Titolo della tesi: Image Analysis of Terahertz Scattering Microscopy for Medical Diagnostics
This Thesis work reports on the characterization of a novel microscope developed by our research group for early diagnosis in-vivo of skin diseases resulting in anomalous buried structures in the skin, which may not be visible at naked eye or with the state-of-the-art diagnostics technologies based on visible, infrared or ultrasound waves. In particular, we have focused on the Basal Cell Carcinoma (BCC), one of the most common skin diseases in Europe and in the United States. The illumination source of our microscope is in the TeraHertz range (10^12 Hz, or THz), a region of the electromagnetic spectrum that has not been widely used in real-life applications because of a lack in practical sources and detectors, an occurrence that is often named as the TeraHertz Gap. In the last 20 years, however, room-temperature sources and detectors for the Terahertz range based on affordable microelectronic technologies have been developed, and in the last 10 years have become available to research groups and optical instrumentation companies for developing practical applications. THz radiation is a good candidate for skin anomaly detection for medical diagnostics for the following three reasons: 1) it can penetrate the skin for many hundreds of microns, much more than near-infrared (few hundreds of microns with strong scattering) or visible light (tens of microns); 2) it is very sensitive to the water content, or skin hydration, unlike other penetrating radiation such as X-rays; 3) as opposed to high-energy radiation or particles, it is not-ionizing (photon energy of a few meV compared to the dozens of keV of the X-rays or to almost MeV-range of gamma rays resulting from positron beams).
All the instrumentation we used has been acquired in the framework of the project Micoted: Microscopio Terahertz per la Dermatologia funded by the Regional research fund for small and medium enterprises, co-financed by Regione Lazio and the European Union. This Thesis work and the Micoted project have run in parallel since the beginning. Many research groups around the world are trying to diagnose skin diseases using the THz radiation, but we believe that we have proposed an approach that is different from all previous approaches. The main differences are in the type of THz source and detector used, and in the physical observables that are measured: while most other groups use a wideband source and a single-pixel detector in the so-called THz Time-Domain Spectroscopy (THz-TDS) technique, we use a continuous-wave monochromatic source at 0.6 THz but we can exploit a THz camera for the detection of the signal reflected by the skin, in the form of the real-space image of the pointspread function of the microscope. The point-spread function of our quasi-ideal confocal microscope is an Airy disk, which is then modified by the presence of the sample. Our idea is precisely to study the modifications of the Airy disk pattern produced by the presence of the sample.
As a consequence of our novel approach, we did not have much bibliography to look at for describing our new approach. That is why we had to start from skin simulants, in order to obtained controlled and reproducible experiments to develop the technique itself and to study the features that characterize our distorted Airy disk images. During the data analysis we have found out that, in the absence of a strong theory that could support data interpretation, a Machine Learning approach could have helped us to find which features were the most significant. In the last part of the experimental work, therefore, we have developed a Machine Learning approach that is thoroughly presented here. The results of this Thesis work form the solid basis on which a future project on human skin samples and, possibly, even on in-vivo patients, could develop further our microscope.