Implementation of a knowledge-based decision support system for treatment plan auditing through automation.

Area of Science:

• Medical Physics
• Radiotherapy Quality Assurance
• Machine Learning in Healthcare

Background:

• Independent auditing is crucial for radiotherapy quality assurance (QA) and continuous quality improvement (QI).
• Manual, time-intensive audits of cross-campus treatment plans were performed annually by senior physicists.
• The goal was to standardize planning, update policies, and provide training.

Purpose of the Study:

• To develop a knowledge-based automated anomaly-detection algorithm for decision support in retrospective radiotherapy plan auditing.
• To standardize and enhance the efficiency of external beam radiotherapy (EBRT) treatment plan assessment across eight campuses.

Main Methods:

• 843 EBRT plans from 721 lung patients were automatically acquired.
• 44 parameters were extracted and pre-processed from each plan.
• An "isolation forest" (iForest) algorithm identified anomalous plans based on an anomaly score.

Main Results:

• 75.6% of plans with the highest iForest anomaly scores indicated concerning qualities needing actionable recommendations.
• Auditing time decreased from 20.8 min (manual) to 14.0 min (iForest-guided), saving 6.8 min per chart.
• Annual time savings of approximately 30 hours were estimated for 250 audited charts.

Conclusions:

• iForest effectively detects anomalous radiotherapy plans, strengthening manual auditing with decision support and improving standardization.
• Automation enhances efficiency, making this method suitable for establishing a standard, more frequent plan auditing procedure.