MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Detectability of Patient Anatomy Changes Via Real-Time Exit Fluence Monitoring

M Ahmed*, H Nourzadeh , W Watkins , J Siebers , University of Virginia Health System, Charlottesville, VA

Presentations

(Sunday, 7/29/2018) 3:00 PM - 6:00 PM

Room: Exhibit Hall

Purpose: To determine the detectability of significant patient anatomy changes and the distinguishability of patients via analysis of predicted and measured frame-by-frame exit-fluence images.

Methods: Data from 2 different VMAT prostate patients were utilized. For the 1st patient, we predicted gantry-angle resolved images through the planning CT yielding set P1.1 and measured gantry resolved images through the day of treatment anatomy yielding M1.2. The 2nd patient had predicted image sets through three different CTs (P2.1, P2.2, P2.3) and no measurements. The ability to discern references P1.1 and P2.1 from tests M1.2, P2.2, P2.3 was performed with various metrics using ROC analysis. Image set P2.2 has a large air pocket compared to P2.3 and P2.1.

Results: Verifying P1.1&P2.1 with M1.2 shows that with current prediction algorithms, detection of the wrong patient is possible with ~65-70% confidence using mean gamma and percent intensity difference. Gamma index (gamma < 1) improved the area under ROC curves when increasing the dose tolerance and distance-to-agreement criteria. On verifying Ref:P1.1&P2.1 and Test:P2.3, AUCs shows that minimizing the inherent noise boosts the confidence to 90%. On verifying Ref:P1.1&P2.1 and Test:P2.2, the ROC analysis illustrates poor confidence for all metrics but the Kolmogorov–Smirnov (KS) test. The KS metric is capable of distinguishing different patients and major anatomic changes, while the other metrics are unable to distinguish the difference between these.

Conclusion: Significant patient anatomic variations and patient identification can be achieved with per-frame exit fluence monitoring using the KS test, but not the other metrics. Fluence deviations strongly depends on the accuracy of the prediction algorithm and the frame noise. Robustness of prediction to normal daily anatomy changes, setup errors, and couch positioning improves SNR, which in turn improves the detectability of significant anatomical variations. 

Keywords

Not Applicable / None Entered.

Taxonomy

Not Applicable / None Entered.

Contact Email