MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Machine Learning Analysis of Treatment Session Time Components

R Kashani, D Smith*, C Mayo, University of Michigan, Ann Arbor, MI

Presentations

(Sunday, 7/12/2020)   [Eastern Time (GMT-4)]

Room: AAPM ePoster Library

Purpose:
Efficiency and proper allocation of treatment machine time and resources requires a full understanding of factors that impact treatment time. This study aims to utilize machine learning to determine the most important planning and delivery features that contribute to the overall treatment time.

Methods:
We studied 10,968 radiotherapy treatment courses encompassing 14,542 treated plans, representing 160,844 radiotherapy treatment sessions occurring between 2014 and 2019. Treatment session timeline components were extracted, including imaging time, image analysis wait time, and treatment delivery time. The total treatment session time was defined as the summation of these components.

Random Forest regression models were developed using Scikit-Learn. Parameters examined included: number of images (CBCT, KV, MV), number of beams, planning technique, beam parameters, treatment site, motion management strategy, use of flattening filter free vs standard beams, use of non-coplanar beams, and stereotactic body radiation therapy (SBRT) vs non-SBRT. Timeline distributions were analyzed to quantify, and visualize the impact of these parameters.


Results:
Treatment session time and treatment delivery times were 17.5±8.8 minutes and 7.7±5.8 minutes for SBRT treatments, and 8.0±5.5 minutes and 3.7±2.3 minutes for non-SBRT treatments. There were no significant differences in session time for patients treated with standard dose vs high dose rate flattening filter free 6MV beams. First fraction vs subsequent fraction session times were 21.8±10.1 vs 15.8±7.5 minutes for SBRT and 13.6±10.2 vs 7.6±4.7 minutes for non-SBRT treatments due to additional imaging. Session time for abdominal and thoracic SBRT treatments was 26.3±9.2 minutes with breath hold, compared to 16.1±6.5 minutes with free breathing. Our cross validation model predicted session times within 3.5 minutes.

Conclusion:
Impact of various treatment parameters on overall treatment session time was evaluated, providing quantitative data to help guide decisions regarding motion management and planning at the time of simulation.

Download ePoster [PDF]

Keywords

Modeling, Numerical Analysis, Radiation Therapy

Taxonomy

IM/TH- Informatics: Informatics in Therapy (general)

Contact Email