Tunable THz sensing for early detection of skin cancer by deep learning enabled image reconstruction

This research reports deep learning model-based image reconstruction of healthy cells and cancerous cells by deployment of metamaterial absorbers. Two different tunable absorbers have been proposed. In absorber I, tunability is introduced by varying the chemical potential of graphene strips which act as switches while in absorber II, tunability is facilitated by using multiple graphene patches embedded in slotted silver patches. Equivalent circuit models (ECM) have been proposed for modelling the electromagnetic coupling between different constituents in the absorbers by lumped parameters for analysing the reflection characteristics. This study is vital for comprehending the effect of the absorber geometry in determining the resonant frequencies corresponding to peak absorption and reflection nulls. The surface current distribution aids in determining whether electric or magnetic resonances are formed in the absorber. The tunable absorbers achieved a maximum sensitivity of 435 GHz/RIU. Maximum quality factor of 319 and figure of merit (FOM) of 11 have been obtained. The proposed absorbers can be used in detecting cancerous cells of human skin when the latter is placed as an analyte over it. Distinct 2D images of healthy and cancerous cells have been reconstructed from the reflection characteristics of the absorber when placed in vicinity of human skin which ensures that it can be used as a biosensor for non-invasive detection of skin cancer at an early stage. A meticulous analysis of convolutional neural network (CNN) enabled imaging algorithm from the reflectance spectrum has been elucidated. The model achieved 94.3% accuracy, 92.7% sensitivity, 95.8% specificity, and an F1 score of 93.2%.