Encrypted login | home

Program Information

A Predictive Planning Tool for Stereotactic Radiosurgery

no image available
S Palefsky

S Palefsky*, J Roper , E Elder , A Dhabaan , Winship Cancer Institute of Emory University, Atlanta, GA

Presentations

SU-D-BRB-1 (Sunday, July 12, 2015) 2:05 PM - 3:00 PM Room: Ballroom B


Purpose: To demonstrate the feasibility of a predictive planning tool which provides SRS planning guidance based on simple patient anatomical properties: PTV size, PTV shape and distance from critical structures.

Methods: Ten framed SRS cases treated at Winship Cancer Institute of Emory University were analyzed to extract data on PTV size, sphericity (shape), and distance from critical structures such as the brainstem and optic chiasm. The cases consisted of five pairs. Each pair consisted of two cases with a similar diagnosis (such as pituitary adenoma or arteriovenous malformation) that were treated with different techniques: DCA, or IMRS. A Naive Bayes Classifier was trained on this data to establish the conditions under which each treatment modality was used. This model was validated by classifying ten other randomly-selected cases into DCA or IMRS classes, calculating the probability of each technique, and comparing results to the treated technique.

Results: Of the ten cases used to validate the model, nine had their technique predicted correctly. The three cases treated with IMRS were all identified as such. Their probabilities of being treated with IMRS ranged between 59% and 100%. Six of the seven cases treated with DCA were correctly classified. These probabilities ranged between 51% and 95%. One case treated with DCA was incorrectly predicted to be an IMRS plan. The model’s confidence in this case was 91%.

Conclusion: These findings indicate that a predictive planning tool based on simple patient anatomical properties can predict the SRS technique used for treatment. The algorithm operated with 90% accuracy. With further validation on larger patient populations, this tool may be used clinically to guide planners in choosing an appropriate treatment technique. The prediction algorithm could also be adapted to guide selection of treatment parameters such as treatment modality and number of fields for radiotherapy across anatomical sites.



Contact Email: