ESTRO 2025 - Abstract Book

S3749

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2025

1444

Digital Poster Texture analysis of optical coherence tomography angiography for detecting microstructural changes in skin cancer lesions post kV-based radiotherapy Gerd Heilemann 1 , Giulia Rotunno 2 , Lisa Krainz 3 , Mengyang Liu 3 , Richard Haindl 3 , Flavia Lo Bue 2 , Kristen Meiburger 2 , Wolfgang Drexler 3 , Dietmar Georg 1 , Christoph Müller 4 , Cora Waldstein 1 1 Department of Radiation Oncology, Medical University of Vienna, Vienna, Austria. 2 PolitoBIOMed Lab, Department of Electronics and Telecommunications, Politecnico di Torino, Torino, Italy. 3 Center for Medical Physics and Biomedical Engineering, Medical University of Vienna, Vienna, Austria. 4 Department of Dermatology, Medical University of Vienna, Vienna, Austria Purpose/Objective: To evaluate the potential of optical coherence tomography angiography (OCTA) texture analysis in detecting microstructural changes in skin cancer lesions of patients undergoing kV-based radiotherapy. Material/Methods: Twelve lesions in six patients with diagnosed skin cancer treated with kV-based radiotherapy were included in this study. A swept-source OCTA system was used to non-invasively and label-free capture the morphology of the superficial microvasculature in skin lesions. The imaging system provides a lateral resolution of 31.5 μm and an axial resolution of 27.3 μm. An area of 1 cm² was imaged, reaching a depth of 1-1.5 mm per acquisition. Lesions were scanned using OCTA at three time points: pre-irradiation, immediately post-radiotherapy, and three months after treatment completion. Each OCTA acquisition produced a volume of 120 × 512 × 512 pixels, with the first dimension representing depth. OCTA images were processed through z-score normalization and resampled to one third of their original size. Central regions of interest (80 × 480 × 440) were defined within each lesion to minimize artifacts inherent to OCTA. First-order statistical features and texture features were extracted from the processed images using pyradiomics (v3.1, Python). Feature selection was performed using LASSO regression (α=0.001), retaining non-zero coefficient features. Selected features were standardized, and classification was performed using an XGBoost model. Bootstrap resampling (n=100) with Out-of-Bag (OOB) validation was applied to evaluate model robustness. Receiver Operating Characteristic (ROC) analysis was conducted on OOB samples, with the Area Under the Curve (AUC) computed for each bootstrap iteration. Mean AUC and standard deviation were reported to assess classification performance. Results: A combination of first-order and texture features effectively differentiated between pre-irradiation and immediately post-irradiation OCTA images. These features revealed changes due to radiotherapy, indicating detectable alterations in the lesion microstructure through OCTA texture analysis. Classification using XGBoost, trained on LASSO-selected features, achieved a mean ROC with an AUC of 0.70 ± 0.19, demonstrating moderate discrimination capability and robustness across bootstrap iterations. Among the selected features, the most relevant were texture based metrics, including original_ngtdm_Busyness (importance: 0.63), followed by first-order features such as original_firstorder_Skewness (importance: 0.10) and texture metrics like original_glcm_ClusterShade (importance: 0.08). Conclusion: The radiomic feature analysis of OCTA images showed promise as a non-invasive technique for detecting changes in skin cancer lesions after radiotherapy. Moreover, this proof-of-concept study highlights its potential as a tool for monitoring treatment response in skin cancer patients, possibly contributing to personalized patient care in the future.

Keywords: OCTA, Skin cancer, Radiomics