Application of Artificial Neural Networks for Accurate Determination of the Complex Permittivity of Biological Tissue

Abstract

Medical devices making use of radio frequency (RF) and microwave (MW) fields have been studied as alternatives to existing diagnostic and therapeutic modalities since they offer several advantages. However, the lack of accurate knowledge of the complex permittivity of different biological tissues continues to hinder progress in of these technologies. The most convenient and popular measurement method used to determine the complex permittivity of biological tissues is the open-ended coaxial line, in combination with a vector network analyser (VNA) to measure the reflection coefficient (S11) which is then converted to the corresponding tissue permittivity using either full-wave analysis or through the use of equivalent circuit models. This paper proposes an innovative method of using artificial neural networks (ANN) to convert measured S11 to tissue permittivity, circumventing the requirement of extending the VNA measurement plane to the coaxial line open end. The conventional three-step calibration technique used with coaxial open-ended probes lacks repeatability, unless applied with extreme care by experienced persons, and is not adaptable to alternative sensor antenna configurations necessitated by many potential diagnostic and monitoring applications. The method being proposed does not require calibration at the tip of the probe, thus simplifying the measurement procedure while allowing arbitrary sensor design, and was experimentally validated using S11 measurements and the corresponding complex permittivity of 60 standard liquid and 42 porcine tissue samples. Following ANN training, validation and testing, we obtained a prediction accuracy of 5% for the complex permittivity.

Microwave Dielectric Property Retrieval From Open-Ended Coaxial Probe Response With Deep Learning

“…The obtained results are an indication that the proposed approach is robust and can be used to retrieve the dielectric properties of materials with the OECP technique. When this study is compared with the previously reported work in the literature, to the best of authors' knowledge, only one reported work carried out a study utilizing NN to retrieve complex permittivity with OECP response [21]. However, the reported study only used measurements; therefore, train and test data were restricted with the measurements and [24] 111.8 109.23 6.9 5.26 42.00 42.9 3.68±0.03 Methanol [26] 33.64 37.01 5.…”

“…In the literature, an artificial neural network model is proposed in [21] to compute the complex permittivity from the measured reflection coefficients for biomedical applications. In the proposed study [21], 102 experiment samples including standard liquids and biological tissues were measured in order to train, validate and test the neural network model. Prediction accuracy of ±5% was obtained for the complex permittivity.…”

“…Prediction accuracy of ±5% was obtained for the complex permittivity. The study in [21] was limited with collected data from the experiments; therefore, the predictions of the trained neural network model was restricted. Composing a large dataset through measurements that include different materials, frequencies, and probes is labor intensive and costly.…”

“…1) One main contribution of this work is generation of data-set for training the deep neural network, which estimates the Debye parameters. In particular, unlike to [21], the input and output dataset was not measured but synthetically generated by means of a calibration procedure. Designing such procedures is an important topic for inverse electromagnetic estimation problems [22], since it produces large amount of data required to train the deep learning model.…”

State estimation via prediction-based scheme for linear time-varying uncertain networks with communication transmission delays and stochastic coupling

“…In recent years, the analysis methods of complex dynamical networks have played an important role in the big data transmissions, information processing and power grid (Baran & Rzysko, 2020;Bonello et al, 2020;Chen et al, 2019;Ding et al, 2019;Hernandez-torres et al, 2020;Wang et al, 2019). Recently, the networked control technology brings many conveniences to industrial development and technological advancements (Chen et al, 2020;Hu et al, 2017;Zou, Wang, Dong et al, 2020;.…”