MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Evaluation of Three Different Smart Assistants for Treatment Planning of VMAT Head and Neck Patients

R McBeth*, M Lin , D Nguyen , S Jiang , Medical Artificial Intelligence and Automation (MAIA) Laboratory, Department of Radiation Oncology, University of Texas Southwestern Medical Center, Dallas, Texas, USA

Presentations

(Sunday, 7/14/2019)  

Room: ePoster Forums

Purpose: To evaluate the strength and weaknesses of three machine-learning/deep-learning treatment planning assistants being implemented in our clinic.

Methods: Models 1 and 2 are commercial products. Model 1 extracts the features of the current patient and predicts the planning objectives based on relationships between features and objectives learned from historical patient data. Model 2 utilizes machine-learning to define the planning objectives by matching the new patient to historical patients. Model 3 is an in house deep-learning dose prediction algorithm. 500 clinical plans were used for training and twenty patients were used for testing and evaluating the three models. The dose-volume constraints of high-impact OARs predicted by the three models were compared with the delivered clinical plans.

Results: In general, models 1 and 3 produce plans representing the average of the plans used to train the model. Model 2 matches more closely to the clinical intention of the physician by selecting the nearest treated patient. Models 2 and 3 are more flexible in accepting a wide range of plans with different clinical goals while model 1 requires the plans to be carefully selected for more specific prescription dose schemes. Model 1 lacks transparency in model training but provides the user with detailed statistics and a range of DVH curves. Model 2 is a proprietary solution but allows the user to clearly and effectively review the previously treated match cases. Model 3 is the most transparent of the solutions for our clinic as it is developed in house and allows for easy modification.

Conclusion: In general, all of the tools evaluated are valuable additions that can increase the speed, consistency and quality of clinical head and neck plans. Each method had differences in the output of the model but all created plans that are very near to a clinically acceptable plan.

Keywords

Not Applicable / None Entered.

Taxonomy

Not Applicable / None Entered.

Contact Email