Reducing the fluorescence background in microfluidic assays is important in obtaining accurate outcomes and enhancing the quality of detections. This study demonstrates an integrated process including cell labelling, fluorescence background reduction, and biomolecule detection using liquid-liquid extraction in a microfluidic droplet system. The cellular lipids in Chlorella vulgaris and NIH/3T3 cells were labelled with a hydrophobic dye, Nile red, to investigate the performance of the proposed method. The fluorescence background of the lipid detection can be reduced by 85% and the removal efficiency increased with the volume of continuous phase surrounding a droplet. The removal rate of the fluorescence background increased as the surface area to volume ratio of a droplet increased. Before Nile red was removed from the droplet, the signal to noise ratio was as low as 1.30 and it was difficult to distinguish cells from the background. Removing Nile red increased the signal to noise ratio to 22 and 34 for Chlorella vulgaris and NIH/3T3, respectively, and these were 17 fold and 10 fold of the values before extraction. The proposed method successfully demonstrates the enhancement of fluorescence detection of cellular lipids and has great potential in improving other fluorescence-based detections in microfluidic systems.