ESTRO 2025 - Abstract Book

S3826

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2025

Vall D'Hebron Institute of Oncology (VHIO), Barcelona, Spain. 19 Division of Cancer Epidemiology, German Cancer Research Center (DKFZ), Heildelberg, Germany. 20 Division of Cancer Sciences, University of Manchester, Manchester, United Kingdom Purpose/Objective: We aimed to investigate whether a combination of Functional-Data-Analysis (FDA) and Linear-Discriminant-Analysis (LDA) was superior to Equivalent-Uniform-Dose (EUD) modeling approaches when analysing the association between dose distributions and radio-induced toxicity. Material/Methods: We developed the Functional-Discriminant Approach on rectal Dose-Surface Maps (DSMs) from 1560 prostate cancer patients enrolled in an international, prospective study, with follow-ups up to 8 years (median 24-months, 75th-percentile 60-months). Patients received radical or post-prostatectomy radiotherapy according to local protocols. The endpoints were late grade≥1 proctitis, diarrhoea, and faecal incontinence. We generated DSMs using a Python tool and normalised them (latero-lateral/cranio-caudal directions) to a common We applied P-spline smoothing to reduce noise, capturing significant variations and ignoring minor fluctuations. Next, we performed FDA on the ten principal components to reduce dimensionality and extract the most significant ones. We used the Mann-Whitney U-test to assess which components significantly differed between patients with/without toxicity. We applied LDA to classify patients based on FDA scores, seeking the optimal direction maximising the separation between patients with/without toxicity. We calculated a value (t) for each map representing its projection along the LDA direction to assess how "close" or "far" a patient is from the separation between the two classes. Increasing t values indicate more proximity to the spatial dose-pattern of the toxicity class. Figure-1 presents the methods. frame based on median population characteristics. We converted all doses to EQD2Gy (α/β=1.6Gy).

Results: The observed toxicity rates were 16.2%, 13.3% and 6.8% for proctitis, diarrhoea and feacal incontinence. The most significant dose variations for feacal incontinence were in the upper rectum, while for proctitis, smaller variations were in the same region. The largest variations for diarrhoea were in the anal canal. Figure-2 shows DSMs as a function of t for incontinence, highlighting the spatial dose pattern associated with a higher risk of incontinence. t-