ESTRO 2021 Abstract Book

S1552

ESTRO 2021

Conclusion There are significant differences between most images investigated, in feature values and correlations with volume and other features. There is, however, little difference between feature signature for MC scans. Features were prognostic for DR, and outperformed clinical variables, regardless of image type. However, MC scans offer no improvement over single phase images; the latter produces better prognostic models. Motion compensation reduces noise, but tumour radiomics is more impacted by loss of image detail due to residual motion after deformable registration, e.g. due to registration inaccuracies and cardiac motion. PO-1822 Feasibility of spatial normalisation for image-based data mining in breast cancer radiotherapy T. Jaikuna 1 , M. Aznar 1 , P. Hoskin 1 , M. Van Herk 1 , C. West 1 , E. Vasquez Osorio 1 1 The University of Manchester, Division of Cancer Sciences, School of Medical Sciences, Faculty of Biology, Medicine and Health, Manchester, United Kingdom Purpose or Objective The acute and late effects of breast radiotherapy (RT) are well-recognised, but few studies have investigated the impact of the 3D dose distribution in large patient cohorts. Image-based data mining (IBDM) is a new voxel-based analytical approach based on spatial normalisation of the studied population data to a common frame of reference, or reference patient, using image registration. This process is challenging for breast cancer patients as breast shapes vary considerably. This study investigates the feasibility of IBDM in breast RT, assessing the optimal number of reference patients needed and the best choice of registration metric. Materials and Methods 707 breast RT patients from the REQUITE study, treated supine and post-lumpectomy, were included. In a first step, the patients were clustered using K-means on breast cup size, band, and body mass index (BMI), varying k from 2 to 9. The optimal number of clusters was selected using Silhouette analysis. The patients corresponding to the centroid of each cluster were selected (Ref i , where i refers to the cluster number). Similarly, one single reference patient was selected for the complete cohort (“Ref 0 "). In a second step, affine registration followed by deformable image registration (DIR) was