Quantitative assessment of radiotherapy-induced microvasculature changes in non-melanoma skin cancer using OCTA

Non-melanoma skin cancer (NMSC) is a term that refers mainly to basal cell carcinomas (BCCs) and squamous cell carcinomas (SCCs). NMSC is the most common form of cancer in Caucasians, with a growing incidence worldwide. Among the possible treatments, radiotherapy is often preferred as a tissue-preserving non-surgical option, which shows effectiveness in terms of cosmetic results and local control. Vessel morphologies vary with different tumour stages and as a result of treatment, so cutaneous blood flow can be used as a biomarker to diagnose NMSC. Optical coherence tomography angiography (OCTA) can visualize both the microstructure and vasculature of skin non-invasively and could thus be used in treatment planning and follow-up. The objective of this thesis is to understand whether feature extraction on OCTA can quantitatively assess lesion-dependent changes in skin micro-vasculature, as a result of radiotherapy. The dataset was obtained at the Vienna General Hospital by recruiting 20 patients with BCC, SCC, or actinic keratosis. It’s comprised of 146 acquisitions, with 18 of them being healthy volumes, used as a baseline, and the remaining belonging each to a different timepoint: before, right after, 3 months after, and 6 months after radiotherapy. Each time, 4 OCT volumes are acquired and then processed to obtain an OCTA volume through an intensity-based method. These arrays are semi-automatically segmented and further masked to ignore, during feature extraction, any areas that might lack angiographic information. A 2D projection is also computed along the depth axis of the 3D volume. Skeletonizations of both 2D and 3D arrays are computed. The features are chosen to describe the peculiarities of NMSC vascularization. Therefore, some parameters, such as branch tortuosity and entropy, describe how chaotically the vessels are intertwined with each other. Others, like vascular density, number of trees, and mean radius, observe how packed the blood vessels are and if their dimension changes between different timepoints. The variation of certain parameters with depth is calculated as well. These features allow to further compare lesions to healthy skin and see, for example, whether vascularization is the densest closer or further from the surface. The statistical analysis performed on the extracted data shows that, overall, 37 features change significantly across timepoints (p<0.05), with very strong evidence (p<0.0001) for most of them. They describe the vascular architecture as denser, more chaotic and branched, and closer to the surface in lesions than in healthy skin. Moreover, the post-hoc analysis indicates that pairs of timepoints often statistically differ from each other. The foundations for the reproducibility of these results are laid by developing two Python pipelines: one semi-automatic class for the initial computing of the OCTA volume, and an automatic pipeline for feature extraction. This thesis shows that quantitative parameters extracted from OCTA data allow to distinguish different timepoints in NMSC lesions treated with radiotherapy, with a consistent trend of features converging back to healthy values. Notably, these features are easily interpretable and give direct insight into the evolution of the lesions. Future studies that combine these vascular features with radiomics might be able to provide methods that accurately and non-invasively predict NMSC lesions’ response to treatment, thus opening the door to personalized radiotherapy for each patient.