Differentiable Microscopy (∂μ)
Machine Learning Could Help Invent New Microscopes

Udith Haputhanthri1,2
Kithmini Herath1,2
Ramith Hettiarachchi1,2
Hasindu Kariyawasam1,2
Azeem Ahmad3
Balpreet S. Ahluwalia3
Chamira Edussooriya2
Dushan Wadduwage1,4,*
1 Center for Advanced Imaging, Faculty of Arts and Sciences, Harvard University, Cambridge, MA, USA
2 Department of Electronic and Telecommunication Engineering, University of Moratuwa, Sri Lanka
3 Department of Physics and Technology, UiT The Arctic University of Norway, Tromso, Norway
4 John Harvard Distinguished Science Fellows Program, Faculty of Arts and Sciences, Harvard University, Cambridge, MA, USA
* wadduwage@fas.harvard.edu

All-Optical Quantitative Phase Microscope

Timelapse of optimizing the Learnable Fourier Plane Filter

Differentiable Optical-Electronic QPI
End-to-end pipeline: Input field with high contrast phase information is fed into the proposed optical phase feature extraction network. The resultant compressed output intensity field which contains the phase features is captured by the detector. The electronic phase reconstruction network utilizes these features to reconstruct phase information


Ever since the first compound microscope by Zacharias Janssen in the late 16th century, scientists have been inventing new types of microscopes for various tasks. Inventing a novel architecture demands years, if not decades, worth of scientific experience and creativity. Here, we introduce Differentiable Microscopy (∂µ), a machine learning-based design paradigm, to aid scientists design new microscope architectures. first models a physics-based optical system with trainable optical elements. Using pre-acquired data, we then train the model end-to-end for a task of interest. First, we present two all-optical quantitative phase microscope (QPM) designs that require no computational post-reconstruction. Going a step further, we propose to acquire images in a compressed form such that more information can be transferred beyond the existing electronic hardware bottleneck. To this end, we present a learnable optical compression-decompression framework that learns content-specific features. The proposed system consists of learnable optical feature extractors as image compressors, where the intensity representation produced by these networks is then captured by the imaging sensor. Finally, a reconstruction network running on electronic hardware decompresses the QPM images. We believe that this setup opens up a new pathway for achieving end-to-end optimized (i.e., optics and electronic), compact QPM systems that provide unprecedented throughput improvements. The proposed differentiable microscopy framework supplements the creative process of designing new optical systems and would perhaps lead to unconventional but better optical designs.

Paper and Supplementary Material

Differentiable Microscopy Designs an All Optical Quantitative Phase Microscope

(hosted on ArXiv)

From Hours to Seconds: Towards 100x Faster Quantitative Phase Imaging via Differentiable Microscopy

(hosted on ArXiv)

Try our code


Results across tasks

All Optical Quantitative Phase Imaging

SSIM scores across multiple datasets

Optical Model MNIST HeLa [0,Pi] HeLa [0, 2Pi] Bacteria Colab Notebook
Generalized Phase Contrast 0.5134 0.5652 0.4056 0.6740 Open In Colab
Learnable Fourier Filter (LFF) 0.9184 0.7217 0.5921 0.9820 Open In Colab
PhaseD2NN 0.9146 0.6254 0.4854 0.9915 Open In Colab

Differentiable Optical-Electronic QPI