以擠壓流場和基於視覺的物體追蹤實現可攜式光學流道系統用於微粒分離與定量

Abstract

在臨床上，從複雜檢體中分離白血球細胞、循環腫瘤細胞與聚合物等生物分子，具有重大意義。例如，白血球細胞的計數為數百種疾病的診斷、篩查及治療手段的評估提供參考。微流體系統可以藉由微米尺度的流體通道處理微量液體。目前用於微粒分離的微流體晶片，已經可以操控、分離微粒，但大多依賴外部的電場或磁場、多孔濾膜，後者還會有產生孔隙阻塞、吸附效應等問題。此外，微流體晶片的運轉需要高成本的送流系統和搭配顯微鏡使用，限制了微流體晶片本身的應用。為了發展精確、低成本的微流體晶片微粒分離技術及應用，本研究提出基於擠壓流體分離技術的可攜帶式光學微流體系統，採用較低成本的微量注射幫 浦、智慧型手機照相機，實現微粒分離與計數。本研究藉由在兩幅連續影像中識別同一微粒的方法，實現基於影像的粒子追蹤。本研究還採用基於編解碼器的卷積神經網路在粒子追蹤之前先對原始影像進行去雜訊處理。目前，智慧型手機已經非常普及，因此，低成本微流體晶片的開發對未來定點照護檢驗等情境的應用意義非凡。

Separation of biomolecules like WBCs (white blood cells), CTCs (circulating tumor cells), polymers from complex samples have extensive clinical significance. For example, WBC count provides implications for the diagnosis and screening of hundreds of diseases and treatments. Microfluidics is the study of systems that can process small quantities of fluids by using tiny channels having dimensions at the microscale. Several microfluidic chip based particle sorting solutions have been provided which manipulate the particle movement inside micro channels to separate them, however many of these techniques require external electrical or magnetic fields, porous membrane filters which raise clogging and fouling effect. If not the above problems, almost every microfluidic device needs bulky expensive pumping system and lab microscopes which limit the use of these valuable microfluidic design solutions in the lab itself. Now the question is, can we find a cost effective and accurate alternative to lab grade microscope and syringe pump to combine with a simple microfluidic design to do cell sorting? A portable opto-fluidic system for particle separation and quantification is proposed which uses a novel microfluidic design called “pinched flow fractionation” along with a smart phone camera and low cost syringe pump to address this issue. It uses vision based particle tracking by defining an identity mapping between corresponding particles of two consecutive frames. An encoder-decoder based convolutional neural network is used to do pixel wise semantic segmentation which generates completely noise free output images essentially required for above identity mapping and further image processing pipeline. Since smart phones are ubiquitous now, this solution provides a possibility for an automated point of care disease diagnosis tool.