ESTRO 2023 - Abstract Book

S252

Saturday 13 May

ESTRO 2023

Results For CE, median Dice scores were 0.81 (95% CI 0.71-0.83) and 0.82 (95% CI 0.74-0.84), while median HD95 were 5.91 (95% CI 2.8-16.4) and 3.16 (95% CI 2.8-7.1) for Operator-1 and Operator-2, respectively (Figure 2 A). For NE, median Dice scores were 0.65 (95% CI 0.56-0,69) and 0.63 (95% CI 0.57-0.67), while median HD95 were 16.1 (95% CI 10.6-22.2) and 16.7 (95% CI 9.4-23.2), respectively (Figure 2 C). Comparing volume sizes, we found excellent ICC of 0.90 (p<0.001) and 0.95 (p<0.001), for CE, respectively, and 0.97 (p<0.001) and 0.90 (p<0.001), for NE, respectively. Moreover, there was a strong Spearman’s correlation of 0.83 (p<0.001) between RANO-volumes and HD-GLIO-volumes. Taken together, we found that for CE-volumes, the Dice similarity coefficients and HD95 had better scores between operator and HD-GLIO segmentation, than for inter-operator scores. This indicates that the HD-GLIO segments had a shape and location somewhat intermediate between the Operator-1 and Operator-2 manual delineations. Adding dilations further increased the Dice-scores and reduced the relative performance difference between individuals. For NE-volumes, we found that Dice similarity and HD95 showed poorer agreement between operator and HD-GLIO than inter-operator scores. This was largely because manual NE-delineations held substantially larger volumes than HD-GLIO predictions Average processing time was < 6 minutes per dataset.

Conclusion HD-GLIO deep learning predictions demonstrated high geometrical agreement with manual delineations in segmenting glioblastoma tumor compartments on standard multi-parametric MRI. We therefore find HD-GLIO feasible as an oncologist support tool for target delineation. PD-0317 Introduction of AI segmentation to drive improvements in Breast Cancer radiotherapy A. Wowk 1 1 Northern Centre for Cancer Care, Freeman Hospital, Dose Planning, Newcastle-Upon-Tyne, United Kingdom Purpose or Objective An evaluation of AI auto-segmentation for routine clinical practice in breast radiotherapy planning was carried out. Analysis of technique improvements, clinical outcomes and efficiency gains within the department is presented. Materials and Methods We introduced AI segmentation (MVision AI, Helsinki) models for all breast patients requiring nodal therapy, planned using Raystation 9B Planning System (RaySearch Laboratories). AI assisted in the generation of volumes for breast lymph nodes (as per ESTRO guidelines) as well as creation of breast or chest wall CTVs. AI contours were used for field definition, plan optimisation and dose statistics, and replaced the traditional field-based target structure for both the SMLC (Segmental Multi Leaf Collimation) breast planning technique and a newly introduced VMAT technique for Internal Mammary Node (IMN) positive disease. Results AI has facilitated nodal outlining in routine clinical practice, as the demands on clinician time previously limited its use. From an initial review of nodal volumes for 30 patients, clinicians assessed that 67% required no edits or minor edits only. For the SMLC planning technique, AI volumes have allowed Dosimetrists to individualise the match plane based on nodal coverage, rather than the traditional method using bony anatomy, reducing need for clinician review before plan optimisation. We have therefore removed this review activity (Figure 1), and clinicians and dosimetrists report efficiencies when using AI for field placement.