Powerful Biochip Combines Deformability Cytometry with Deep Learning Tracking for Cancer Cell Classification

Cancer requires advanced diagnostic approaches to inform treatment strategies accurately. Dr. Bee Luan Khoo and team unveiled a pioneering methodology harnessing a constriction-based deformability cytometry (cDC) platform and the power of deep learning for the precise classification of cancer cells.
Published in Cancer
Powerful Biochip Combines Deformability Cytometry with Deep Learning Tracking for Cancer Cell Classification
Like

https://www.nature.com/articles/s41378-023-00577-1

Cancer, characterised by its vast heterogeneity and many metastatic subtypes, requires advanced diagnostic approaches to categorise cell types to inform treatment strategies accurately. Conventional diagnostic methods frequently hinge on label-based analytical techniques, which not only come with a hefty price tag but are also notorious for their intricacy and time-intensive nature. In a recent breakthrough, Dr. Bee Luan Khoo and her team, Haojun Hua and Shangjie Zou, at the City University of Hong Kong, have unveiled a pioneering methodology harnessing a constriction-based deformability cytometry (cDC) platform and the power of deep learning for the precise classification of cancer cells.

 

The Power of Cellular Deformability as a Biomarker

Cellular deformability is a promising label-free biomarker in assessing the physiological condition of cells, revealing vital connections to a range of medical conditions, foremost among them being cancer. The deformability of cancer cells directly correlates with their propensity to metastasise. This underscores the significant potential of cellular deformability as a label-free biomarker for gauging the metastatic potential of cancer cells, paving the way for streamlined and cost-effective diagnostic assays.

 

The Constriction-Based Deformability Cytometry (cDC) Platform

Fig 1 Design of the Constriction-based deformability cytometry (cDC) platform. a) Schematic diagram of the microfluidic device. b) Top view of laminar flow velocity simulation of the microfluidic chip. c) Fluid flow velocity across 36 microconstrictions. d) Schematic view of the cDC platform's experimental setup and computational framework. e) Time-lapse imaging showing cell deformation and movement through a microconstriction.

The cDC platform is a highly sensitive, high-throughput, and cost-effective approach for precisely quantifying cell friction and retention, achieving a remarkable rate of approximately 25,000 cells per minute. The seamless operation of this platform is made feasible by integrating a sophisticated computational framework, ATMQcD, hinged on deep learning. This encompassing framework encompasses crucial processes such as automatic training set generation, multi-object tracking, segmentation, and the precise quantification of cellular deformability.

 

Deep Learning for Enhanced Classification

The research team harnessed the power of deep learning to enhance cell detection and tracking substantially. Their innovative multi-object tracking technique, ATMQcD, draws from the robust Yolov5 and Deep SORT frameworks, outpacing other existing methods in speed and efficiency, especially when simultaneously tracking multiple objects at high speeds. This groundbreaking application of deep learning holds immense promise for revolutionising high-throughput analyses in clinical settings. It presents a formidable tool for evaluating cellular deformability and accurately assessing the physiological state of cells.

 

Effective Stratification of Metastatic Subtypes

 Figure 2. Cell Behavior Analysis: a) Deformation of cells after entering the microconstriction. b) Plot of creeping time and cell size of MCF7 and MDA-MB-231 cells. Data points representing MCF7 and MDA-MB-231 were displayed in different colors.

 The cDC platform, in tandem with the ATMQcD computational framework, underwent rigorous validation using diverse cancer cell lines exhibiting varying degrees of metastatic potential. It achieved a classification accuracy of 92.4% when distinguishing between distinct types of breast cancer cell lines. Additionally, it effectively stratified cancer cells before and after hypoxia treatment exposure. Furthermore, the system's capacity to quantify stiffness through the microfluidic setup proved highly adept at evaluating the metastatic potential of samples featuring heterogeneous phenotypes of cancer cells. Notably, the ATMQcD system showcased exceptional proficiency in discerning cancer cells from leukocytes, boasting an impressive accuracy rate of 89.5%.

 

Broad Applicability Across Cancer Types

 The versatile applicability of the cDC + ATMQcD system was further substantiated through assessments involving lung and bladder cancer cell samples. Remarkably, the system discerned noteworthy alterations in invasiveness following hypoxia treatment in both cancer types. This widespread adaptability presents a promising indication of the system's viability for forthcoming clinical applications.

 

Conclusion

 The cDC + ATMQcD system developed by Dr Khoo’s team from the City University of Hong Kong marks a significant leap in cancer cell classification. Merging cellular deformability cytometry with deep learning, this system introduces a sensitive, high-throughput, and cost-effective approach for evaluating the metastatic potential of cancer cells across diverse cancer types. This pioneering system promises to reshape the landscape of cancer diagnostics, offering a means to stratify metastatic subtypes and inform critical treatment choices accurately.

Hua, H., Zou, S., Ma, Z. et al. A deformability-based biochip for precise label-free stratification of metastatic subtypes using deep learning. Microsyst Nanoeng 9, 120 (2023). https://doi.org/10.1038/s41378-023-00577-1

Please sign in or register for FREE

If you are a registered user on Research Communities by Springer Nature, please sign in