Measuring scan durations and predicting them with their uncertainty

Slot lengths in the appointment book are often based on experience values that were set once and rarely revisited. We measure how long exams actually take on your scanners – and model how much that duration varies and how certain the estimate is. That is what makes the calculation per scanner hour reliable: if you only know an average, you only know the contribution margin approximately.

Duration as a range, not a single number

An exam type does not have a fixed duration. We therefore express it as a probability range – in the form “80 % of these exams fall between X and Y”, where X and Y come from your own data. On top of that comes how certain that range is: an exam type with few cases is estimated more cautiously than one that occurs daily.

What influences the duration

• Protocol and scanner: which sequences are actually run, on which scanner, in which operating window
• Exam-related attributes (pseudonymised): contrast administration, limited ability to be positioned or immobility, claustrophobia, inpatient or outpatient
• Day dynamics: late arrivals, no-shows, gaps that open up, urgent cases fitted in, and delays that propagate to subsequent appointments

These factors explain part of the variation. What they do not explain is reported as residual uncertainty rather than averaged away.

Data we typically use

• DICOM header timestamps from MRI and CT modalities (study, series and acquisition times), pseudonymised
• Protocol and sequence names from the DICOM headers
• RIS exports with exam type, planned slot length and status timestamps
• selected exam-related attributes from the RIS (e.g. contrast administration, inpatient or outpatient), pseudonymised

Timestamps are only as good as their source: clock drift between scanners or manually set RIS statuses are explicitly checked during analysis. The underlying metrics are explained in Throughput KPIs for MRI and CT.

Outcome

• Dashboard with measured exam durations by exam type and scanner, each with its spread
• Slot length proposals based on the measured distribution rather than on the mean
• Prediction of the expected duration as a range with a statement of uncertainty, usable in scheduling
• Foundation for cost and revenue per scanner hour, protocol prioritisation and waiting times

How it runs

1. Analysis: evaluation of a period of historical exams and comparison with planned slots.
2. Report and workshop: measured distributions and slot proposals, reviewed with management and the radiographer team.
3. Implementation (optional): continuous measurement and duration forecasting, so that changes to scanners or protocols reach the schedule.