A research team led by Professor Kevin Tsia, program director of the Biomedical Engineering Program under the Faculty of Engineering of the University of Hong Kong (HKU), has developed an AI-driven imaging tool that enables speedy and precise diagnosis of cancer patients, greatly enhancing the effectiveness of their medical treatment.
In a joint collaboration with HKU’s Li Ka Shing Faculty of Medicine (HKUMed) and Queen Mary Hospital, the team headed by Professor Tsia, has successfully demonstrated the effective use of their latest generative AI method, the Cyto-Morphology Adversarial Distillation (CytoMAD), on lung cancer patients as well as drug tests.
Combined with their proprietary microfluidic technology, CytoMAD allows fast and cost-effective “label-free” imaging of human cells to help clinicians assess a patients’ tumor at the precision of individual cells, and also determine whether patients have the risk of metastasis.
CytoMAD uses AI to automatically correct cell imaging inconsistencies, enhance cell images, and extract previously undetectable information from cell images. Such all-round capability in CytoMAD ensures accurate and reliable downstream data analysis and diagnosis. CytoMAD’s capabilities have the potential to revolutionize cell imaging for meaningful analysis of cell properties and related health and disease information.
“Until now, there was no cost-effective technique to do single-cell analysis through imaging mainly because of the limitation in scale. Under the traditional methods, the imaging throughput is not fast enough and the cell images are not clear and informative enough,” said Professor Tsia.
The team collaborated with Professor James Ho from the Department of Medicine under the School of Clinical Medicine and Professor Michael Hsin from the Department of Surgery under the School of Clinical Medicine at HKUMed. The research was published recently in an article entitled “Information-Distilled Generative Label-Free Morphological Profiling Encodes Cellular Heterogeneity” in the journal Advanced Science.
Uncovering cellular information
Faced with the challenge of low visibility of cell samples placed under the microscope, medical practitioners often resort to the usual method of applying stain and labels to the samples. But such an approach is time-consuming and far from cost-effective in the cumbersome process. This also means patients have to wait for a period of time before the results of their cell analysis, for example, through blood samples, are known.
A key advantage of Professor Tsia’s AI technology is that it is “label free,” hence requiring fewer steps to prepare patient or cell samples. This saves much time and manpower, adding to the speed and efficiency of the diagnosis and drug discovery process. “We use Generative AI technology to render much clearer label-free images with useful information such as whether a treatment has had a positive effect,” he said.
CytoMAD allows simultaneous label-free image contrast translation to reveal additional cellular information. “Our work primarily focuses on label-free imaging modalities (i.e., bright-field (BF) to quantitative phase image (QPI) translation) due to their growing significance in biomedicine in recent years.
“A classical bright-field cell image typically looks like a vague photo full of scattered fainted blobs—nowhere close to informative for meaningful analysis of the cell properties and thus the related health and disease information. Nevertheless, CytoMAD, as a generative AI model, can be trained to extract the information related to mechanical properties and molecular information of cells that was undetectable to the human eye in a brightfield image.
“In other words, we could uncover important properties of cells that underpin cell functions, bypassing the use of standard fluorescence markers and their limitations in costs and time,” explained Dr. Michelle Lo, a postdoctoral researcher in the Department of Electrical and Electronic Engineering of the Faculty of Engineering, who is the main developer of CytoMAD in this project.
Unbiased diagnosis
The novel approach also addresses the challenge of “batch effect”—common unspoken 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.
Current solutions, including those based on machine learning techniques, often require different types of a priori knowledge or assumptions about the data, making them not generalizable enough to be easily adopted in different applications. “Our AI model doesn’t require the need for any assumption. Hence, it allows unbiased cell image analysis and diagnosis.”
This powerful deep-learning model benefits from the ultrafast optical imaging technology that was also developed by Professor Tsia’s team. “This technology allows us to capture cell images at great speed. Every day, tens of millions of images can be generated. Therefore, leveraging this single system, we are in a unique position, among many AI innovations, to accelerate the advanced AI R&D—from training, optimization to deployment,” noted Professor Tsia.
The use of CytoMAD is not limited to lung cancer patients alone, though lung cancer remains a top killer among all cancer diseases globally and ranks as the No.1 cancer risk. It could reduce the often-lengthy process of drug screening, through the adoption of the time-saving “label free method,” as well as its advantages of high-speed imaging and diagnostic function powered by generative AI.
Looking ahead, a prime goal is to train the model to enable medical practitioners to predict cancer or other diseases for potential patients. “Making predictions based on a vast amount of data is the most powerful aspect of AI application in biomedicine,” said Professor Tsia.
Professor Tsia’s team has applied for research funding to conduct clinical trials among lung cancer patients over a three-year period. “We plan to accumulate adequate data and track patients’ progress using our imaging and AI technology.”
More information:
Michelle C. K. Lo et al, Information‐Distilled Generative Label‐Free Morphological Profiling Encodes Cellular Heterogeneity, Advanced Science (2024). DOI: 10.1002/advs.202307591
Citation:
AI-driven tool speeds up cancer diagnosis with precise cell imaging (2025, February 17)
retrieved 17 February 2025
from https://medicalxpress.com/news/2025-02-ai-driven-tool-cancer-diagnosis.html
This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no
part may be reproduced without the written permission. The content is provided for information purposes only.
A research team led by Professor Kevin Tsia, program director of the Biomedical Engineering Program under the Faculty of Engineering of the University of Hong Kong (HKU), has developed an AI-driven imaging tool that enables speedy and precise diagnosis of cancer patients, greatly enhancing the effectiveness of their medical treatment. In a joint collaboration with HKU’s Li Ka Shing Faculty of Medicine (HKUMed) and Queen Mary Hospital, the team headed by Professor Tsia, has successfully demonstrated the effective use of their latest generative AI method, the Cyto-Morphology Adversarial Distillation (CytoMAD), on lung cancer patients as well as drug tests. Combined with their proprietary microfluidic technology, CytoMAD allows fast and cost-effective “label-free” imaging of human cells to help clinicians assess a patients’ tumor at the precision of individual cells, and also determine whether patients have the risk of metastasis. CytoMAD uses AI to automatically correct cell imaging inconsistencies, enhance cell images, and extract previously undetectable information from cell images. Such all-round capability in CytoMAD ensures accurate and reliable downstream data analysis and diagnosis. CytoMAD’s capabilities have the potential to revolutionize cell imaging for meaningful analysis of cell properties and related health and disease information. “Until now, there was no cost-effective technique to do single-cell analysis through imaging mainly because of the limitation in scale. Under the traditional methods, the imaging throughput is not fast enough and the cell images are not clear and informative enough,” said Professor Tsia. The team collaborated with Professor James Ho from the Department of Medicine under the School of Clinical Medicine and Professor Michael Hsin from the Department of Surgery under the School of Clinical Medicine at HKUMed. The research was published recently in an article entitled “Information-Distilled Generative Label-Free Morphological Profiling Encodes Cellular Heterogeneity” in the journal Advanced Science. Faced with the challenge of low visibility of cell samples placed under the microscope, medical practitioners often resort to the usual method of applying stain and labels to the samples. But such an approach is time-consuming and far from cost-effective in the cumbersome process. This also means patients have to wait for a period of time before the results of their cell analysis, for example, through blood samples, are known. A key advantage of Professor Tsia’s AI technology is that it is “label free,” hence requiring fewer steps to prepare patient or cell samples. This saves much time and manpower, adding to the speed and efficiency of the diagnosis and drug discovery process. “We use Generative AI technology to render much clearer label-free images with useful information such as whether a treatment has had a positive effect,” he said. CytoMAD allows simultaneous label-free image contrast translation to reveal additional cellular information. “Our work primarily focuses on label-free imaging modalities (i.e., bright-field (BF) to quantitative phase image (QPI) translation) due to their growing significance in biomedicine in recent years. “A classical bright-field cell image typically looks like a vague photo full of scattered fainted blobs—nowhere close to informative for meaningful analysis of the cell properties and thus the related health and disease information. Nevertheless, CytoMAD, as a generative AI model, can be trained to extract the information related to mechanical properties and molecular information of cells that was undetectable to the human eye in a brightfield image. “In other words, we could uncover important properties of cells that underpin cell functions, bypassing the use of standard fluorescence markers and their limitations in costs and time,” explained Dr. Michelle Lo, a postdoctoral researcher in the Department of Electrical and Electronic Engineering of the Faculty of Engineering, who is the main developer of CytoMAD in this project. The novel approach also addresses the challenge of “batch effect”—common unspoken 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. Current solutions, including those based on machine learning techniques, often require different types of a priori knowledge or assumptions about the data, making them not generalizable enough to be easily adopted in different applications. “Our AI model doesn’t require the need for any assumption. Hence, it allows unbiased cell image analysis and diagnosis.” This powerful deep-learning model benefits from the ultrafast optical imaging technology that was also developed by Professor Tsia’s team. “This technology allows us to capture cell images at great speed. Every day, tens of millions of images can be generated. Therefore, leveraging this single system, we are in a unique position, among many AI innovations, to accelerate the advanced AI R&D—from training, optimization to deployment,” noted Professor Tsia. The use of CytoMAD is not limited to lung cancer patients alone, though lung cancer remains a top killer among all cancer diseases globally and ranks as the No.1 cancer risk. It could reduce the often-lengthy process of drug screening, through the adoption of the time-saving “label free method,” as well as its advantages of high-speed imaging and diagnostic function powered by generative AI. Looking ahead, a prime goal is to train the model to enable medical practitioners to predict cancer or other diseases for potential patients. “Making predictions based on a vast amount of data is the most powerful aspect of AI application in biomedicine,” said Professor Tsia. Professor Tsia’s team has applied for research funding to conduct clinical trials among lung cancer patients over a three-year period. “We plan to accumulate adequate data and track patients’ progress using our imaging and AI technology.” More information: Citation: This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no
Uncovering cellular information
Unbiased diagnosis
Michelle C. K. Lo et al, Information‐Distilled Generative Label‐Free Morphological Profiling Encodes Cellular Heterogeneity, Advanced Science (2024). DOI: 10.1002/advs.202307591
AI-driven tool speeds up cancer diagnosis with precise cell imaging (2025, February 17)
retrieved 17 February 2025
from https://medicalxpress.com/news/2025-02-ai-driven-tool-cancer-diagnosis.html
part may be reproduced without the written permission. The content is provided for information purposes only.
