High-Precision Cell Classification via Mechanical Traits Within Microfluidic Systems

Accurate sorting of cells without using external labels is often challenging, especially when dealing with closely related cell types. Commonly, fluorescent markers are used for identification, but they can be time-intensive and affect cell viability. A purely image-based approach may also miss subtle differences between cells. Researchers therefore looked for a microfluidic method that integrates both morphological and biomechanical indicators to achieve higher classification precision. The authors introduced MIML (Multiplex Image Machine Learning) to combine conventional brightfield images with simultaneously measured mechanical properties by employing a microfluidic device. By analyzing images alongside cell deformability, maximum velocity, and transit time, they sought to improve classification accuracy. This approach was specifically tested on white blood cells (WBCs) and colorectal tumor cells (HCT116), two populations that overlap in size but differ in stiffness and deformation behavior.

They microfabricated a microfluidic chip consisting of narrow channels that force cells to squeeze through, creating measurable differences in their deformation, velocity, and transit times. These mechanical features are then extracted alongside brightfield snapshots of individual cells. Mechanical data, normalized to account for variations in flow and cell size, are used to quantify how easily each cell changes shape. In parallel, brightfield images of the same cells are captured at key stages of the microfluidic channel entry and exit. Both data streams—mechanical features and image information—are aggregated into a single machine-learning framework that leverages convolutional networks for the image analysis and neural architectures for the feature-based classification.

“In this study, we address these limitations through the development of a novel machine learning framework, Multiplex Image Machine Learning (MIML). This architecture uniquely combines label-free cell images with biomechanical property data, harnessing the vast, often underutilized biophysical information intrinsic to each cell. By integrating both types of data, our model offers a holistic understanding of cellular properties, utilizing cell biomechanical information typically discarded in traditional machine learning models.“, the authors explained.

A convolutional neural network processed the images extracted from the microfluidic chip, focusing on morphological cues, while a separate neural network handled the mechanical features. These outputs were merged into a unified classification framework, enabling the system to consider both shape patterns in brightfield images and the cells’ physical responses to mechanical stress. By training MIML on 2521 samples, the authors were able to classify WBCs versus HCT116 cells with an accuracy of 98.3%. This constituted roughly an eight-percent improvement beyond what could be attained using images alone. The confusion matrix showed a corresponding F1 score of about 96.3%, reflecting a low misclassification rate across both cell types. The authors also highlighted a strong correlation—approximately 0.96—between a cell’s maximum velocity and the time it needed to traverse the narrow microfluidic channel, underscoring how these traits often separate more deformable cells (tumor cells) from stiffer ones (WBCs). Their microfluidic-approach thus demonstrated that combining brightfield images with mechanical readouts yields a more discriminating classification tool.

MIML provides an effective label-free technique for distinguishing cells based on shape and physical properties. By capturing and merging real-time data on cell deformation and passage dynamics, the framework surpasses the limitations of single-modality classifiers. This work underscores the significance of mechanical traits as underused cell markers and offers an example of how morphological and biophysical information can be integrated to identify different cell populations with high fidelity. The authors propose that such combined approaches may be extended to other challenging classification tasks—particularly where different cell types share similar appearances but respond to mechanical stress in distinct ways.

“Future work should explore the application of MIML to a wider range of cell types and investigate its integration into clinical workflows. Addressing limitations such as the need for specialized equipment to measure biomechanical properties and exploring methods to incorporate readily available clinical data could further enhance the model’s practicality. “, the authors concluded

Figures are reproduced from Islam, K., Paul, R., Wang, S. et al. MIML: multiplex image machine learning for high precision cell classification via mechanical traits within microfluidic systems. Microsyst Nanoeng 11, 43 (2025). https://doi.org/10.1038/s41378-025-00874-x under a CC BY 4.0 Attribution 4.0 International license.

Read the original article: MIML: multiplex image machine learning for high precision cell classification via mechanical traits within microfluidic systems

For more insights into the world of microfluidics and its burgeoning applications in biomedical research, stay tuned to our blog and explore the limitless possibilities that this technology unfolds. If you need high quality microfluidics chip for your experiments, do not hesitate to contact us.