Cervical Cancer Cell Real-time Screening System Based On Multi-spectral Imaging

So far,the commonly used screening methods for cervical cancer cells include TBS(The Bethesda System)classification and cellular DNA quantitative analysis.The use of multiple staining methods(ie,simultaneous Pap staining of the cytoplasm on the same cell smear and Feulgen staining of the nucleus)for cervical cancer cell screening remains a blank.The difficulty with this multiple staining screening method is that the absorbance of non-DNA material interferes with the absorbance of the DNA material.First,Aiming at the difficulty of absorbance aliasing in the case of composite dyeing,this paper proposes a multi-spectral image stripping model established by multiple linear regression method.The model can be used to strip the spectral response of the aliasing into the spectral response of each single band.The model extracts the absorbance of the DNA material for DNA quantitative analysis,and the remaining bands are pseudo-color synthesis for TBS screening,achieving a perfect combination of two common methods.Second,Multi-spectral imaging using self-developed high-speed high-resolution CMOS camera,multi-color LED for time-sharing imaging of cell slide absorption characteristics,self-made FPGA acquisition card for data acquisition for high-speed data transmission,while achieving continuous focusing and more Layer focusing function.It has the advantages of high spectral resolution,fast spectral response,no moving parts,and low cost.Third,using the flexibility of FPGA,multi-channel parallel pipeline operation of multi-spectral image stripping model is realized by hardware algorithm to realize real-time stripping operation and output of multi-spectral image.A series of experiments have proved that the data transmission of the self-made high-speed camera is stable.The absorbance of the DNA material stripped by the model is not significantly different from the absorbance of the measured DNA material.The scanning platform composed of the high-speed camera and the PCIe(Peripheral Component Interconnect express)acquisition card can realize 100,000 stable collections of cell images in 3 minutes.The depth of the deep learning cell classification model of the analysis platform can reach 99%.In addition,the multi-spectral image stripping model of the system is only related to the wavelength of the absorption spectrum of the absorbing material(ie,incident light),and can be used to establish new models for other similar medical screening,agricultural product testing and other applications,and has great market application potential.