Seeing beyond the surface: Quantifying Actinic Keratosis Progression with OCT and OCTA

Actinic keratosis, a precancerous skin lesion, can transform into squamous cell carcinoma if left untreated, passing through critical stages such as Bowen's disease. Diagnosis of actinic keratosis can be made clinically, by visually analyzing the shape and size of the lesion, or dermoscopically, by investigating the more superficial layer of the skin. However, since these are qualitative analyses, limited by the inability to explore the deeper layers of the skin, they cannot distinguish an actinic keratosis from a tumor. This is why biopsy, an invasive procedure that ensures certain diagnosis of cancerous and non-cancerous condition, is often used. What if it were possible to diagnose actinic keratosis and follow its evolution without invasive intervention? This thesis project proposes an alternative: it explores the use of OCT (optical coherence tomography) and OCTA (OCT angiography), non-invasive imaging techniques that, by exploiting light and the principle of interference, can capture high-resolution images and volumes of biological tissues structure and vasculature. The aim is to demonstrate how these techniques, already used in medical settings, can support the qualitative and quantitative distinction between actinic keratosis and its possible developmental stages. Using a dataset of 13 patients recruited at the Vienna General Hospital, 48 OCT volumes were collected. From these, OCTA volumes were extracted and processed, including artifact reduction and adjustments to contrast and brightness to improve data visualization. Volume segmentation and subsequent skeletonization were then performed. The project runs through two pathways to achieve the presented goal: OCTA processing to obtain meaningful vascular parameters and OCT volume analysis to identify lesion depth. Considering that as the tumor stage progresses, the vascular network tends to be increasingly disorganized and to extend to the deep layers of the skin, parameters such as tortuosity, vascular density, entropy, and fractal dimension were chosen to distinguish different vascular pattern, and were calculated separately for the superficial and deep layers, with the aim of demonstrating differences between the four patient groups: healthy, actinic keratosis, Bowen's disease, and squamous cell tumor. To validate these differences, a statistical analysis, both on individual parameters and multiparametric, was carried out, and a logistic regression-based classifier was then introduced to predict the stage of pathology in output. Parametric analysis effectively distinguished healthy from diseased groups, revealing statistically significant differences in parameters such as vascular density and entropy. However, in the differentiation between actinic keratosis and tumor, the results were partially limited by some critical issues, such as the unbalanced dataset and the presence of artifacts and noise. On the other hand, OCT volume analysis made it possible to determine the depth of lesions, showing that tumor lesions tend to involve deeper layers than non-tumor lesions, facilitating the distinction between actinic keratosis and later stages. This thesis guides through each step from image acquisition to morphologic analysis of OCTA and OCT volumes, offering a noninvasive method to support the diagnosis and distinction between actinic keratoses and cancerous lesions.