The six-year BIGPICTURE project that began in February 2021 will collect three million scanned, digital pathology slides from various European hospitals, research organisations and pharmaceutical companies. One of the participating scientists is pathologist Yossra HS Zidi-Mouaffak, co-coordinator of BIGPICTURE’s Finnish node. She is focusing on how artificial intelligence can be used in pathology.
The project is participated in by 45 organisations from 15 countries. Finland’s contribution to the project is provided by Helsinki University Hospital (HUS), mainly through the Helsinki Biobank, and CSC – IT Center for Science. The BIGPICTURE platform is being built collectively by pathologists, researchers, AI developers, patient advocates, and industry representatives. The files are stored in a repository, which enables the creation of new, efficient AI applications to promote digitalization of diagnostic pathology and to bring novel methods for tissue analyses. Samples can be analysed with artificial intelligence.
Yossra HS Zidi-Mouaffak currently works as a pathologist for Helsinki Biobank (HUS) and is also a PhD student and researcher at the University of Helsinki in Professor Olli Carpén’s research group. One of Yossra Zidi-Mouaffak’s projects involves digital pathology and colorectal cancer.
“Colorectal cancer (CRC) is the second most deadly and the third most commonly diagnosed cancer in the world. It is also the second most common type of cancer in Finland. Most of the CRC patients are treated with surgery and oncological treatments depending on the stage of the disease,” says Zidi-Mouaffak.
Oncological treatments can involve chemotherapy and radiation therapy.
”In our project, we are focusing on a particular set of patients with stage II colorectal cancer for whom the risk benefit ratio of adjuvant chemotherapy in often marginal.”
Stage II colorectal cancer is considered as an early stage of the disease where tumor invasion remains “local” without metastatic dissemination to other distant parts of the body. The tumor will penetrate through the entire intestinal wall and may also extend to adipose tissue or an adjacent organ, but it does not yet spread to lymph nodes or to distant organs. About 75% of patients with stage II will remain cancer-free 5 years after surgery.
“Unfortunately, 25% of the patients will not and these patients could benefit from post-operative chemotherapy”, says Zidi-Mouaffak.
“The question thus is: how to assess which patients are at high risk of recurrence?” Our project’s ultimate aim is to provide a predictive tool in stage II colorectal cancer, requiring ideally a considerable amount of data and images for more reliable results. BIGPICTURE is providing both large amounts of data and AI tools for researchers. This obviously makes progress much faster in this area of research.”
Zidi-Mouaffak selects, annotates and analyses scanned microscope images obtained from cancer patients’ surgical tissue samples stained with Hematoxylin and Eosin (H&E). This results in tissue parts dyed according to their PH. H&E is a routine stain for pathologists that allows them to analyse through the microscope, the morphology of the cells as well as the other components of the tissue.
Two Finnish biobanks, Auria and Helsinki Biobank, are among institutions providing datasets, which include whole slide images and associated curated metadata. Such datasets are used to create machine learning models by means of convolutional neural networks. The artificial intelligence models analyse the images that have been previously selected and annotated.
“As a pathologist, I believe that machine learning has the potential of improving pathologists’ output. The machine learning algorithms can be used as diagnostic tools to achieve routine tasks where they would be obviously faster, and more accurate than a human eye.”
Zidi-Mouaffak gives a few simple examples of AI diagnostic tools: recognize and count cell divisions (called mitosis), count the number of certain immune cells in specific areas, or accurately assess cell proliferation indexes.
“However, AI tools used for making predictions of a disease’s outcome based on image data, are very challenging to develop. They still require long phases of testing and validation before they can actually be used in clinical practice.”
The BIGPICTURE project first creates a storage infrastructure that enables processing, storage and sharing of extremely large image files. Pathological images may be up to several gigabytes in size. Slide images are provided with metadata. This material can be used to develop artificial intelligence tools, such as algorithms. Deep-learning algorithms are taught to classify morphologically similar cohorts, that is, they analyse shapes and structures in the samples. Artificial intelligence is able to detect cancer signs, or biomarkers, and they can then be verified.
“Based on recent studies, we believe that artificial intelligence applied on pre-selected digital slides from well-curated cohorts, could provide an interesting alternative to the existing molecular and morphological predictive markers.”
The purpose of the research team that Zidi-Mouaffak is part of is to develop and verify a new, “predictive marker” that could facilitate the stratification of stage II colorectal cancer patients. The focus is on the morphological features of the tumour.
According to Zidi-Mouaffak, deep-learning algorithms can make surprisingly accurate predictions of certain types of cancer, but in many cases it is not known how the algorithm reaches its decision.
“It is a sort of black box. This clearly deserves more research, and here is where repositories like the ones developed by BIGPICTURE, become extremely relevant. This kind of research requires huge databases with very big numbers of high quality digital slides and metadata, which is the aim of BIGPICTURE.”
Ari Turunen
10.2.2022
Read article in PDF
Citation
Ari Turunen, Yossra HS Zidi-Mouaffak, & Tommi Nyrönen. (2022). BIGPICTURE helps pathology go digital. https://doi.org/10.5281/zenodo.8154477
More information:
BIGPICTURE
HUS Helsinki University Hospital
Helsinki Biobank
https://www.helsinginbiopankki.fi/en/
Auria Biobank
https://www.auria.fi/biopankki/en/
CSC – IT Center for Science
is a non-profit, state-owned company administered by the Ministry of Education and Culture. CSC maintains and develops the state-owned, centra- lised IT infrastructure.
https://research.csc.fi/cloud-computing
ELIXIR
builds infrastructure in support of the biological sector. It brings together the leading organisations of 21 Euro- pean countries and the EMBL European Molecular Bio- logy Laboratory to form a common infrastructure for biological information. CSC – IT Center for Science is the Finnish centre within this infrastructure.
https://www.elixir-finland.org
ELIXIR is partly funded by the European Commission