AI-Driven Tool to Accelerate Cancer Diagnosis

In order to address the challenge of low visibility when examining cell samples under a microscope, medical professionals typically use staining and labeling techniques. However, this process is not only time-consuming but also costly. As a result, patients often face delays in receiving the results of their cell analysis, such as blood sample tests. Another significant issue is the "batch effect," which refers to technical variations across different experimental batches and conditions, such as changes in instrument settings or image acquisition protocols. These variations can hinder the accurate biological interpretation of cell morphology. Existing solutions, including machine learning-based approaches, often rely on specific prior knowledge or assumptions about the data, making them less adaptable and harder to implement in diverse applications. Researchers have now developed an AI-driven imaging tool that enables faster and more accurate diagnosis of cancer patients, significantly improving the effectiveness of their treatment.

In collaboration with other institutions, researchers from the University of Hong Kong (HKU, Hong Kong) successfully demonstrated their latest generative AI method, Cyto-Morphology Adversarial Distillation (CytoMAD), on lung cancer patients and drug tests. Combined with their proprietary microfluidic technology, CytoMAD facilitates fast, cost-effective, "label-free" imaging of human cells. This innovation enables clinicians to assess tumors at the precision of individual cells and determine if the patient is at risk for metastasis. Published in the journal Advanced Science, the study highlights how CytoMAD uses AI to automatically correct inconsistencies in cell imaging, enhance cell images, and extract previously undetectable details. This comprehensive capability of CytoMAD ensures reliable and accurate data analysis and diagnosis. The technology holds the potential to revolutionize cell imaging, providing critical insights into cell properties and related health and disease information.

A significant advantage of this AI technology is its label-free nature, which simplifies the preparation of patient or cell samples. This reduces time and labor, enhancing the speed and efficiency of diagnosis and drug discovery. CytoMAD also enables simultaneous label-free image contrast translation, revealing additional cellular details. Moreover, this novel approach addresses the issue of the "batch effect." The deep-learning model is supported by ultra-fast optical imaging technology, developed by the same research team. While lung cancer remains one of the most lethal cancers globally and a top cancer risk, CytoMAD’s utility is not limited to lung cancer patients. The technology could streamline drug screening processes, thanks to the time-saving "label-free" method, alongside its advantages in high-speed imaging and diagnostic capabilities powered by generative AI. Looking ahead, a key goal is to further train the model to help medical practitioners predict cancer and other diseases in potential patients.

“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,” said Dr. Michelle Lo, the main developer of CytoMAD in this project. “Nevertheless, CytoMAD, as 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.”