Logo: to the web site of Uppsala University

Microwave sensors can be used within medicine as they use non-ionizing radiation, are often low cost, and can be designed for a specific purpose. The application of microwave sensors for diagnostics and monitoring can be improved using appropriate data analysis. The multi-layered structure of the human body makes the measurements on people complex. A tremendous effort is required to create an analytical model of the body. In this context a data-driven approach, building a model that learns from previous measurements, is more suitable to analyze the data. This thesis aims to address statistical and data-driven approaches based on microwave sensor data for biomedical applications.

A significant part of this thesis deals with microwave sensors for assessing muscle quality. It details the progress from initial clinical campaign to the creation of a machine learning algorithm to assess the local body composition. Such a device would be suitable for screening age-related muscle disorders like sarcopenia and muscle atrophy. Statistical analysis following the initial clinical campaign revealed no significant differences in the microwave data. Therefore, new sensor designs were evaluated. The most promising sensor was used in a small clinical campaign where it was able to detect a change in muscle size for one patient with multiple measurements over time. Successive measurements followed on tissue emulating phantoms and volunteers. For data analysis a machine learning algorithm was designed to predict the skin, fat, and muscle properties. This changes the aim from assessing muscle quality to assessing local body composition. For phantom data the algorithm was accurate for skin and fat and for volunteer data for fat and muscle. Crucially, the algorithm also performed better with more data available, meaning that results should improve if more data is collected.

Microwave sensors have also been employed to assess bone. The first of two applications was to monitor the bone healing progression post surgery treating craniosynostosis. No substantial conclusions could be drawn from the statistical analysis most likely due to measurement uncertainties. The second application used a purpose-built setup for controlled measurements in ex vivo bone samples submerged in liquid, to simulate an in vivo environment. The purpose was to estimate the dielectric properties of bone. The derived bone properties were lower than expected, probably due to air trapped inside the sample.

The Muscle Analyzer System (MAS) project wants to create a standalone microwave device that can assess the muscle quality, called the MAS device. To achieve that an algorithm that can derive the properties of skin, fat and muscle from the measurements is needed. This paper presents a machine learning algorithm that aims to do precisely that. The algorithm relies on first predicting the skin using the data from the MAS device, then predicting the fat again using the data from the MAS but also the predicted skin value and lastly the muscle is predicted using the microwave data together with the skin and fat predictions. Data have been collected in phantom experiments, materials that mimick the dielectric properties of human tissues. The algorithm is trained to predict the properties of said phantoms. The results show that the prediction for skin thickness works well, the fat thickness prediction is okay but the muscle prediction struggles. This is partly due to the error from the skin and fat layers are propagated to the muscle layer and partly because the muscle layer is farthest away from the sensor, which makes getting information from that layer harder.

In this paper a standalone microwave device is evaluated for its ability to assess local body composition with the ultimate goal to assess muscle quality. Data have been collected from volunteers who were measured on their thigh using the microwave device and ultrasound. A machine learning algorithm with three stages is designed that utilizes the stacked nature of the tissues in the thigh to predict skin and fat thickness and the cross-sectional area of the rectus femoris muscle. The input to the algorithm is the signal response from the microwave sensor and also the prediction from the previous layers. The ultrasound measurements are used as the ground truth labels for each tissue to train the machine learning models. The measurements were performed with two sensors, where usage of the combined data from both sensors produced the best results for fat and muscle, 0.57 and 0.63 in R 2 score, respectively. In the drop analysis, a step where a select proportion of the data is temporarily removed, the identified models showed increased scores with a larger amount of data available indicating the learning of the models improves with more data. Although the results are encouraging more data is ultimately needed to further study the algorithm.

We have developed a new transmission-based, open-ended coaxial probe for assessing vertebrae strength during spinal fusion surgery. The approach exploits the fact that the probes are within the far field of each other implying that the phase varies linearly with respect to propagation distance. Determining the absolute phase is critical for recovering the associated tissue dielectric properties from which bone strength will be determined. Unfortunately, unwanted multi-path signals corrupt the signals at the lower end of the operating frequency range from which our conventional unwrapping strategy depends. Our new approach requires only three measurements within the prime frequency range and can be determined robustly with a minimum of computations. This will be vital to developing a commercial device since the signal levels will be extremely low power requiring longer than usual data acquisition times, which will be mitigated by measuring the data at only a few frequencies. Fast and efficient operation will be critical for clinical success.

We are developing a new transmission-based probe for assessing vertebrae strength via the differences between the dielectric properties of normal and osteoporotic bone. The probes are simple open-ended coaxial cables that will fit easily into the pedicle canals drilled by the surgeons before implanting screws to secure the supporting instrumentation. If the screws pull because of bone weakness, it can inflict a wide range of complications. In the progression towards actual clinical trials, we have already validated the approach on different liquids with varying dielectric properties. In this study, we assess actual animal bone samples as a precursor to human investigations. We exploit the fact that the trabecular bone is considerably different on opposite sides of the growth plate of long bones such as the femur - somewhat mimicking differences between normal and osteoporotic bone. For these experiments, we show that the properties of the distal and proximal sections are considerably different and present opportunities to exploit them in a diagnostic setting

In this thesis the primary focus is on the evaluation of biomedical microwave sensor to be used in the muscle analyzer system. Lower muscle quality is one indicator that a patient can have sarcopenia. Therefore the muscle analyzer system can be a tool used in screening for sarcopenia. Sarcopenia is a progressive skeletal muscle disorder that typically affects elderly people. It is characterized by several different things, one of them is that there is an infiltration of fat into the muscle. At microwave frequencies the dielectric properties of fat are vastly different than the muscles. So, this fat infiltration creates a dielectric contrast compared to muscle without this fat infiltration that the sensors aim to detect.

The muscle analyzer system is proposed to be a portable device that can be employed in clinics to assess muscle quality. The sensors are evaluated on their ability to distinguish between normal muscle tissue and muscle of lower quality. This is achieved via electromagnetic simulations, clinical trials, where the system is compared against established techniques, and phantom experiments, where artificial tissue emulating materials is used in a laboratory setting to mimick the properties of human tissues.

In a initial clinical pilot study the split ring resonator sensor was used, but the results raised concerns over the penetration depth of the sensor. Therefore, three new alternative sensors were designed and evaluated via simulations. Two of the new sensors showed encouraging results, one of which has been fabricated. This sensor was used in a another clinical study.This study only had data from 4 patients, 8 measurements in total, meaning it was hard to draw any conclusions from it. The sensors used in the clinical setting as well as another were evaluated in the phantom experiments. Those experiments were exploratory because a wider frequency range was used, although some problems in the experiments were found.

A secondary approach in this thesis is devoted to a data-driven approach, where a microwave sensor is simulated. The data from it is simulated and used to train a neural network to predict the dielectric properties of materials. The network predicts these properties with relatively high accuracy. However, this approach is currently limited to simulations only. Several ideas on how to improve this approach and extend it to measurements is given.

A microwave sensor to safely measure quality of muscle tissue for diagnosis and screening of diseases and medical conditions characterized by fat infiltration in muscle is presented. Fat infiltration in muscle may be seen by a lower dielectric constant of muscle at microwave frequencies corresponding to the large contrast between fat and muscle tissues. A planar resonator based on a bandstop filter and optimized to noninvasively interrogate muscle in the thigh on tissue quality is proposed. Currently, a study based on clinical trials is carried out, and, here, we present a preliminary correlation between skin and fat thicknesses and rectus femoris cross sectional area (CSA) measured with ultrasound and the proposed sensor's resonance frequency. CST simulations based on the ultrasound information guide the analysis. We see that although there are signs of a potential correlation between CSA and resonance, skin and fat variability is still an issue to overcome.

Microwave-based sensing for tissue analysis is recently gaining interest due to advantages such as non-ionizing radiation and non-invasiveness. We have developed a set of transmission sensors for microwave-based real-time sensing to quantify muscle mass and quality. In connection, we verified the sensors by 3D simulations, tested them in a laboratory on a homogeneous three-layer tissue model, and collected pilot clinical data in 20 patients and 25 healthy volunteers. This report focuses on initial sensor designs for the Muscle Analyzer System (MAS), their simulation, laboratory trials and clinical trials followed by developing three new sensors and their performance comparison. In the clinical studies, correlation studies were done to compare MAS performance with other clinical standards, specifically the skeletal muscle index, for muscle mass quantification. The results showed limited signal penetration depth for the Split Ring Resonator (SRR) sensor. New sensors were designed incorporating Substrate Integrated Waveguides (SIW) and a bandstop filter to overcome this problem. The sensors were validated through 3D simulations in which they showed increased penetration depth through tissue when compared to the SRR. The second-generation sensors offer higher penetration depth which will improve clinical data collection and validation. The bandstop filter is fabricated and studied in a group of volunteers, showing more reliable data that warrants further continuation of this development.

This paper presents an extension to previous work, using neural networks to characterize materials in microwave frequencies, to extend the applicability of a deep learning model to be able to characterize the dielectric properties of biological tissues. A neural network model using convolutional and fully connected layers is designed to predict the permittivity and loss tangent using the scattering parameters from a coplanar waveguide transmission sensor. Simulated data from the sensor provide a large dataset, with a wide range of values for the permittivity and loss tangent, which is used to train and test the model. The trained network is validated by predicting the output parameters on the test set. Compared with previous work, by using convolutional layers the applicable parameter space is vastly extended while keeping satisfying levels of accuracy. A complete system with a trained network is proposed to be used in a lab or in clinics.

Microwave sensing technique is a possible and attractive alternative modality to standard Xrays,magnetic resonance imaging, and computed tomography methods for medical diagnostic applications.This technique is beneficial since it uses non-ionizing radiation and that can be potentially used for themicrowave healthcare system. The main purpose of this paper is to present a microwave sensing techniqueto analyze the variations in biological tissue thickness, considering the effect of physiological and biologicalproperties on microwave signals. With this goal, we have developed a two-port non-invasive sensor systemcomposed of two split ring resonators (SRRs) operating at an Industrial, Scientific, and Medical frequencyband of 2.45 GHz. The system is verified using the amplitude and phase of the transmitted signal in ex-vivomodels, representing different tissue thicknesses. Clinical applications such as the diagnosis of muscularatrophy can be benefitted from this study.