A team of researchers from INESC TEC, the Netherlands Cancer Institute (NKI) and IMP Diagnostics has just published an important study in npj Digital Medicine (a journal from the Nature portfolio), which could shape the future of breast cancer diagnosis. The study reviews various virtual staining techniques that use advanced Artificial Intelligence (AI) and could enable healthcare professionals to make molecular diagnoses more accurately and efficiently, thereby facilitating the timely start of personalised treatment – with proven benefits for breast cancer patients.
And there are many patients: according to data from the Liga Portuguesa Contra o Cancro, close to 9,000 new cases of breast cancer are diagnosed in Portugal each year. So how is the disease currently diagnosed? The process involves several steps aimed at detecting the disease as early as possible, confirming the presence of malignant tumours, and tailoring treatment accordingly. One of the first steps in this process is tissue analysis using haematoxylin and eosin (H&E) staining, which allows doctors to assess the tissue’s morphology (but not the molecular biomarkers.
One of the most used methods to identify these biomarkers in breast cancer is immunohistochemistry (IHC), which detects whether certain receptors or proteins are present in tumour cells. This is crucial for identifying the cancer subtype and guiding treatment but requires multiple consecutive tissue slices and several hours of preparation and analysis.
With this challenge in mind, the team from INESC TEC, NKI, and IMP Diagnostics explored a new approach: virtual staining. This method uses advanced AI techniques to transform an image of a single H&E-stained tissue sample into a virtual IHC image – eliminating the need for physical staining.
“This new study presents a comprehensive review of the latest techniques for generating these virtual stains. In addition to explaining the main approaches, we also carried out a novel benchmarking study, evaluating the performance of different AI models using two publicly available breast cancer datasets,” explained Jaime Cardoso, INESC TEC researcher.
Now published in one of the world’s most prestigious scientific journals, npj Digital Medicine, the study will be a valuable resource for researchers and clinical staff seeking to understand or develop virtual staining techniques for breast cancer, said Sara Oliveira, a postdoctoral researcher at NKI. “Virtual staining has attracted considerable interest in the computational pathology community in recent years, with models being developed for various applications. This study provides a broad overview of the state of the art and provides insight into the performance levels of different methodologies,” she added.
The high prevalence of breast cancer makes research in this area particularly relevant – not only to help reduce the number of cases and deaths, but also to enable more effective treatment.
In this sense, Diana Montezuma Felizardo, an anatomical pathologist and researcher at IMP Diagnostics, states: “Virtual staining could revolutionise the way we analyse breast tumours, allowing us to extract essential information without additional tests, more quickly and at lower cost – potentially having a real impact on patients’ lives.”
The researchers mentioned that several state-of-the-art models can now perform this image-to-image transformation with high precision, preserving clinically important information from the original image. Most of these models use a technology called Generative Adversarial Networks (GANs), but alternative approaches are emerging (including diffusion-based models, the same technology behind some AI image generators).
In this study, INESC TEC co-led the technical supervision, IMP Diagnostics handled the clinical component, and the NKI was responsible for the literature review and comparative analysis.
The article, “H&E to IHC virtual staining methods in breast cancer: an overview and benchmarking”, is available here.
News, current topics, curiosities and so much more about INESC TEC and its community!
