Avoiding adverse effects of staining reagents on cellular viability and cell signaling, label-free cell imaging and analysis is essential to personalized genomics, drug development, and cancer diagnostics. By analyzing the images of cells, imagebased cell analytic methodologies offer a relatively simple and economical way to understand the cell heterogeneities and developments. Owing to the developments in high-resolution image sensors and high-performance computation processors, the emerging lens-less digital holography techniques enable a simple and cost-effective approach to obtain label-free cell images with large field of view and microscopic spatial resolution. In this work, the lens-less digital holography technique is adopted for image-based cell analysis. The holograms of three kinds of cells which are MDA-MB231, EC-109 and MCF-10A respectively were recorded by a lens-less digital holography system composed of a laser diode, a sample holder, a sensor and a laptop computer. The acquired holograms are first high-pass filtered. Then the amplitude images were reconstructed using the angular spectrum method and the sample to sensor distance was determined using the autofocusing criteria based on the sparsity of image edges and corner points. The convolutional neural network (CNN) was used to classify the cells. The experiments show that an accuracy of 97.2% can be achieve for two type cell classification and 91.2% for three type cell classification. It is believed that the lens-less holography combining with machine learning holds great promise in the application of stainless cell imaging and classification.
Raman tomography can provide quantitative distribution of chemicals in a three-dimensional volume with a non-invasive and label-free manner. In view of the problems of existing data collection strategy, a frequency modulation and spatial encoding based Raman tomography was proposed, which aims to improve the data collection scheme and reduce the data collection time. In this scheme, the laser beam was divided into several sub-beams to use as multipoint excitation light sources. These sub-beams were first modulated with different frequencies and then incident on the different points of sample surface simultaneously. Because the excited Raman signals would carry such modulation information, the Raman signals from which excitation position can be distinguished with the demodulation process. In detection end, the Raman scattering light first passed through a spatial-encoding mask and then was directed to the single photomultiplier tube. By changing the pattern of the mask and then performing recovery with sparse reconstruction, the distribution of the Raman signals on the sample surface can be obtained based on compressive sensing theory. Preliminary results showed that our scheme can recover the Raman images to the certain extent with a better signal-to-noise ratio, demonstrating the proposed scheme is feasible.