ESTRO 2025 - Abstract Book

S3802

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2025

2825

Digital Poster Tumor nuclear size as a biomarker for post-radiotherapy survival in gynecological malignancy: development of a multivariable prediction model Yujing Zou 1 , Harry Glickman 1 , Manuela Pelmus 2 , Farhad Maleki 3 , Boris Bahoric 4 , Magali Lecavalier-Barsoum 4 , Shirin Abbasinejad Enger 1,5 1 Medical Physics Unit, Department of Oncology, Faculty of Medicine, McGill University, Montreal, Canada. 2 Department of Pathology, Faculty of Medicine, McGill University, Montreal, Canada. 3 Department of Computer Science, University of Calgary, Calgary, Canada. 4 Department of Radiation Oncology, Jewish General Hospital, Montreal, Canada. 5 Lady Davis Institute for Medical Research, Jewish General Hospital, Montreal, Canada Purpose/Objective: Radiotherapy aims to destroy cancer cells by targeting their nuclei, which contain the DNA essential for cellular function and survival. Tumor radiosensitivity may be influenced by the relative size of the cancer cell nuclei. This study aimed to investigate a potential causal link between tumor nuclei size distribution features and post radiotherapy patient outcomes in gynecological squamous cell carcinoma (SCC). Multivariate gynecological SCC mortality and progression prediction models were developed and validated using Cox proportional hazards (CoxPH) analysis. Material/Methods: Our cohort consisted of gynecological SCC patients from Montreal’s Jewish General Hospital (n = 45) and The Cancer Genome Atlas - Cervical Squamous Cell Carcinoma and Endocervical Adenocarcinoma 1 data (n = 134). All included patients were non-metastatic, received radiotherapy, and had pre-treatment diagnostic whole slide images (WSIs) available. Quality-assured binary nuclei masks 2 were overlaid onto WSI segmentations (tumor, stroma, inflammatory, necrosis, and others) from a pre-trained U-Net with a ResNet50 backbone 3 , generating nuclei masks for each tissue class (Figure 1). Each WSI contained an average of approximately half a million nuclei. Tumour nucleus size distribution mean, tumour nucleus size distribution standard deviation, patient age, and clinical stage dichotomized into early and locally advanced disease 4 were used to predict overall survival (OS) and progression free interval (PFI).

Results: Multivariate CoxPH analysis results (Table 1) revealed a potential protective effect of nucleus size distribution mean for OS (HR 0.39, 95% CI 0.15 - 1.02, p = 0.064) and PFI (HR 0.35, 95% CI 0.13 - 0.92, p = 0.040), and a harmful effect of nucleus size distribution standard deviation for OS (HR 10.76, 95% CI 1.72 - 67.27, p = 0.016) and PFI (HR 13.75, 95% CI 2.21 - 85.62, p = 0.0083). The bootstrap-validated C-statistic was 0.61 (optimism 0.05) for OS and 0.58 (optimism 0.05) for PFI, and graphical calibration showed underestimation of early OS and PFI and overestimation of late OS and PFI.