All-optical classification of real biomedical cell images using a diffractive neural network: a simulation study

TL;DR

This paper demonstrates a simulated all-optical cell classification system using a diffractive neural network, achieving high accuracy with potential for ultrafast biomedical imaging applications.

Contribution

It introduces a novel in-silico design of an all-optical DNN for biomedical image classification, optimized for real-world cell images, with performance comparable to traditional CNNs.

Findings

Achieved 93.6% classification accuracy.

Utilized experimentally acquired phase and amplitude images.

Showcased potential for ultrafast, energy-efficient optical computing.

Abstract

We report an in-silico demonstration of an all-optical cell classification system using a single-layer diffractive neural network (DNN) optimized for real-world biomedical images. Implemented virtually with a spatial light modulator (SLM), the DNN was numerically trained via backpropagation to differentiate breast and lung cancer cells. The training utilized experimentally acquired phase and amplitude images from optofluidic time-stretch quantitative phase imaging. Classification was simulated by computing the optical intensities at the detection plane. The optimized DNN achieved 93.6% accuracy, approaching that of conventional convolutional neural networks. This study highlights the potential of SLM-based DNNs for ultrafast, energy-efficient biomedical image processing in practical optical computing scenarios.