A research team at the University of Hong Kong (HKU) has developed an artificial intelligence-driven imaging tool designed to improve the speed and accuracy of cancer diagnosis. Led by Professor Kevin Tsia from the Faculty of Engineering, the team has introduced Cyto-Morphology Adversarial Distillation (CytoMAD), a generative AI-based method that enables precise single-cell analysis without requiring conventional labelling techniques. The technology has been tested in collaboration with HKU’s Li Ka Shing Faculty of Medicine and Queen Mary Hospital, demonstrating its effectiveness in assessing lung cancer patients and supporting drug-screening processes.
CytoMAD enhances cell imaging by automatically correcting inconsistencies, improving image clarity, and extracting previously undetectable information. This advancement allows for more reliable data analysis, facilitating improved medical decision-making. The technology integrates with a proprietary microfluidic system, enabling rapid and cost-effective imaging of human cells. By providing high-resolution single-cell imaging, it supports clinicians in evaluating tumour characteristics and assessing metastasis risk.
Traditional imaging methods require staining and labelling of cell samples, a process that is both time-consuming and labour-intensive. CytoMAD eliminates this requirement, streamlining sample preparation and accelerating diagnostic workflows. The AI model translates standard bright-field images into more detailed representations, revealing cellular properties that are typically difficult to analyse. This transformation is achieved through generative AI algorithms trained to extract information related to mechanical and molecular properties that would otherwise remain undetectable.
A key limitation of existing cell imaging techniques is their reliance on slow and costly processes that may delay critical treatment decisions. Many current solutions depend on fluorescence markers, which require additional steps and increased costs. CytoMAD provides a label-free alternative, reducing these constraints while maintaining accuracy. By leveraging generative AI, the system converts low-contrast bright-field images into more informative visualisations, offering deeper insights into cell morphology without the need for chemical staining.
Another challenge in cell imaging is the variability introduced by differences in equipment configurations and imaging protocols, often referred to as the “batch effect.” Such inconsistencies can hinder accurate biological interpretation. Many existing machine learning solutions rely on predefined assumptions about data, limiting their adaptability. CytoMAD overcomes this limitation by functioning without predefined data constraints, allowing for a more objective and generalisable approach to cell image analysis.
The system benefits from an ultrafast optical imaging technology developed by the research team, enabling the capture of millions of cell images daily. This high-throughput capability accelerates AI model training, optimisation, and implementation. By leveraging this technology, the research team aims to refine AI-driven biomedical imaging solutions further. The ability to rapidly process vast amounts of cellular data positions CytoMAD as a powerful tool for both clinical applications and medical research.
Beyond lung cancer diagnosis, CytoMAD has the potential to expedite drug discovery by reducing the time required for screening processes. Its combination of high-speed imaging and AI-driven analysis offers a more efficient alternative to traditional methods. The ability to rapidly assess cellular responses to treatments could improve drug development timelines, making the technology valuable for pharmaceutical research.
In the long term, the research team envisions expanding the application of CytoMAD to predictive healthcare, aiming to train the model to detect early signs of cancer and other diseases. Future developments may focus on integrating the system into clinical practice for real-time patient monitoring and personalised treatment planning. AI’s ability to analyse vast datasets and detect subtle cellular changes could enhance early disease detection, leading to improved patient outcomes.
The research team seeks funding for a three-year clinical trial on lung cancer patients, using AI-enhanced imaging to track outcomes. This study could drive wider AI adoption in medical diagnostics, improving efficiency and scalability in healthcare solutions.
