ESTRO 37 Abstract book

ESTRO 37

S547

as a weak correlation, else it is defined as a strong correlation. Results Median tumor volume was 35 cc (range 11-490 cc) and 49 cc (range 12-544 cc) for the multicenter and single center cohort respectively. Median SUV max was 15.0 (4.0-45.0) and 15.4 (6.3-33.2). In the multicenter patient cohort, 41 out of 57 PET TF were only weakly correlated to SUV max , SUV mean , and/or MATV using fixed bin counts (Table 1). This number decreased to 3 out 57 PET TF when using a fixed bin width. For the single center patient cohort 19 out of 57 PET TF were weakly correlated using fixed bin counts and no features showed a weak correlation when using a fixed bin width. Interestingly, PET TF in a large multicenter cohort had more independent textural features than the single center cohort. Figure 1 shows the results in greater detail.

Conclusion Our preliminary data demonstrate an abnormal vascular profile of brain metastases prior to SRS, followed by high inter-patient variations in vascular responses. In normal brain tissue, vascular responses were generally transient, and with no differences in response depending on prescribed SRS dose or delivered dose. PO-0985 The relationship between PET textural features, SUV metrics and metabolic active tumor volume T. Konert 1 , J. Van de Kamer 2 , W. Vogel 1 , J. Sonke 2 1 Netherlands Cancer Institute, Nuclear Medicine, Amsterdam, The Netherlands 2 Netherlands Cancer Institute, Radiation Oncology, Amsterdam, The Netherlands Purpose or Objective PET radiomics features may have prognostic or predictive value and could therefore assist treatment decisions. SUV metrics and tumor volume could be closely correlated to PET textural features (PET TF), and therefore it may be possible that resulting prognostic PET TF would rather act as a surrogate for other well-established prognostic parameters than as an independent variable. This study investigates if PET textural features could act as a prognostic factor, independent of the well-established prognostic factors SUV max , SUV mean , and metabolic active tumor volume (MATV) and their variability between two patient cohorts. Material and Methods A multicenter patient dataset (n = 268) with stage IIB-IIIB non-small cell lung cancer (NSCLC) and an independent external validation single center patient cohort (n = 41) with stage I-IIIB NSCLC were included. Pre-treatment FDG PET/CT scans were acquired on PET/CT scanners from Philips and GE. Only the primary tumor was included and a VOI was created by applying a threshold of 40% of the SUV max . Per VOI, with both in-house developed software and the Py-Radiomics toolkit, 57 second and higher-order PET TF were computed and these were compared with SUV max , SUV mean , and MATV. In order to calculate PET TF, the SUV was binned using two common methods: a fixed bin count of 64 and a fixed bin width of 0.25. Associations between PET TF and SUV max , SUV mean , and MATV were assessed with the Spearman rank correlation coefficient to determine a possible non-linear relationship. A correlation coefficient between -0.5 and 0.5 was defined

Conclusion This study demonstrates that in both cohorts, PET textural features derived using a fixed bin count are largely independent from SUV max , SUV mean , and MATV. Their independency decreases when using a fixed bin width. Clear guidelines for PET feature analysis, including SUV binning, and feature selection are necessary to avoid collinearity. The next step is to assess the prognostic value of PET textural features and its variability in multiple patient cohorts in both pre-treatment and response imaging.

Poster: Physics track: Images and analyses