Detection of carcinoembryonic antigen specificity using microwave biosensor with machine learning

Early diagnosis and screening of tumor markers are essential for effective cancer treatment and improve the treatment efficiency and prognosis of tumor recurrence and metastasis. In this study, a split-ring resonator (SRR) circuit based on an interdigital electrode structure was developed and applied to microwave biosensors along with machine learning to detect extremely low concentrations of Carcinoembryonic Antigen (CEA). CEA was detected using a microwave sensor operating at a resonance frequency of 4.33 GHz. When the microwave sensor is exposed to CEA analytes, it generates a new frequency in the range of 1-2 GHz. The position and intensity of the newly generated frequency can be used to characterize and predict the concentration of the CEA solution. The proposed sensor exhibits excellent resonance linearity for various concentrations of CEA (R² = 0.999), as well as a very low detection limit (39 pg/mL) and high sensitivity (27.5 MHz/(ng/mL)). A machine learning approach was implemented to predict the CEA concentration in blood samples. The results showed close concurrence with the CEA concentration detected by the sensor. Western Blot (WB) was used to compare the CEA contents of four different cell types, and a biosensor was used for validation; the results of the two experiments showed good agreement. This is the first demonstration of the validation of biosensor reliability at the cellular level. The proposed concept exhibits outstanding detection performance with convenient and rapid tumor marker detection. Hence, it has important implications as an auxiliary diagnosis method for cancer.