Themed collection AI in Microfluidics

Keisuke Goda, Hang Lu, Peng Fei, and Jochen Guck introduce the AI in Microfluidics themed collection, on revolutionizing microfluidics with artificial intelligence: a new dawn for lab-on-a-chip technologies.

This review outlines the current advances of high-throughput microfluidic systems accelerated by AI. Furthermore, the challenges and opportunities in this field are critically discussed as well.

This review summarizes the implementations of droplet microfluidics based on AI, including droplet generation, biological analysis, and material synthesis.

We discuss the recent trends in integrating deep-learning (DL) and optofluidic imaging. A holistic understanding of them could incentivize DL-powered optofluidic imaging for advancing a wide range of novel applications in science and biomedicine.

In this review article, we surveyed the applications of machine learning in microfluidic design and microfluidic control.

GPT-4 was utilized to generate 3D and 2D CAD designs for common microfluidic device components.

A microfluidic ratchet sorting device is used to separate RBCs based on deformability. Sorted cells are imaged using optical microscopy and are used to train and test a deep learning network to classify the cells based on deformability.

This work presents two new quality metrics for droplet generation, versatility and stability.

We experimentally justify the advantages of jumping on the deep learning trend for image-activated budding yeast sorting and validate its applicability towards morphology-based yeast mutant screening.

A real-time single-cell imaging and classification system can directly identify cell types from motion-blur images using a deep learning algorithm.

We developed a compact perfusion cell culture with integrated wireless detection device for real-time optical monitoring. The platform enables long-term cell growth and cytotoxicity assay where cell viability is quantified using AI software.

Introducing meta-optimizer as a new multi-model Bayesian optimization algorithm, consisting of multiple surrogate models addressing the challenge of model selection for autonomous chemical experimentation.

A new algorithm based on deep learning analyzes angiogenic morphogenesis images taken from angiogenesis on a chip. This method can assess the morphology of angiogenesis in great depth using multiple indicators and extract 3D indices from 2D images.

Using microfluidics, we isolate cancer cells under fluid flow mimicking sinusoidal capillaries. With deep-learning and FUCCItrack, we analyze 2D/3D time-lapse multi-channel images to study cell cycle dynamics, motility, volume, and morphology.

A new approach to combine digital microfluidics and machine learning algorithms to enable applications that require high throughput analysis.

We present a novel approach for the design of capillary-driven microfluidic networks using a machine learning genetic algorithm (ML-GA).

This work demonstrates how a small set of motion parameters uniquely measures a wide range of cell deformability in microfluidics.

Deep learning-enabled smartphone-based image processing has significant advantages in the development of point-of-care diagnostics.

A dual model object detection system for high precision monitoring of cell encapsulation statistics in microfluidic droplets with comparisons from YOLOv3 and YOLOv5 performance.

D-CryptO is a deep learning-based tool that can be used to analyze colon organoid structural maturity directly from brightfield images. D-CryptO can be applied in many cases such as analyzing organoids following chemotherapeutic drug treatment.

Improving surfactant-laden microdroplet size prediction using data-driven methods.

We proposed an explainable deep learning-based method to classify similar fluorescence colors for multiplex digital PCR in a single fluorescent channel.

To address the persistence of the COVID-19 pandemic, we have developed a novel point-of-care SARS-CoV-2 biosensor. This sensor has a limit of detection within an order of magnitude of traditional PCR and can provide an accurate measure of viral load.

Machine learning applied to impedance cytometry data enables biophysical recognition of cellular subpopulations over the apoptotic progression after gemcitabine treatment of pancreatic cancer cells from tumor xenografts.

Machine learning algorithms for cell classification via on-chip fluorescence microscopy are shown to be robust to microfluidic distortions due to cell displacement during acquisition.

We developed a method for label-free image identification and classification of peripheral blood nucleated cells flowing in a microfluidic channel, based on the subcellular structures of quantitative phase microscopy images.

Upper: predictions using the machine learning surrogate model with ensemble latent assimilation; bottom: recorded experimental images of each corresponding timestep.

Microspheres array based intelligent nanoscope processed data collection for deep learning training. The trained convolutional neural network model classified the different sizes of nanoparticle samples.

Replacing a human operator by an open source optical feedback control loop for acoustofluidic focusing of biological cells (e.g. cancer cells in different resonance modes), micro- and nanometer particles results in an improved device performance.

The handheld, do-it-yourself ptychographic whole slide scanner for high-throughput digital pathology applications.

Our multicellular coculture array with the integration of machine learning analysis is able to predict adverse cutaneous drug reactions.

A successful outcome of the coupling between microfluidics and AI: neural networks tackle the signal processing challenges of single-cell microfluidic impedance cytometry.

A deep-learning-based image restoration method enhances the performance of imaging flow cytometry.

Lightweight and reliable deep-CNN for speeding up the computation of the quantitative phase maps of flowing/rolling cells and for retrieving the 3D tomograms of each cell by holographic flow cytometry modality.

We propose to employ NN-enhanced IFC to achieve both real-time single-cell intrinsic characterization and intrinsic metric-based cell classification at high throughput.

A quantitative particle agglutination assay using mobile holographic imaging and deep learning is demonstrated for point-of-care testing.