Adaptive Weighted Vector Means Optimization for Healthy and Malignant Skin Modeling at Microwave Frequencies Using Clinical Data

The dielectric properties of normal and cancerous skin vary with frequency due to changes in water content and tissue composition. Developing a reliable microwave system for skin cancer detection requires accurate characterization of that change in the dielectric properties. A possible choice is the Cole-Cole model, which can accurately fit the measured dielectric data for tissues. However, fitting the non-linear Cole-Cole model parameters with the measured data requires a sophisticated optimization algorithm. This study proposes an adaptive weighted vector means optimization algorithm, which employs adaptive initialization, logarithmic spaces, and enhanced local search mechanism, resulting in improved accuracy with fewer iterations. The algorithm is evaluated using dielectric data from healthy skin, basal cell carcinoma, squamous cell carcinoma, and melanoma and is found to outperform other relevant algorithms. One of the salient features of this study is that a set of clinical melanoma dielectric data is acquired, analyzed, and physically interpreted in terms of relaxation frequency and dispersion across 0.3 GHz to 14 GHz. It is found that melanoma closely follows the second-order Debye model, which is a special case for the second-order Cole-Cole model with a zero-valued dispersion broadening parameter. Although melanoma data is obtained from one lesion because of the low incidence rate, the research findings will contribute to a better understanding skin cancer at microwave frequencies. A triangular plot, which shows model fitness accuracy and the number of iterations, is presented to summarize the advantages of the algorithm.