-
Dr. Ole-Johan Skrede Successfully Defends his Doctoral Thesis
26.04.2024
We extend our congratulations to Dr. Ole-Johan Skrede for successfully defending his doctoral thesis titled "Selected Studies on the Application of Histological Image Analysis in Cancer Diagnostics Using Deep Learning" on Friday, April 26, 2024. The dissertation took place at the Department of Informatics, Faculty of Mathematics and Natural Sciences, in the namesake's Ole-Johan Dahle’s House.
.
From left: Xing Cai, Supervisor Fritz Albregtsen, Anne Solberg, Ole-Johan Skrede, Anders Lundevold and Paul J van Diest Foto: Petter Bjørklund/UiT
.
Dr. Skrede's research focuses on advancing cancer diagnostics through the application of deep learning techniques to analyze histological images. One significant outcome of his work is the development of a method to estimate the prognosis of patients who have undergone colorectal cancer surgery. This innovative approach involves digital microscope image analysis to identify cancerous regions and assess their severity. By training deep learning models on tissue sections from approximately 2,500 patients, Dr. Skrede and his colleagues have developed a deep learning model that enhances the accuracy of prognosis predictions, leading to better stratification of patients to adjuvant chemotherapy after surgery. The research team has rigorously evaluated this methodology on over 1,000 patients to demonstrate its validity and usefulness in clinical practice. Notably, the new method allows identification of substantially more patients that could be spared from unnecessary adjuvant therapy.
.
Dr. Skrede's doctoral thesis comprises three papers published in high-impact journals, significantly contributing to medical research and the recent DoMore project, an ICT Lighthous Project supported by the Research Council of Norway. The papers highlight the integration of deep learning with traditional pathological markers to optimize treatment for patients suffering from colorectal cancer, the possibility to automatically segmented any type of tumor, perhaps even rare types not included in the model development, as well as laying the foundation for better design of deep learning studies in cancer diagnostics and beyond.
.
Before defending his thesis, Dr. Skrede presented a trial lecture at the same venue, on the subject: “Foundation models in cancer research”.
.
The adjudication committee
- Professor Paul J van Diest, Department of Pathology, University Medical Center Utrecht, the Netherlands
- Professor emeritus Arvid Lundervold, Department of Biomedicine, University of Bergen, Norway
- Professor Anne Solberg, Department of Informatics, University of Oslo, Norway
Supervisors
.
Ole-Johan Skrede's supervisors throughout his doctoral journey have been Professor Emeritus Fritz Albregtsen at the Department of Informatics, UiO, Norway, and the late Professor Håvard E. Danielsen, at the Institute for Cancer Genetics and Informatics (ICGI), Oslo University Hospital, Norway.
.
We are so grateful Ole-Johan will continue his research at ICGI, being an important contributor to many of our most prestigious projects.
We extend our gratitude to the committee members for their invaluable insights and to Professor Xing Cai for chairing the defense
.
-
Establishing guidelines for prediction models in medical deep learning is essential
15.01.2024
The increase in scientific publications on deep learning for cancer diagnostics in recent years is impressive, but the conversion of promising prototypes into automated systems for medical utilisation is still moderate. In a recent issue of the scientific journal "Nature Machine Intelligence", Paula Dhiman and colleagues published a comment highlighting the importance of planning evaluations of deep learning systems in advance by predefining study protocols.
Andreas Kleppe, Ole-Johan Skrede and Knut Liestøl from the Institute for Cancer Genetics and Informatics at Oslo University Hospital acclaim the recent initiative by Dhiman and colleagues, and have now published a response to this comment in the January issue of "Nature Machine Intelligence".
Challenges in validations of prediction models
Prototypes for medical deep learning systems frequently claim to perform comparable with or better than clinicians. Even among the best studies evaluating external cohorts, few predefine the primary analysis, which can lead to over-optimistic results due to adaptations of the system, patient selection, or analysis methodology. The lack of stringent evaluation of external data and the development or evaluation of systems on narrow or inappropriate data for the intended medical setting are significant concerns. This over-promising will erode trust in the technology, and may hinder its adoption in the medical clinic. More concerning is the utilisation of prediction models that have not been properly tested, which may result in harm to patients due decisions made based on ill-founded evidence.
Recommended guidelines
In an article published in Nature Reviews Cancer in 2021, "Designing deep learning studies in cancer diagnostics", Kleppe et al. defined a list of recommended protocol items for external cohort evaluation of a deep learning system (PIECES). Among other recommendations, PIECES advocates explicit specification of the primary analysis and any pre-planned secondary analyses that authors wish to commit themselves to report on, and requests that authors describe precisely how the proposed system was developed and how its performance will be assessed.
Since the PIECES article was published, many publications have cited it in support of the need for predefined analyses and external cohort validation, and some have explicitly followed the guidelines.
By implementing these guidelines, medical utilization of deep learning systems can be enhanced, by the way of proper evaluation and translation of promising prototypes into verified systems in clinical practise. Kleppe and colleagues additionally suggest incentives that may increase the uptake of the practice — for example, through endorsement from investors, funders and publishers.
-
The world’s first clinical study using AI on tissue sections to guide the choice of therapy for real patients
15.01.2024
Andreas Kleppe and Tarjei Sveinsgjerd Hveem will conduct the clinical study
A Norwegian study led by researchers at the Institute for Cancer Genetics and Informatics at Oslo University Hospital aims to determine whether AI can help doctors decide which patients need chemotherapy after colorectal cancer surgery. The study will involve about 2,000 patients from Norway, United Kingdom and other countries, and will test whether AI can assist clinicians in providing more personalized treatment.
The AI method, developed by the institute and called Histotyping (video below), works by analyzing digital images of biopsies processed into tissue sections. Specialized doctors in diagnostics and interpretations of changes caused by disease, pathologists, analyze the H&E-stained sections to determine the patient's prognosis and more. AI has been shown to provide supplemental information so that the combination of assessments by AI and pathologists is better than each of them are individually. The new study aims to show that this combination leads to more personalized treatment and benefits the patients.
The study's main investigator, Andreas Kleppe, believes that AI can help many colorectal cancer patients avoid unnecessary chemotherapy. This is the first clinical study to use AI in this way.
Just a few days into the new year, one of Norway’s main newspapers, Aftenposten, wanted to learn more about our study. Several members of our staff were captured by the photographer "in action", as Andreas Kleppe and Tarjei S. Hveem discussed the project with the journalist. The article can be read (in Norwegian) on aftenposten.no.
A video demonstrating Histotyping
-
Personalizing treatment for colorectal cancer patients by combining tissue-based biomarkers and ctDNA
01.12.2023
Combining artificial intelligence-generated digital pathology tools, conventional histopathological assessment and circulating tumor DNA (ctDNA) analysis can improve treatment stratification of patients with colorectal cancer after surgery. Kerr and colleagues outline this novel paradigm for personalized adjuvant treatment of colorectal cancer in a new publication in Nature Reviews Clinical Oncology.
Cancer recurrence is estimated to occur in 80% of patients with colorectal cancer (CRC) within 3 years after surgery. The selection of adjuvant therapy depends on conventional histopathological staging procedures, which constitute a blunt tool for patient stratification. The benefits of adjuvant therapy are relatively marginal, and it is clear that there is a need for better methods for selecting patients who will benefit the most from the treatment whilst sparing those who will not derive benefit.
-"The better we understand the likelihood of cancer recurrence, the better we can tailor our adjuvant therapy, providing a more truly personalized treatment", emphasizes David Kerr, Professor at the University of Oxford and former president of the European Society for Medical Oncology (ESMO)
Liquid biopsies detecting ctDNA have been shown to have clinical utility for early detection of recurrence through surveillance and thus have the potential to personalize the management of CRC patients. However, the analysis of ctDNA is costly, and the initial assessment of a patient's status usually occurs at least four weeks following curative surgery and two weeks after completing systemic therapy. This delay is due to the persistence of elevated levels of cell-free DNA for several weeks post-treatment. Given the uncertain consequences of delaying potential chemotherapy and the fact that some patients may not show detectable ctDNA at their initial follow-up assessment, we propose using tissue-based biomarkers to facilitate an early pre-selection of treatment.
Improved patient management
Current clinicopathological markers are insufficient to stratify patients with early-stage CRC accurately. In 2020, Skrede et al. demonstrated how artificial intelligence (AI) can be used to predict CRC patient outcome in a study in The Lancet (Skrede et al., The Lancet 2020). The AI marker, named DoMore-v1-CRC, predicts the likelihood of cancer-specific death directly from images of routine histopathology sections. Building on these findings, the marker has since then been integrated with established clinicopathological markers to provide a clinical decision support system (CDSS) for guiding the choice of adjuvant chemotherapy in stage II and III CRC without residual disease after surgery (Kleppe et al., Lancet Oncology 2022).
Compared to conventional risk stratification for adjuvant therapy, the proposed CDSS identifies a much larger group of patients with an excellent prognosis that are likely to have similar survival with and without adjuvant chemotherapy and can, therefore, be spared the severe side effects of the treatment.
Since the CDSS's recommendation can be determined within a few days after surgery, patients identified as high-risk can begin treatment soon after surgery. In addition, the CDSS would identify additional strong candidates for adjuvant chemotherapy among those who are ctDNA negative at first assessment. Patients classified as low risk by the CDSS would then enter a ctDNA monitoring program and receive treatment upon ctDNA detection, if any.
- "I believe that integrating tissue and blood-borne prognostic biomarkers, as we suggest in this article, does make sense in regard to a more personalized treatment", says Professor Kerr. With this combined approach, more than half the patients with high-risk stage II and III CRC can be spared from adjuvant treatment, as they are very unlikely to benefit from it. This novel paradigm will reduce the economic cost and personnel requirements and improve patient management by more truly personalized treatment – which is ultimately the goal!
Illustration of patient management using the combination of tissue-based biomarkers and ctDNA
-
The Institute for Cancer Genetics and Informatics receives funding from the Norwegian Cancer Society
27.11.2023
Research funding from the Norwegian Cancer Society is raised by the Norwegian public and is an important contribution to Norwegian cancer research environments. The number of applications for the annual call for proposals was again numerous this year, and the competition is strong. Our project aims to use artificial intelligence (AI) to improve risk stratification in patients with colorectal cancer that has spread to the liver. We are grateful that the Norwegian Cancer Society has chosen to support our work and ambitions, which will enable the Institute to continue its work to guide in tailoring treatment options for patients with colorectal cancer.
About the project
An ageing population comes with an increase in cancer incidence. Despite the many advancements in diagnosis, surgical technique, screening, and molecular characterisation, colorectal cancer (CRC) remains a major global health problem, being the second most common cancer and the second most common cause of cancer death in Norway. About 20% of CRC patients are diagnosed with distant metastasis at primary diagnosis, and an additional 25% develop distant metastasis after surgery for localised colorectal cancer. Treatment of colorectal liver metastasis (CLRM) is inconsistent, but resection and chemotherapy are the standard treatment methods in patients who are eligible for surgery. Among patients undergoing liver resection, approximately 40% develop recurrences within one year after surgery, illustrating the need for better tools to identify the proper treatment for each patient.
Artificial intelligence (AI) radically transforms our society, including healthcare and medical diagnostics. Deep learning (DL) is a subfield of AI that is well-suited to perform complex visual recognition tasks and has proven particularly useful in medical image analysis. Based on long-term experience in digital pathology, the Institute for Cancer Genetics and Informatics (ICGI) at Oslo University Hospital has, over the last 8 years, built a competent computing environment for DL in medical image analysis. Deep learning has been used to predict patient outcomes from Whole Slide Images (WSIs) of routine haematoxylin and eosin (H&E)-stained tissue sections from cancers and similar methodology will be utilised in the current project. The project aims to develop deep learning models for predicting recurrence and survival in patients with colorectal liver metastases treated with surgery, to tailor adjuvant treatment and surveillance programmes which in turn will improve survival and quality of life. By linking these predictions with a characterisation of cells and tissue, including morphology and cell types, the project aims to reveal biological mechanisms involved in metastasis and poor patient outcomes. Overall, the project's objectives are to improve risk stratification and identify patients who will benefit from aggressive treatment or those who should not undergo surgery based on their frailty and treatment prospects.
-
Welcome to the 5th CRC Network Meeting in Oxford (UK), 30 - 31. March 2023
27.02.2023
The themes at this years annual conference are on the biology and management of colorectal cancer. This is the 5th CRC Network meeting to be held in Queen’s College, Oxford.
We bring together scholars who enjoy the intimate atmosphere of an Oxford College, enabling discussion and potential collaboration between all attendees.
The CRC Network meeting is free, and open to all interested in advances in research and treatment options for patients suffering from colorectal cancer. Due to limited capacity, registration is required.
For more information, go to the website crcnetwork.net
-
By leveraging a deep learning marker, 1 of 3 colorectal cancer patients with local (lymph node) metastasis may safely avoid adjuvant chemotherapy.
15.08.2022
New research shows how to reduce morbidity, mortality, and costs associated with treatment after surgery for colorectal cancer by using deep learning models!
We previously demonstrated that deep learning models can predict whether or not a patient will die of colorectal cancer after surgery by analysing images of tissue sections commonly used in routine histopathological examinations. The accuracy was high compared to other markers, but no marker is 100% accurate.
Our new study in The Lancet Oncology shows precisely how the deep learning marker should be integrated with the markers currently used in the clinic and that the clinical decision support system combining all markers allows a better and more individualised selection of adjuvant chemotherapy. In particular, 1 of 3 patients with local (lymph node) metastasis may safely avoid adjuvant chemotherapy. The current standard of care for these patients, double-agent chemotherapy, is associated with substantial side-effects and even causes some deaths.
Read our new paper here: https://authors.elsevier.com/a/1fZcb5EIIgH-tC
Illustration: Decision tree combining DoMore-v1-CRC marker with T and N stage, and number of lymph nodes.
-
Institute Director Professor Håvard E. Danielsen awarded the Excellent Researcher prize 2022 from Oslo University Hospital
10.06.2022
Three scientists received awards for their outstanding research at a ceremony at Oslo University Hospital, Rikshospitalet, held on the 10th June 2022. Silje Fjellgård Jørgensen and Geir Ringstad both received the "Early career award", while Håvard E. Danielsen was awarded NOK 300.000 and the prestigious "Excellent Researcher prize". These annual prizes honour excellent scientific work at the hospital.
The awarding process is organized by the hospital's research committee, while an external Scientific Advisory Board has evaluated the candidates. Read about the award winners on the Oslo University Hospital's research-website: https://www.ous-research.no/home/ous/Homepage%20news/22909
-
Danielsen is this years recipient of the King Olav Vs Cancer Research Prize
03.02.2022
We congratulate Håvard E. Danielsen as the recipient of King Olav V's Cancer Research Prize 2022. He has been awarded the prize for having developed methods using artificial intelligence (AI) giving cancer patients a more precise prognosis and counteract overtreatment.
Danielsen is considered a pioneer and a world-leading expert in digital pathology and artificial intelligence. At Oslo University Hospital, he heads the Institute of Cancer Genetics and Informatics (ICGI).
King Olav V’s Cancer Research Prize
In 1992, the year after King Olav’s death, King Harald of Norway established a new prize for cancer research in honour of his late father.
The King Olav V Cancer Research Foundation, established by King Harald and the Norwegian Cancer Society awards a prize of 1,000,000 Norwegian kroner each year to a Norwegian cancer researcher or researchers who have contributed to the promotion of Norwegian cancer research.
His Majesty the King is responsible for the solemn presentation of the award during a ceremony most often held in the university’s auditorium in Oslo during the month of May.
Professor Danielsen and the Institute for Cancer Genetics and Informatics are deeply honored to receive this Award.
Many years experience at the Norwegian Radium hospital
With funding from the Norwegian Cancer Society, the now 64-year-old Danielsen began his career at The Norwegian Radium Hospital in 1987, as a research fellow within the field of image analysis and electron microscopy. Since 1992, he has held various management positions at the hospital, which became part of Oslo University Hospital in 2005.
In 2004, the hospital (Radiumhospitalet-Rikshospitalet) established an integrated institute that linked IT and biology, Norway's first institute for medical informatics. The Institute has been headed by Danielsen since.
Employees describe Danielsen as an innovative, creative, and visionary leader who throughout his career has been good at challenging established ideas. In addition to leading the institute, Danielsen also holds a Professor II position at the University of Oslo's Department of Informatics, and a "Visiting Professor of Cancer Informatics" position at Oxford University in the UK.
His interests in culture and music, golf and boating occupy a lot of his time, but this has not stopped him in publishing over 160 articles so far, in leading medical journals. Not surprisingly, 15 patent applications have also been filed.
- With King Olav V's Cancer Research Prize awarded from the same organization that supported my doctorate degree, the circle is now complete, says Danielsen. We have just started using artificial intelligence within health care. I am very grateful for this recognition and the funding will give us the opportunity to further develop new methods for the benefit of cancer patients, he adds.
Please enjoy the Norwegian Cancer Society's interview with Håvard E. G. Danielsen
Other articles:
From Det norske kongehus:
Delte ut Kong Olav Vs kreftforskningspris
From Oslo Universitetssykehus:
Håvard Danielsen tildelt Kong Olav Vs kreftforskningspris
From Oslo Cancer Cluster:;
AI Researcher gets canceraward
From NRK (Norwegian public broadcaster):
Håvard Danielsen at NRK Nyhetsmorgen
From Dagens Medisin:
Håvard Danielsen får Kreftforeningens pris for kreftforskning
Finner diagnose og prognose med kunstig intelligens
Vil finne ny kunnskap om «gammelt» konsept
From University of Oslo, Faculty of Mathematics and Natural Sciences
Håvard E.G. Danielsen årets vinner av Kreftforskningsprisen
From forskning.no:
Kunstig intelligens-forsker får kreftforskningsprisen
From Computerworld:
Kreftforskningsprisen 2022: Banebrytende bruk av KI
The Norwegian Cancer Society:
About King Olav V´s Cancer Research Award - including an overview of previous winners
Our Institute for Cancer Genetics and Informatics is so proud having now two employees who have been awarded the "Oscar for Norwegian Cancer Researchers" . In 2011 the head of the Section for Cancer Gynetics, Sverre Heim also received the award.
-
Artificial intelligence-based biomarkers in active surveillance of prostate cancer
03.02.2022
Active surveillance of prostate cancer aims to avoid or delay treatment for patients with indolent tumours, without compromising survival and quality of life. The patients are monitored regularly and only treated if they show signs of disease progression.
The ICGI recently received funding from Helse Sør-Øst for the project titled "Artificial intelligence-based biomarkers in active surveillance of prostate cancer". The project is a collaboration between four institutions in the South-Eastern Norway Regional Health Authority; Oslo University Hospital, Vestfold HT, Vestre Viken HT and Telemark HT.
We aim to develop artificial intelligence-based biomarkers to be used with the above-mentioned patient group, as there is currently no prognostic marker recommended for routine clinical use. We will do this by combining both new and existing molecular and image markers using machine learning and make a fully automated system that analyses all available cells in all available tissue samples.
Our hypothesis is that previous attempts to improve clinical risk classification has failed because they did not properly include tumour heterogeneity, spatiality and cellular feature co-occurrences in their designs. In-house developed software allows us to spatially align and analyse different features within tissue sample.
Finally, these results will be combined with clinicopathological parameters currently used for risk stratification of prostate cancer patients, into a new risk stratification tool.