[10p-N302-1~10] AI for Measurement Science: Toward New Measurement Paradigms Enabled by the Deep Integration of Information and Measurement Sciences

4:45 PM - 5:15 PM JST(7:45 AM - 8:15 AM UTC)

[10p-N302-8] Toward all-optical classification of biomedical cell images

〇Kotaro Hiramatsu¹ (1.Kyushu Univ.)

Keywords:

Bioimaging,Diffractive neural network

The growing demand for high-throughput, energy-efficient computing has spurred interest in optical computing, including diffractive neural networks (DNNs) that compute passively at the speed of light. However, most DNN demonstrations have used simple, standardized datasets such as MNIST, leaving their real-world utility largely unverified. Here, we present an in-silico demonstration of all-optical cell classification using a single-layer DNN based on a spatial light modulator (SLM). Trained via backpropagation, the network classifies lung cancer, breast cancer, and white blood cells from phase and amplitude images experimentally acquired by optofluidic time-stretch quantitative phase imaging. By measuring optical intensities at the detection plane, it achieves 96.1% accuracy, approaching that of conventional CNNs. Building on this work, we discuss the outlook toward experimental all-optical analysis of biomedical images.