From Pixels to Diagnosis: Applications of X-ray Virtual Histology in Clinical Pathology

Histopathology is the medical discipline that studies the signs and characteristics of human disease processes in biological tissues. In the field of clinical diagnostics, the histopathological examination is a fundamental step in the process of defining the diagnosis, prognosis, and treatment plan. Despite the advancement of technology and the production of increasingly accurate and robust results, the two-dimensional information obtained through histopathology remains constrained by the sectioning plane. In contrast, X-ray virtual histology (XVH) represents an emerging imaging technique that produces high-resolution three-dimensional insights at the microscopic level, thereby overcoming the two-dimensional limitation of conventional histology. By leveraging the intrinsic characteristics of X-rays to capture comprehensive microarchitectural information, XVH eliminates the necessity for physical sectioning of the biological specimen, thereby preserving its structural integrity. This thesis aims to apply synchrotron-based XVH to two of the most aggressive and heterogeneous cancer types, which are melanoma skin cancer and non-small cell lung cancer (NSCLC). The need for superior image resolution and contrast provided by synchrotron light is essential for detecting subtle morphological variations within tumors. These capabilities are of particular significance for the investigation of infiltrative characteristics that drive tumor invasion, a process that may require further examination to gain a comprehensive understanding of its underlying mechanisms. In this work, the invasive phenotypes of melanoma, such as melanocytic nests and pagetoid spread, were translated from histopathology to XVH for the first time. Moreover, three-dimensional reconstruction of some adverse prognostic factors in NSCLC, such as spread through air spaces (STAS), pleural invasion, and lympho-vascular invasion, demonstrates the superiority of XVH over other experimental techniques in terms of image quality and diagnostic potential. To effectively manage and analyze the large datasets generated by XVH, an important contribution of this thesis lies in the integration of advanced deep learning algorithms for the automated segmentation and classification of tumor tissues. This approach enhances the identification of critical features such as cellular architecture, tumor boundaries, and vasculature, thereby streamlining the interpretation of complex histological data. In addition to synchrotron-based imaging, this thesis also provides a preliminary exploration of the potential of laboratory-based X-ray virtual histology systems. By qualitatively comparing the performance of two phase-contrast X-ray imaging laboratories in Trieste, Italy, this study anticipates their potential to broaden access to XVH technology in both clinical and research environments where synchrotron resources are not readily available. The overarching objective of this thesis is to advance the technical development of X-ray virtual histology while also enhancing its clinical relevance through close collaboration with medical professionals. This work aims to highlight the promising potential of X-ray virtual histology in cancer diagnostics and research by serving as a critical bridge between the technical advancements pursued by engineers and physicists and the clinical insights provided by medical doctors. Through this interdisciplinary approach, the thesis facilitates the translation of this innovative imaging technology into practical medical applications.