Preliminary investigation on Microwave Sensing for early detection of Lymphoedema

In this work we investigate the potential of microwave sensors for the early detection of Lymph Oedema (LO) which could contribute in avoiding complications and reducing costs for the healthcare system. LO is a chronic disease characterized by the over-accumulation of lymphatic fluids in the body, which generally causes the swelling located in one arm or leg, sometimes both arms and/or legs. Although many medical solutions have been explored targeting the possibility of preventing and detecting the illness, especially considering high-risk factors and implying additional expansive tests, like Magnetic Resonance Imaging (MRI) or Computerized Tomography (CT) Scan, the idea of using microwave sensors can represent a valid, efficient and cost-effective alternative. Microwave sensor measurements paired with Machine Learning techniques may provide a feasible solution in the understanding of disease penetration in the body and, possibly, offer a more general comprehension on the structure laying underneath the skin. This thesis analyzes readings performed with a Split Ring Resonator (SRR) on patients undergone surgery, trying to apply clustering and classification methods to discover if any correlation between measurements and patient’s metadata is present. In parallel, to interpret this processed information, simulations using CST Studio Suite software are performed emulating the three-layered structure of the human body, composed of skin, fat and muscle. In the simulations same sensor used for the real measurements is present and different scenarios are tested, including changing electro-magnetic parameters of the body materials and addition of Lymph layer in the 3D model. In the end, a deep description on the methodology to be adopted for future data collection will be proposed. This is done to avoid same problems occurred during previous acquisition and to allow better results in the machine learning phase. As a matter of facts, through some experiments performed for this thesis work, some erroneous behaviors have been discovered. Unfortunately, both for this reason and for lack of enough data, machine learning results do not seem to provide any robust result, but in future, possibly with an additional collection of patients’ meta-data, this could be feasible.