Explore classification, detection and segmentation as conceptual tasks. Distinguish research demonstrations from tools approved and validated for a particular clinical use.

Review how sample selection, labels and dataset splits influence evaluation. Use synthetic teaching examples to explore missing context and annotation uncertainty without interpreting patient scans.

Discuss sensitivity, specificity and error types using fictional results. Recognise that a headline accuracy figure cannot establish suitability for every population or setting.

Compare hypothetical performance across settings and sample groups. Identify questions about representative data, external validation and changes in acquisition conditions.

Map responsibilities and escalation points in fictional supervised workflows. Discuss automation bias and situations where unsupported outputs must not guide clinical action.

Organise a tool-description review covering intended use, evidence, limitations and unanswered questions. Present a non-clinical evaluation checklist for discussion with qualified specialists.