ESTRO 2024 - Abstract Book

S601

Clinical - Breast

ESTRO 2024

Figure 1: workflow of dosiomic analysis for prediction of skin toxicity after radiotherapy.

Results:

Thirty- two patients (42%) experienced acute skin toxicity (grade ≥2) following breast radiation therapy. In terms of prediction performance, dosiomics prediction models showed superior results, achieving a higher AUC value of 0.78 (Figure 2), while DVH prediction models showed 0.66. Among the applied algorithms, SVM exhibited the best performance, with high values for Sensitivity (0.88), Specificity (0.66), Accuracy (0.77), and an AUC of 0.78 (Table 1). SVM identified 8 critical features, with Major Axis Length from the shape class being one of the most relevant features for skin toxicity (grade ≥2). Other significant features included Energy and MCC, primarily belonging to the GLCM class, as well as Breast CTV, V105cc skin, V 110% Skin, chemo regimen, Receptor State, and Hormonal Treatment.(4)