[16p-A37-1~11] CS.2 Code-sharing Session of 3.2 & 4.4

2:15 PM - 2:30 PM JST(5:15 AM - 5:30 AM UTC)Presentation by Applicant for JSAP Young Scientists Presentation AwardEnglish Presentation

[16p-A37-3] Quantitative Zernike Phase-Contrast Microscopy with an Untrained Neural Network

〇(D)Zinan Zhou¹, Keiichiro Toda¹, Rikimaru Kurata², Kohki Horie¹, Ryoichi Horisaki², Takuro Ideguchi¹ (1.UTokyo (Science), 2.UTokyo (IST))

Keywords:

quantitative phase imaging,phase-contrast microscopy,deep learning

Our work discusses advancements in Zernike phase-contrast microscopy (PCM), which transforms phase shifts in a sample into intensity contrast but lacks quantitative data about the specimen. Previous efforts to quantify PCM involved complex algorithms with heavy reliance on regularization and optimization, limiting practical use. The study introduces a revised PCM phase retrieval algorithm using an untrained neural network (uNN) inspired by deep image prior (DIP) methodology. This new approach removes the need for ad hoc regularization by employing uNN as a structural prior, allowing effective quantitative phase retrieval from a single PCM image. The performance of this method was compared to ground truth and previous methods, showing promising results with a root-mean-square error of 0.17 radians and a structural similarity index of 0.799. The study explored various neural network architectures, including encoder-decoder and U-Net, demonstrating the potential for practical applications in biological and clinical settings.