": 99.53,
  "entropy": null
}

Preliminary findings

preliminaryFindings is a set of preliminary suspicions and their likelihoods, expressed as percentages from 0 to 100. Any field can be null when not applicable to the case.

• hasCondition: likelihood that a condition is present.
• malignancy: likelihood that the condition is malignant.
• adjustedMalignancy: malignancy probability after applying clinical adjustments (for example, lesion-density or context-based corrections). A number between 0 and 100 when the adjustment model has run for the case, or null otherwise. Note: in the FHIR view this field is hard-coded null; the raw view is the only place where a numeric value can appear.
• pigmentedLesion: likelihood that the lesion is pigmented.
• urgentReferral: probability that the patient should be seen by a doctor within the next 48 hours. Higher values indicate a stronger recommendation to bring the patient forward into urgent care.
• highPriorityReferral: probability that the patient should be seen by a doctor within the next 15 days. It complements urgentReferral by capturing cases that are not 48-hour-urgent but still warrant a fast-tracked appointment ahead of the routine queue.

"preliminaryFindings": {
  "hasCondition": 100,
  "malignancy": 0.02,
  "adjustedMalignancy": null,
  "pigmentedLesion": 0,
  "urgentReferral": 100,
  "highPriorityReferral": 100
}

iaSeconds

iaSeconds: wall-clock seconds the inference pipeline spent processing the case (image plus AI calls). Number with decimals; can be null if timing was not recorded.