Dickson M. D. Siu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Dickson M. D. Siu has a total of 37 co-authors. Here are the top co-authors by number of shared publications:
- Kevin. K. Tsia
Department of Electrical and Electronic Engineering, The University of Hong Kong, Pokfulam, Hong Kong
10 shared publications - Kelvin C. M. Lee
Department of Electrical & Electronic Engineering, The University of Hong Kong, Pokfulam Road, Hong Kong
7 shared publications - Michelle C. K. Lo
Department of Electrical & Electronic Engineering, The University of Hong Kong, Pokfulam Road, Hong Kong
5 shared publications - Justin S. J. Wong
Department of Electrical and Electronic Engineering, The University of Hong Kong, Hong Kong, China
4 shared publications Bob M. F. Chung
The University of Hong Kong, Pokfulam Road, Hong Kong, Hong Kong
3 shared publications- James Chung-Man Ho
Division of Respiratory Medicine Department of Medicine School of Clinical Medicine LKS Faculty of Medicine The University of Hong Kong Hong Kong SAR China
3 shared publications
Latest Publications
- Open Access
- 11 December 2025
- Published by Springer Nature in Nature Communications
- Vol. 16 (1)
Generalizable morphological profiling of cells by interpretable unsupervised learning
The intersection of advanced microscopy and machine learning is transforming cell biology into a quantitative, data-driven field. Traditional cell profiling depends on manual feature extraction, which is labor-intensive and prone to bias, while deep learning provides alternatives but faces challenges with interpretability and reliance on labeled data. We present MorphoGenie, an unsupervised deep-learning framework for single-cell morphological profiling. By combining disentangled representation learning with high-fidelity image reconstruction, MorphoGenie creates a compact, interpretable latent space that captures biologically meaningful features without annotation, overcoming the “curse of dimensionality.” Unlike previous models, it systematically links latent representations to hierarchical morphological attributes, ensuring semantic and biological interpretability. It also supports combinatorial generalization, enabling robust performance across diverse imaging modalities (e.g., fluorescence, quantitative phase imaging) and experimental conditions, from discrete cell type/state classification to continuous trajectory inference. This provides a generalized, unbiased strategy for morphological profiling, revealing cellular behaviors often overlooked by expert visual examination.
- 11 December 2025
- Published by Springer Nature in Nature Communications
- Vol. 16 (1)
Open AccessGeneralizable morphological profiling of cells by interpretable unsupervised learning
The intersection of advanced microscopy and machine learning is transforming cell biology into a quantitative, data-driven field. Traditional cell profiling depends on manual feature extraction, which is labor-intensive and prone to bias, while deep learning provides alternatives but faces challenges with interpretability and reliance on labeled data. We present MorphoGenie, an unsupervised deep-learning framework for single-cell morphological profiling. By combining disentangled representation learning with high-fidelity image reconstruction, MorphoGenie creates a compact, interpretable latent space that captures biologically meaningful features without annotation, overcoming the “curse of dimensionality.” Unlike previous models, it systematically links latent representations to hierarchical morphological attributes, ensuring semantic and biological interpretability. It also supports combinatorial generalization, enabling robust performance across diverse imaging modalities (e.g., fluorescence, quantitative phase imaging) and experimental conditions, from discrete cell type/state classification to continuous trajectory inference. This provides a generalized, unbiased strategy for morphological profiling, revealing cellular behaviors often overlooked by expert visual examination.
- Open Access
On-The-Fly Live-Cell Intrinsic Morphological Drug and Genetic Screens by Gigapixel-per-second Spinning Arrayed Disk Imaging
Next-generation drug discovery and functional genomics require rapid, unbiased single-cell profiling at scale—demands that exceed the limited speed, throughput, and labor-intensive labeling constraints of conventional high-content image-based screening. We introduce spinning arrayed disk (SpAD), a high-throughput, label-free imaging platform for live-cell imaging that integrates continuous circular scanning, ultrafast quantitative phase imaging (QPI), and a novel circular array of 96 culture chambers. SpAD achieves an order-of-magnitude reduction in imaging time compared to traditional fluorescence-based workflows, while remaining compatible with standard cell culture workflows. By extracting rich biophysical features using intrinsic morphological (InMorph) profiling and machine learning, SpAD enables sensitive, large-scale screening of drug responses and CRISPR gene knockouts without labeling. Critically, label-free biophysical readouts from SpAD reveal mechanism-linked changes in mass, refractive index, subcellular textures, and light scattering that fluorescent labels often obscure. SpAD thereby resolves subtle phenotypes and heterogeneous subpopulations with high reproducibility, providing a robust, scalable foundation for precision cellular morphological assays.
Open AccessOn-The-Fly Live-Cell Intrinsic Morphological Drug and Genetic Screens by Gigapixel-per-second Spinning Arrayed Disk Imaging
Next-generation drug discovery and functional genomics require rapid, unbiased single-cell profiling at scale—demands that exceed the limited speed, throughput, and labor-intensive labeling constraints of conventional high-content image-based screening. We introduce spinning arrayed disk (SpAD), a high-throughput, label-free imaging platform for live-cell imaging that integrates continuous circular scanning, ultrafast quantitative phase imaging (QPI), and a novel circular array of 96 culture chambers. SpAD achieves an order-of-magnitude reduction in imaging time compared to traditional fluorescence-based workflows, while remaining compatible with standard cell culture workflows. By extracting rich biophysical features using intrinsic morphological (InMorph) profiling and machine learning, SpAD enables sensitive, large-scale screening of drug responses and CRISPR gene knockouts without labeling. Critically, label-free biophysical readouts from SpAD reveal mechanism-linked changes in mass, refractive index, subcellular textures, and light scattering that fluorescent labels often obscure. SpAD thereby resolves subtle phenotypes and heterogeneous subpopulations with high reproducibility, providing a robust, scalable foundation for precision cellular morphological assays.
- 5 September 2025
- Published by Royal Society of Chemistry (RSC) in Lab on a Chip
- Vol. 25 (20) , 5329-5341
High-throughput multimodal optofluidic biophysical imaging cytometry
Traditional biophysical cytometry has been limited by its low-dimensional phenotyping characteristics, often relying on only one or a few cellular biophysical phenotypes as readouts. This has perpetuated the perception that biophysical cytometry lacks the power to determine cellular heterogeneity. Here, we introduce a multimodal biophysical cytometry platform, termed quantitative phase morpho-rheological (QP-MORE) cytometry, which simultaneously captures a collection of high-resolution biophysical and mechanical phenotypes of single cells at ultrahigh throughput (>10 000 cells per s). Combined with a microfluidic constriction channel design, QP-MORE integrates ultrafast single-cell quantitative phase imaging (QPI) and high-throughput deformability cytometry to resolve subcellular structures and whole-cell rheology in a single pass. QP-MORE's optofluidic design enables label-free, multi-contrast imaging of cells flowing at ∼1 m s−1, achieving subcellular resolution unmatched by existing deformability-based platforms. To validate its precision, we developed a robust calibration protocol ensuring high accuracy in morpho-rheological measurements. We also deployed QP-MORE to dissect drug-induced biophysical heterogeneity in HL60 leukemia and MDA-MB-231 breast cancer cells treated with latrunculin B (actin depolymerizer) and cytochalasin D (actin capping agent). QP-MORE not only revealed drug-specific subcellular biophysical signatures, but also achieved 99% accuracy in classifying drug mechanisms, surpassing deformability cytometry (78–94%). This underscores the potential of QP-MORE in expanding the capability of biophysical cytometry, especially in advancing our understanding of cellular heterogeneity and drug interactions.
- 5 September 2025
- Published by Royal Society of Chemistry (RSC) in Lab on a Chip
- Vol. 25 (20) , 5329-5341
High-throughput multimodal optofluidic biophysical imaging cytometry
Traditional biophysical cytometry has been limited by its low-dimensional phenotyping characteristics, often relying on only one or a few cellular biophysical phenotypes as readouts. This has perpetuated the perception that biophysical cytometry lacks the power to determine cellular heterogeneity. Here, we introduce a multimodal biophysical cytometry platform, termed quantitative phase morpho-rheological (QP-MORE) cytometry, which simultaneously captures a collection of high-resolution biophysical and mechanical phenotypes of single cells at ultrahigh throughput (>10 000 cells per s). Combined with a microfluidic constriction channel design, QP-MORE integrates ultrafast single-cell quantitative phase imaging (QPI) and high-throughput deformability cytometry to resolve subcellular structures and whole-cell rheology in a single pass. QP-MORE's optofluidic design enables label-free, multi-contrast imaging of cells flowing at ∼1 m s−1, achieving subcellular resolution unmatched by existing deformability-based platforms. To validate its precision, we developed a robust calibration protocol ensuring high accuracy in morpho-rheological measurements. We also deployed QP-MORE to dissect drug-induced biophysical heterogeneity in HL60 leukemia and MDA-MB-231 breast cancer cells treated with latrunculin B (actin depolymerizer) and cytochalasin D (actin capping agent). QP-MORE not only revealed drug-specific subcellular biophysical signatures, but also achieved 99% accuracy in classifying drug mechanisms, surpassing deformability cytometry (78–94%). This underscores the potential of QP-MORE in expanding the capability of biophysical cytometry, especially in advancing our understanding of cellular heterogeneity and drug interactions.
- Open Access
Generalizable Morphological Profiling of Cells by Interpretable Unsupervised Learning
The intersection of advanced microscopy and machine learning is revolutionizing cell biology into a quantitative, data-driven science. While traditional morphological profiling of cells relies on labor-intensive manual feature extraction susceptible to biases, deep learning offers promising alternatives but struggles with the interpretability of its black-box operation and dependency on extensive labeled data. We introduce MorphoGenie, an unsupervised deep-learning framework designed to address these challenges in single-cell morphological profiling. Enabling disentangled representation learning integrated with high-fidelity image reconstructions, MorphoGenie possesses a critical attribute to learn a compact, generalizable and interpretable latent space. This facilitates the extraction of biologically meaningful features without human annotation, additionally overcoming the "curse of dimensionality" inherent in manual methods. Unlike prior models, MorphoGenie introduces a systematic approach to mapping disentangled latent representations to fundamental hierarchical morphological attributes, ensuring both semantic and biological interpretability. Moreover, it adheres to the concept of combinatorial generalization—a core principle of human intelligence— which greatly enhances the model’s capacity to generalize across a broad spectrum of imaging modalities (e.g., quantitative phase imaging and fluorescence imaging) and experimental conditions (ranging from discrete cell type/state classification to continuous trajectory inference). The framework offers a new, generalized strategy for unbiased and comprehensive morphological profiling, potentially revealing insights into cellular behavior in health and disease that might be overlooked by expert visual examination.
Open AccessGeneralizable Morphological Profiling of Cells by Interpretable Unsupervised Learning
The intersection of advanced microscopy and machine learning is revolutionizing cell biology into a quantitative, data-driven science. While traditional morphological profiling of cells relies on labor-intensive manual feature extraction susceptible to biases, deep learning offers promising alternatives but struggles with the interpretability of its black-box operation and dependency on extensive labeled data. We introduce MorphoGenie, an unsupervised deep-learning framework designed to address these challenges in single-cell morphological profiling. Enabling disentangled representation learning integrated with high-fidelity image reconstructions, MorphoGenie possesses a critical attribute to learn a compact, generalizable and interpretable latent space. This facilitates the extraction of biologically meaningful features without human annotation, additionally overcoming the "curse of dimensionality" inherent in manual methods. Unlike prior models, MorphoGenie introduces a systematic approach to mapping disentangled latent representations to fundamental hierarchical morphological attributes, ensuring both semantic and biological interpretability. Moreover, it adheres to the concept of combinatorial generalization—a core principle of human intelligence— which greatly enhances the model’s capacity to generalize across a broad spectrum of imaging modalities (e.g., quantitative phase imaging and fluorescence imaging) and experimental conditions (ranging from discrete cell type/state classification to continuous trajectory inference). The framework offers a new, generalized strategy for unbiased and comprehensive morphological profiling, potentially revealing insights into cellular behavior in health and disease that might be overlooked by expert visual examination.
- 30 July 2024
- Published by Royal Society of Chemistry (RSC) in Lab on a Chip
- Vol. 24 (17) , 4182-4197
Dispersion-free inertial focusing (DIF) for high-yield polydisperse micro-particle filtration and analysis
We developed dispersion-free inertial focusing to overcome the inherent (size-)dispersion of inertial focusing which imposes challenges on applications requiring uniform positioning of polydisperse particles, e.g. , microfiltration and flow cytometry.
- 30 July 2024
- Published by Royal Society of Chemistry (RSC) in Lab on a Chip
- Vol. 24 (17) , 4182-4197
Dispersion-free inertial focusing (DIF) for high-yield polydisperse micro-particle filtration and analysis
We developed dispersion-free inertial focusing to overcome the inherent (size-)dispersion of inertial focusing which imposes challenges on applications requiring uniform positioning of polydisperse particles, e.g. , microfiltration and flow cytometry.
- Open Access
- 12 June 2024
- Published by Wiley in Advanced Science
- Vol. 11 (29) , e2307591
Information‐Distilled Generative Label‐Free Morphological Profiling Encodes Cellular Heterogeneity
Image‐based cytometry faces challenges due to technical variations arising from different experimental batches and conditions, such as differences in instrument configurations or image acquisition protocols, impeding genuine biological interpretation of cell morphology. Existing solutions, often necessitating extensive pre‐existing data knowledge or control samples across batches, have proved limited, especially with complex cell image data. To overcome this, “Cyto‐Morphology Adversarial Distillation” (CytoMAD), a self‐supervised multi‐task learning strategy that distills biologically relevant cellular morphological information from batch variations, is introduced to enable integrated analysis across multiple data batches without complex data assumptions or extensive manual annotation. Unique to CytoMAD is its “morphology distillation”, symbiotically paired with deep‐learning image‐contrast translation—offering additional interpretable insights into label‐free cell morphology. The versatile efficacy of CytoMAD is demonstrated in augmenting the power of biophysical imaging cytometry. It allows integrated label‐free classification of human lung cancer cell types and accurately recapitulates their progressive drug responses, even when trained without the drug concentration information. CytoMAD also allows joint analysis of tumor biophysical cellular heterogeneity, linked to epithelial‐mesenchymal plasticity, that standard fluorescence markers overlook. CytoMAD can substantiate the wide adoption of biophysical cytometry for cost‐effective diagnosis and screening.
- 12 June 2024
- Published by Wiley in Advanced Science
- Vol. 11 (29) , e2307591
Open AccessInformation‐Distilled Generative Label‐Free Morphological Profiling Encodes Cellular Heterogeneity
Image‐based cytometry faces challenges due to technical variations arising from different experimental batches and conditions, such as differences in instrument configurations or image acquisition protocols, impeding genuine biological interpretation of cell morphology. Existing solutions, often necessitating extensive pre‐existing data knowledge or control samples across batches, have proved limited, especially with complex cell image data. To overcome this, “Cyto‐Morphology Adversarial Distillation” (CytoMAD), a self‐supervised multi‐task learning strategy that distills biologically relevant cellular morphological information from batch variations, is introduced to enable integrated analysis across multiple data batches without complex data assumptions or extensive manual annotation. Unique to CytoMAD is its “morphology distillation”, symbiotically paired with deep‐learning image‐contrast translation—offering additional interpretable insights into label‐free cell morphology. The versatile efficacy of CytoMAD is demonstrated in augmenting the power of biophysical imaging cytometry. It allows integrated label‐free classification of human lung cancer cell types and accurately recapitulates their progressive drug responses, even when trained without the drug concentration information. CytoMAD also allows joint analysis of tumor biophysical cellular heterogeneity, linked to epithelial‐mesenchymal plasticity, that standard fluorescence markers overlook. CytoMAD can substantiate the wide adoption of biophysical cytometry for cost‐effective diagnosis and screening.
- Open Access
Dispersion-free inertial focusing (DIF) for high-yield polydisperse micro-particles filtration and analysis
Inertial focusing excels at the precise spatial ordering and separation of microparticles by size within fluid flows. However, this advantage, brought by its inherent size-dependent dispersion, could turn into a drawback that challenges applications requiring consistent and uniform positioning of polydisperse particles, such as microfiltration and flow cytometry. To overcome this fundamental challenge, we introduce Dispersion-Free Inertial Focusing (DIF). This new method minimizes particle size-dependent dispersion while maintaining the high throughput and precision of standard inertial focusing, even in a highly polydisperse scenario. We demonstrate a rule-of-thumb principle to reinvent inertial focusing system and achieve an efficient focusing of particles ranging from 6 to 30 µm in diameter onto a single plane with less than 3 µm variance and over 95% focusing efficiency at highly scalable throughput (2.4-30 mL/hr) – a stark contrast to existing technologies that struggle with polydispersity. We demonstrated that DIF could be applied in a broad range of applications, particularly enabling high-yield continuous microparticle filtration and large-scale high-resolution single-cell morphological analysis of heterogeneous cell populations. This new technique is also readily compatible with the existing inertial microfluidic design and thus could unleash more diverse systems and applications.
Open AccessDispersion-free inertial focusing (DIF) for high-yield polydisperse micro-particles filtration and analysis
Inertial focusing excels at the precise spatial ordering and separation of microparticles by size within fluid flows. However, this advantage, brought by its inherent size-dependent dispersion, could turn into a drawback that challenges applications requiring consistent and uniform positioning of polydisperse particles, such as microfiltration and flow cytometry. To overcome this fundamental challenge, we introduce Dispersion-Free Inertial Focusing (DIF). This new method minimizes particle size-dependent dispersion while maintaining the high throughput and precision of standard inertial focusing, even in a highly polydisperse scenario. We demonstrate a rule-of-thumb principle to reinvent inertial focusing system and achieve an efficient focusing of particles ranging from 6 to 30 µm in diameter onto a single plane with less than 3 µm variance and over 95% focusing efficiency at highly scalable throughput (2.4-30 mL/hr) – a stark contrast to existing technologies that struggle with polydispersity. We demonstrated that DIF could be applied in a broad range of applications, particularly enabling high-yield continuous microparticle filtration and large-scale high-resolution single-cell morphological analysis of heterogeneous cell populations. This new technique is also readily compatible with the existing inertial microfluidic design and thus could unleash more diverse systems and applications.
- Open Access
Information-Distilled Generative Label-Free Morphological Profiling Encodes Cellular Heterogeneity
Image-based cytometry faces constant challenges due to technical variations arising from different experimental batches and conditions, such as differences in instrument configurations or image acquisition protocols, impeding genuine biological interpretation of cell morphology. Existing solutions, often necessitating extensive pre-existing data knowledge or control samples across batches, have proved limited, especially with complex cell image data. To overcome this, we introduce Cyto-Morphology Adversarial Distillation (CytoMAD), a self-supervised multi-task learning strategy that distills biologically relevant cellular morphological information from batch variations, enabling integrated analysis across multiple data batches without complex data assumptions or extensive manual annotation. Unique to CytoMAD is its “morphology distillation”, symbiotically paired with deep-learning image-contrast translation - offering additional interpretable insights into the label-free morphological profiles. We demonstrate the versatile efficacy of CytoMAD in augmenting the power of biophysical imaging cytometry. It allows integrated label-free classification of different human lung cancer cell types and accurately recapitulates their progressive drug responses, even when trained without the drug concentration information. We also applied CytoMAD to jointly analyze tumor biopsies across different non-small-cell lung cancer patients’ and reveal previously unexplored biophysical cellular heterogeneity, linked to epithelial-mesenchymal plasticity, that standard fluorescence markers overlook. CytoMAD holds promises to substantiate the wide adoption of biophysical cytometry for cost-effective diagnostic and screening applications.
Open AccessInformation-Distilled Generative Label-Free Morphological Profiling Encodes Cellular Heterogeneity
Image-based cytometry faces constant challenges due to technical variations arising from different experimental batches and conditions, such as differences in instrument configurations or image acquisition protocols, impeding genuine biological interpretation of cell morphology. Existing solutions, often necessitating extensive pre-existing data knowledge or control samples across batches, have proved limited, especially with complex cell image data. To overcome this, we introduce Cyto-Morphology Adversarial Distillation (CytoMAD), a self-supervised multi-task learning strategy that distills biologically relevant cellular morphological information from batch variations, enabling integrated analysis across multiple data batches without complex data assumptions or extensive manual annotation. Unique to CytoMAD is its “morphology distillation”, symbiotically paired with deep-learning image-contrast translation - offering additional interpretable insights into the label-free morphological profiles. We demonstrate the versatile efficacy of CytoMAD in augmenting the power of biophysical imaging cytometry. It allows integrated label-free classification of different human lung cancer cell types and accurately recapitulates their progressive drug responses, even when trained without the drug concentration information. We also applied CytoMAD to jointly analyze tumor biopsies across different non-small-cell lung cancer patients’ and reveal previously unexplored biophysical cellular heterogeneity, linked to epithelial-mesenchymal plasticity, that standard fluorescence markers overlook. CytoMAD holds promises to substantiate the wide adoption of biophysical cytometry for cost-effective diagnostic and screening applications.
- 5 January 2023
- Published by Royal Society of Chemistry (RSC) in Lab on a Chip
- Vol. 23 (5) , 1011-1033
Optofluidic imaging meets deep learning: from merging to emerging
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.
- 5 January 2023
- Published by Royal Society of Chemistry (RSC) in Lab on a Chip
- Vol. 23 (5) , 1011-1033
Optofluidic imaging meets deep learning: from merging to emerging
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.
- 9 September 2020
- Published by Royal Society of Chemistry (RSC) in Lab on a Chip
- Vol. 20 (20) , 3696-3708
Deep-learning-assisted biophysical imaging cytometry at massive throughput delineates cell population heterogeneity
An optofluidic imaging cytometry strategy that hierarchically analyzes single-cell biophysical morphology beyond millions of cells at high sensitivity and specificity.
- 9 September 2020
- Published by Royal Society of Chemistry (RSC) in Lab on a Chip
- Vol. 20 (20) , 3696-3708
Deep-learning-assisted biophysical imaging cytometry at massive throughput delineates cell population heterogeneity
An optofluidic imaging cytometry strategy that hierarchically analyzes single-cell biophysical morphology beyond millions of cells at high sensitivity and specificity.