Quantitative Detection Of Immunochromatographic Strips Based On Deep Learning And PSO Algorithms

Lateral flow immunoassay(LFIA)is a new immunoassay technology based on the combination of specific antigen-antibody reaction and chromatography method.Especially,gold immunochromatographic strip(GICS)has become a popular membrane-based diagnostic tool in a variety of settings due to its sensitivity,simplicity and rapidness.Currently,most GICSs are limited to qualitative or semi-quantitative tests due to the limitation of technique and the existence of interference.On the one hand,the results obtained directly by naked eyes will increase relatively labor-intensive and also contain some subjective deviations.On the other hand,the results provide too little information,which cannot satisfy the clinical needs.Therefore,this paper aims to optimize the dynamical model and develop a framework of automatic image recognition to further improve the sensitivity as well as the quantitative performance of the GICS systems.Firstly,this paper develops a dynamical model containing the time-delay parameter for LFIA according to the biochemical reactions.Then,a novel switching delayed particle swarm optimization(SDPSO)algorithm is proposed for the parameter estimation problem of the LFIA.In the proposed SDPSO algorithm,the velocity of the particle jumps from one mode to another based on an evolutionary factor and Markov chain.The experimental results show that the SDPSO can not only enrich the interaction information but also avoid the swarm from premature.The new SDPSO is also successfully exploited to identify the unknown time-delay parameters of in the nonlinear state-space model of LFIA.Secondly,the deep belief network(DBN)is applied for the first time,to quantitative analysis of GICS images with hope to segment the test and control lines with a high accuracy.Three features(including intensity,distance and difference)are selected as the inputs for the DBN method in order to successfully distinguish the test and control lines from the region of interest(ROI)that is obtained by pre-processing the GICS images.Experiments results show the feasibility and effectiveness of the DBN for quantitative analysis of GIGS.Finally,a new dynamic state-space model is established in this paper for transforming the segmentation of test and control lines into the problem of state estimation.Then,an innovative particle filter(PF)with a hybrid proposal distribution,named deep belief network-based particle filter(DBN-PF)is proposed to further improve the performance of DBN.In particular,DBN provides an initial recognition result in the hybrid proposal distribution,and the PSO algorithm moves particles to regions of high likelihood.The performance of proposed DBN-PF method is comprehensively evaluated on not only an artificial dataset but also the GICS images.Experiment results demonstrate that the proposed approach is effective,and also can be utilized as a novel method for quantitative analysis of GICS.