Healthcare (Commonwealth Union) – Diagnosis that occurs quick and accurately, plays a key role in its treatment. Scientists at QIMR Berghofer, Australia have unveiled a new artificial intelligence (AI) tool that could transform cancer diagnosis by giving pathologists a kind of “super vision” to detect hidden genetic markers in routine tissue samples.
The machine learning tool, named STimage, uses advanced spatial biology analysis to identify subtle molecular patterns that conventional pathology methods can miss. According to new research published in Nature Communications, STimage successfully predicted breast, skin, and kidney cancers, as well as a liver immune disorder. The study found the tool to be reliable, cost-effective, and capable of delivering results quickly in a format pathologists can easily interpret.
Associate Professor Quan Nguyen, who led the development of STimage with QIMR Berghofer’s National Centre for Spatial Tissue and AI Research (NCSTAR), described the technology as a breakthrough for precision medicine.
Nguyen indicated that it is like giving pathologists the super resolution vision of Superman or Superwoman.
He further pointed out that they can scan millions of invisible biomarkers in a tiny tissue sample and identify the few that show signs of cancer and this is critical for earlier detection, more precise diagnosis, and better-informed treatment decisions.
The innovation has the potential to enhance access to specialist expertise, particularly for patients in regional and remote areas, and could mark the start of a new era in digital pathology and personalised medicine.
The researchers foresee this tool assisting pathologists in rural, regional, and urban settings by easing high workloads, improving diagnostic accuracy, and cutting the time needed for sample screening and analysis—offering access to vital molecular insights that are currently confined to specialized research centers.
“The STimage tool does not replace the experience and expertise of pathologists. Rather, it assists them in their important and technically challenging work, by providing extra information about cell types and genetic activity that they can’t see with their own eyes,” explained Associate Professor Nguyen.
Spatial biology is an emerging discipline that investigates the intricate molecular interactions within tissue environments, aiming to reveal the underlying causes of cancer and other illnesses. It offers insights that cannot be obtained through traditional pathology methods, which rely on examining hematoxylin and eosin (H&E)-stained slides under a microscope.
H&E staining has been the standard, cost-effective technique used by pathologists worldwide for over a century. The familiar blue and pink hues highlight cellular structures and tissue organization, helping doctors detect structural abnormalities, but they do not provide direct information about the tissue’s molecular activity.
According to Associate Professor Nguyen, the STimage tool performs spatial analysis on H&E slides to produce biologically-informed predictions of disease by identifying molecular patterns present in the tissue.
“It makes a diagnostic prediction and mathematically computes the level of certainty about the result. In a first, there is transparency about the result with the tool showing the reasons that led to the prediction, like specific tissue or cellular features, to help pathologists evaluate findings,” added Associate Professor Nguyen.
In recent years advancements in AI and other technologies have made it possible to analyse large samples within seconds, which has the potential to greatly enhance diagnostics and advance new treatments.
The tool was also able to accurately forecast patient outcomes and responses to treatment, correctly identifying individuals at high or low risk of survival and predicting whether they were likely to have a full or partial response to available medications. These capabilities are still in the early stages and are undergoing further refinement.
The team trained the model using machine learning and statistical approaches to analyze spatial patterns from anonymized datasets of breast, skin, and kidney cancers, as well as the liver condition primary sclerosing cholangitis.
While only a handful of similar tools exist in this emerging field, the researchers demonstrated that STimage outperformed these alternatives, offering additional advantages in terms of reliability and clarity in interpreting the model’s predictions.
