AI in medical practice

Technology enters everyday practice

AI-enabled medical devices can identify patterns in data and provide information that supports diagnosis or treatment. FDA examples include imaging and retinal image analysis. The list of devices authorised for marketing in the United States illustrates the field's breadth; it does not establish that any AI application is suitable for clinical use. [1]

Trust requires evidence

An output must be assessed alongside the intended use, target patients and research evidence. Transparency principles from the FDA, Health Canada and MHRA emphasise information about a system's development, performance, limitations and role in professional decision-making. [2]

Potential biases in the data also matter, as does whether the intended patients are represented in the development and test datasets. Human–AI teamwork needs to be examined in the actual work environment. [2]

Evaluation continues after implementation

Some systems may change as new information becomes available. Evaluation therefore extends from development to implementation, monitoring and modification. Performance in one situation alone does not describe safety throughout the lifecycle. [1]

Questions for learning

Educational reflection questions from Akamedia

• What task is the system intended to perform?
• How has its performance been assessed in this particular patient group?
• How can users recognise uncertainty and changes in performance?

Plain-language summary

AI's value in clinical work depends on its intended use, research evidence and the professional's understanding of the limits of its output.