Predicting slot lengths of MRI exams to decrease observed discrepancies between planning and execution.

Area of Science:

• Radiology
• Medical Imaging
• Health Informatics

Background:

• Planned slot times (T_plan) for abdomen MRI exams often do not match actual scan durations (T_act).
• Discrepancies lead to scheduling inefficiencies, with 30% of exams exceeding planned durations and others having unnecessarily long slots.

Purpose of the Study:

• To identify discrepancies between planned and actual abdomen MRI slot lengths.
• To develop a machine learning model for predicting individual MRI exam slot lengths to improve scheduling accuracy.

Main Methods:

• Retrospective analysis of 3038 abdomen MRI exams across 17 protocols.
• Training a Random Forest Regression model on historical data to predict slot lengths (T_pred) based on patient and exam context.
• Comparing T_pred with T_plan against T_act using Pearson correlation and error metrics.

Main Results:

• The Random Forest model achieved a Pearson correlation of 0.66 between predicted (T_pred) and actual (T_act) slot lengths, outperforming planned times (T_plan) with a correlation of 0.50.
• Schedule adherence improved, with a 28% reduction in root mean squared error and a 16% reduction in the standard deviation of time differences.
• Analysis of liver protocol exams indicated patient condition and sequence selection influence duration but have limited predictive value.

Conclusions:

• Machine learning-based prediction of individual MRI slot lengths significantly enhances scheduling accuracy compared to protocol-based planning.
• The developed model offers a more precise approach to managing MRI workflow and resource allocation.
• While clinical context offers insights, its direct integration into predictive models shows limited improvement potential.