Isotropic quantitative differential phase contrast (iDPC) microscopy based on pupil engineering has made significant
improvement in reconstructing phase image of weak phase objects. In previous researches, the pupil designs have been
investigated for enhancing the data acquisition efficiency. To further improve the phase retrieval procedure in iDPC, we
adapt deep neural networks to achieve isotropic phase distribution from half-pupil based quantitative differential phase
contrast (qDPC) microscopy. In this study, we utilized U-net model for mapping from 1-axis phase reconstruction to 12-
axis one. The results show that the deep neural network we proposed achieved expecting performance. The final testing
loss value of our model after 1000 epochs of training achieved 6.7e-5 after normalized. The peak signal to noise ratio
improvement is from 26dB to 30dB.
Quantitative differential phase contrast (qDPC) imaging is a specific technique for observing the transparent object. qDPC method adopts the structured-light illumination to provide the quantitative phase reconstruction, and it has lesser hardware requirement compared with other quantitative phase imaging (QPI) method. Conventionally, to achieve isotropic phase retrieval with better uniformity by utilizing qDPC system, it needs multiple measurements with different asymmetric illumination pattern along the transverse direction. However, it takes much more time to reconstruct the phase distribution while increasing the number of measurements. Therefore, here we applied the deep neural network (DNN) model for approaching isotropic phase retrieval and minimizing the acquisition time simultaneously. To achieve the isotropic phase distribution with less measurements, the U-net architecture was adopted in this study. The U-net model was utilized for converting the result from 1-axis qDPC method (the phase retrieval which has lesser measurements) to the result from 12-axis qDPC method (the phase retrieval which has more measurements to cover all the spatial frequency information in the spatial-frequency domain). For the model training stage, we prepared 5 different types of cells to provide sufficient training datasets. To evaluate the performance of our trained model, we prepared another 2 distinct types of cells referred as testing dataset. The results showed our model can recover the insufficient phase value in the sample. The morphology of cells can be analysis after applying our proposed DNN model.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.