ESTRO 2023 - Abstract Book

S1339

Digital Posters

ESTRO 2023

Our quantitative analysis shows a difference in the similarity indexes in the results obtained which demonstrates a statistically significant variability between the GS and the AI-corrected contours (mean DSC 0.95, mean 95HD 2.54, and mean MDC 0.66). In the independent grading system done by the 3 RO experts: grade 1 was designated to 66.67% of the GS and 78.33% of the AI-c; Grade 2 was accorded to 33.33% of the GS and 21.67% of the AI-c. No major corrections (Grade 3) were registered. Conclusion The mix between AI algorithms and ROs can successfully generate good and superior quality delineations (the AI prediction can properly generate the lateral and medial border of the breast tissue). Differences are primarily found in the cranio caudal direction between the data sets. Retraining AI algorithms on standardized reference datasets have the potential to further enhance performance, increasing the usage of AI algorithms for clinical investigations and research.

PO-1642 Gaze angle prediction for ocular proton therapy from anatomical features

D. Björkman 1 , D. Björkman 2 , R. Via 1 , A. Lomax 1,2 , G. Baroni 3 , D.C. Weber 1,4,5 , J. Hrbacek 1

1 Paul Scherrer Institute, Center for Proton Therapy, Villigen, Switzerland; 2 ETH Zurich, Department of Physics, Zurich, Switzerland; 3 University Hospital Bern, Department of Radiation Oncology, Bern, Switzerland; 4 Inselspital, Bern University Hospital, University of Bern, Department of Radiation Oncology, Bern, Switzerland; 5 University Hospital Zurich, Department of Radiation Oncology, Zurich, Switzerland Purpose or Objective Proton beam delivery into the eye of a patient has proven to be a highly successive and reliable method of treating intra ocular tumours. The current clinical workflow for treatment in the Optis treatment room at the Center for Proton Therapy (CPT) at Paul Scherrer Institute in Switzerland, includes the utilization of the treatment planning system Eyeplan. Treatment plan optimization for ocular proton therapy means finding a suitable gaze angle of the eye to be treated to minimize exposure of healthy eye structures. The optimization requires time and experience and may introduce observer related biases. The methodology discussed in this paper attempts to solve the aforementioned issues by predicting the patient gaze angle based on large data set of historical treatment patient data describing the patient anatomy and corresponding treatment plans. Materials and Methods Patient gaze angle is described in terms of its polar and azimuth angles, which are inferred from predicting the position of the fixation light. Predictions of fixation point are obtained by training a set of machine learning algorithms with historical data describing both the anatomy and treatment positioning for 1091 patients. These patients all satisfied the inclusion criteria of being treated without wedge compensator and used the treated eye for patient positioning. Dose-volume histograms of relevant structures are extracted from Eyeplan and used to compare predicted plans to reference plans for a subset of the testing data set. Results The investigations indicate that the anatomical information present in the CPT historical patient database is sufficient to predict patient gaze orientation within seconds using the machine learning framework. Figure 1 shows the numerical analysis of 190 patient cases which proves that the treatment gaze angle can be predicted within 11.67 degrees for all patient cases with a median angle difference of 3.00 degrees. Furthermore, 75 percent of the predicted gaze angles are below 4.74 degrees compared to the corresponding clinical plans. The four predicted treatment plans that were investigated in further detail are shown in figure 2 and indicate similar plan quality to the clinical plan with varying relative exposure to intra-ocular organs at risk.