Skip to content

Clinical interpretation

CALMaR aims to make imaging-derived information usable and traceable. It should support clinical reasoning rather than convert a scan directly into a treatment decision.

Evidence layers in a report

A report should distinguish:

Layer Example
Measured or segmented Lesion mask volume
Spatial calculation Percentage overlap with an atlas region
Normative inference Estimated structural disconnection
Literature association Published association with an impairment
Clinical observation Behavioural language assessment
Decision Goal and therapy plan agreed with the person and clinician

Why behavioural assessment remains essential

People with similar lesions can show different communication profiles and recovery trajectories. Premorbid characteristics, distributed damage, comorbidities, treatment, time, environment, multilingual experience, and measurement all contribute.

Imaging can refine a hypothesis, expose a mechanism, or add prognostic context. It does not replace direct communication assessment or the person's priorities.

Group evidence and individuals

An association demonstrated across a cohort can be informative without being accurate enough for individual prediction. Before presenting an individual-level statement, ask:

  • Was the finding validated for individual prediction?
  • Does the current person resemble the study sample?
  • Are the acquisition, preprocessing, and outcome compatible?
  • Is uncertainty calibrated and visible?
  • Would an error change access to therapy or expectations about recovery?

CALMaR's knowledge base can surface candidate therapy ingredients and their supporting evidence. Ranking is not prescribing. A clinically responsible output should expose:

  • The measured feature that triggered the finding
  • The supporting citation and population
  • Evidence quality and confidence
  • Conflicting or absent evidence
  • Stroke stage and other applicability conditions
  • A statement that assessment, goals, feasibility, and clinician judgement remain necessary

Reporting uncertainty

Uncertainty can originate in the image, lesion segmentation, registration, atlas definition, normative reference, statistical association, knowledge-base entry, and clinical applicability.

Avoid collapsing these sources into one confidence score unless its construction and interpretation are defensible. Specific warnings are often more useful than a single number.