Spatio-Temporal Processing for Multichannel Biosensors Using Support Vector Machines

Rapid-response biosensing systems are necessary to counteract threats due to foreign and high-consequence pathogens. A yes/no multichannel biosensor is an important tool that enables simultaneous detection of different pathogens, independent of their relative concentration level. This paper proposes a novel multichannel biosensing technique, which combines multiclass support vector machines (SVMs) with multichannel immunosensors. The method combines spatial and temporal information generated by the multichannel immunosensor for rapid and reliable discrimination between pathogens of interest. This paper demonstrates that by including temporal and cross-reactive spatial signatures, the accuracy of the system can be improved at low pathogen concentration levels and for discrimination between closely related strains of pathogens. Compensation of systematic and biosensor fabrication errors is achieved by the use of a supervised SVM training which is also used in system calibration. Experimental results, with a prototype multichannel biosensor used for discriminating strains of E. coli (K12 and O157 : H7) and Salmonella enterica serovar Thompson, show an accuracy of 98% for concentration levels, 100-108colony forming units per milliliter, and total detection time of less than 6 min