The Digital Revolution in Your Cells: How AI is Redefining Cancer Diagnosis
For over 150 years, the cornerstone of cancer diagnosis has remained largely unchanged: a pathologist examining a thin slice of tissue on a glass slide through a microscope. This traditional method, while effective, relies heavily on manual observation and requires significant time and specialised expertise. However, a quiet revolution is unfolding as researchers digitise these glass slides into high-resolution pixels, allowing Artificial Intelligence (AI) to assist in the diagnostic process. This field, known as Digital Pathology, is transforming the landscape of modern medicine.
From Glass Slides to Digital Maps
The transition to digital pathology begins with converting physical tissue samples into Whole Slide Images (WSIs). These files are massive, often containing billions of pixels, which allow a pathologist to zoom from a broad “tissue view” down to the level of individual cells. Beyond simple viewing, this digitisation enables the instantaneous sharing of samples across the globe, allowing a clinician in a remote area to consult with world-leading experts in seconds.
Research at Sunway University
In Malaysia, this cutting-edge research is one of the key areas of interest at 㽶Ƶwithin the Department of Data Science and Artificial Intelligence. The efforts are led by Dr Wan Siti Halimatul Munirah Wan Ahmad, who brought extensive experience in digital pathology from her previous academic appointments. The core of the research involves the development of sophisticated computational frameworks designed to handle the complexities of human tissue. While the team is currently in the foundational phase of their projects at the university, the work has already garnered international industry support. For instance, the department’s technical capability was recognised with an NVIDIA Academic Grant, providing the high-performance computing power necessary to train the complex AI models at the heart of their studies.
Advancing the “UNI2-h” Foundation Model
A key pillar of the current work involves the use of UNI2-h, a state-of-the-art Foundation Model. In the context of AI, a foundation model is one that has been pre-trained on a massive diversity of data, giving it a broad, generalised understanding of tissue architecture. The team is focused on enhancing this model by adding a specialised multi-scaled Segmentation Head. This technical addition is designed to achieve robust, instant segmentation, the process of automatically identifying and “colouring in” the boundaries of different cell types and structures. Unlike older models that might struggle with variations in image quality or staining techniques, this enhanced UNI2-h model is designed to be highly resilient, providing consistent and accurate data regardless of the lab environment.
The Impact on Clinical Outcomes
The integration of sophisticated models like UNI2-h offers several transformative benefits for the medical field:
- Precision: By automating the identification of cell boundaries, the AI provides a level of quantitative detail that is difficult for a human to replicate manually.
- Efficiency: Processes that traditionally took hours of manual counting can be completed almost instantly, allowing pathologists to focus on complex diagnostic decisions.
- Predictive Insights: These models can detect subtle morphological patterns that may predict how a specific tumour will respond to treatment, paving the way for more personalised patient care.
Conclusion
The field of digital pathology is evolving from a simple image-viewing tool into a sophisticated diagnostic partner. Through the research conducted at Sunway University, the “art” of pathology is being augmented by data-driven science. Supported by high-performance technology, these advancements promise to make cancer diagnosis faster, more consistent, and ultimately more effective for patients worldwide.
This article is from a recent unpublished work by 㽶Ƶteam led by Dr Wan Siti Halimatul Munirah Wan Ahmad, funded by the NVIDIA Academic Grant. Her work and code on digital pathology can be found in Google Scholar (http://bit.ly/wshmunirah) and GitHub (https://github.com/mizjaggy18).
Dr Wan Siti Halimatul Munirah Wan Yahya @ Wan Ahmad
Faculty of Engineering & Technology
School of Computing and Artificial Intelligence
Email: @email