Already a world leader in image enhancement, ContextVision is now taking its medical analysis know-how into a new field. With the development of our INIFY™ portfolio, we aim to empower pathologists in the new digital era.
Digital Pathology
Intelligent support for life-saving decisions
INIFY™ Prostate – based on the latest breakthroughs in deep learning
After years of intense research and development, we are proud to announce our upcoming decision support software for digital pathology: INIFY™ Prostate, with image analysis algorithms based on the latest breakthroughs in AI and deep learning.
INIFY™ Prostate will automatically separate slides containing cancer from those that do not, before the pathologist has even opened the case. Thus, pathologist’s worklists will be presented in a worst-first order, meaning that the patient cases that contain most cancer – as well as the slides within each case – are immediately shown at the top.
Built on advanced artificial intelligence technology – deep learning – INIFY™ Prostate has the potential to drastically speed up diagnosis of scanned, H&E stained prostate biopsy samples.
INIFY™ Prostate will automatically separate slides containing cancer from those that do not, before the pathologist has even opened the case. Thus, pathologist’s worklists will be presented in a worst-first order, meaning that the patient cases that contain most cancer – as well as the slides within each case – are immediately shown at the top.
Built on advanced artificial intelligence technology – deep learning – INIFY™ Prostate has the potential to drastically speed up diagnosis of scanned, H&E stained prostate biopsy samples.
Groundbreaking technology
The complex patterns in histopathology tissue samples make them challenging to evaluate with classic AI algorithms. However, new machine learning technology has proved to be very powerful for pattern recognition – handling tasks that are very challenging even for the human brain.
Our Decision Support Tools are built on deep learning algorithms, which in turn draw on the latest advances in Convolutional Neural Networks (CNNs). These visual analysis tools – widely used in facial recognition and image categorization technologies – are exceptionally powerful in their ability to identify patterns of interest in images, with a hierarchical information representation similar to the human visual system.
The extremely sophisticated algorithms have been developed in collaboration with deep learning experts and pathology specialists. Based on a patent-filed, highly accurate method for auto-annotation of cancer tissue, they provide relevant and objective outcomes, helping pathologists conclude the diagnosis for each individual patient.
Our Decision Support Tools are built on deep learning algorithms, which in turn draw on the latest advances in Convolutional Neural Networks (CNNs). These visual analysis tools – widely used in facial recognition and image categorization technologies – are exceptionally powerful in their ability to identify patterns of interest in images, with a hierarchical information representation similar to the human visual system.
The extremely sophisticated algorithms have been developed in collaboration with deep learning experts and pathology specialists. Based on a patent-filed, highly accurate method for auto-annotation of cancer tissue, they provide relevant and objective outcomes, helping pathologists conclude the diagnosis for each individual patient.
Supporting the pathologist
Today’s pathologists are faced with an ever-increasing number of pathology samples. The actual process of image slide analysis is repetitive and time-consuming; for example a great deal of time and effort is spent on scrutinizing images that in the end turn out to be benign. Other tasks are very tedious – checking lymph nodes for metastases, or estimating the amount of cancer in the sample.
Combined with an all-too-common shortage of qualified pathologists, this risks creating bottlenecks in the treatment process, not to mention stressful working environments.
By speeding up the diagnostic workflow and taking over tedious routine tasks, our Digital Pathology portfolio will allow pathologists to realize their full potential as skilled specialists, helping them make the most of their valuable time. Our upcoming Decision Support Tools will provide relevant, objective information for faster, easier diagnoses, complete with quantified H&E data for smooth delivery to oncologists. The pathologist can easily check and confirm results, maintaining full control throughout the analysis and evaluation process.
Combined with an all-too-common shortage of qualified pathologists, this risks creating bottlenecks in the treatment process, not to mention stressful working environments.
By speeding up the diagnostic workflow and taking over tedious routine tasks, our Digital Pathology portfolio will allow pathologists to realize their full potential as skilled specialists, helping them make the most of their valuable time. Our upcoming Decision Support Tools will provide relevant, objective information for faster, easier diagnoses, complete with quantified H&E data for smooth delivery to oncologists. The pathologist can easily check and confirm results, maintaining full control throughout the analysis and evaluation process.
The experience behind our products
Our Digital Pathology team includes many highly experienced specialists from a broad range of fields: experts in immunology and biology, highly skilled project managers, QA/regulatory experts, software engineers, experts on deep learning and AI as well as marketers with a solid business focus. In addition, we have impressive external networks of pathologists, research labs, industry collaborators and other partners.
The development of our Digital Pathology portfolio builds on the findings of SL-DESUTO, a now finished Eurostars-sponsored research collaboration between ContextVision and the University of Applied Sciences Western Switzerland. ContextVision Advisory Board plays a key role in further strategic research & development within the business unit, providing us with unique and invaluable insights and feedback. The board features some of the most prominent names in the field:
The development of our Digital Pathology portfolio builds on the findings of SL-DESUTO, a now finished Eurostars-sponsored research collaboration between ContextVision and the University of Applied Sciences Western Switzerland. ContextVision Advisory Board plays a key role in further strategic research & development within the business unit, providing us with unique and invaluable insights and feedback. The board features some of the most prominent names in the field:
Anil Parwani
Professor of Pathology at
The Ohio State University, USA
The Ohio State University, USA
Dan Ciresan
CEO, Conndera Research,
Romania
Romania
Jeroen van der Laak
Associate Professor at Radboud UMC,
Nijmegen, the Netherlands
Nijmegen, the Netherlands
Junya Fukuoka
MD.PhD. Chair, Professor,
Dept. of Pathology,
Nagasaki University, Japan
Dept. of Pathology,
Nagasaki University, Japan
Fredrik Pontén
MD, PhD, Dept. of Immunology,
Genetics & Pathology,
Uppsala University, Sweden
Genetics & Pathology,
Uppsala University, Sweden
Henning Müller
Prof. Dr.
University of Applied Sciences and
Arts Western Switzerland
University of Applied Sciences and
Arts Western Switzerland
I am impressed by the company’s vision and their amazing efforts within AI and machine learning for radiology. I believe that these strengths can help solve some of the current clinical issues that we encounter in pathology today.
User-friendly tools and high-quality data
As part of the development process, we have created easy-to-use, interactive web-based viewers and annotation tools, as well as platforms for image processing.
Data and data quality are essential prerequisites for developing highly accurate deep learning networks on which our auto-annotation tool is based. By processing high amounts of digitalized tissue samples, the algorithm is trained to deliver accurate, objective annotations with the utmost precision.
A major source of data comes from ContextVision’s collaboration with ALAB in Poland, one of Poland’s largest providers of pathology services. Thanks to a number of research collaborations with academia and pathology departments throughout the world, additional data sources are available, opening up powerful opportunities for future product development.
Data and data quality are essential prerequisites for developing highly accurate deep learning networks on which our auto-annotation tool is based. By processing high amounts of digitalized tissue samples, the algorithm is trained to deliver accurate, objective annotations with the utmost precision.
A major source of data comes from ContextVision’s collaboration with ALAB in Poland, one of Poland’s largest providers of pathology services. Thanks to a number of research collaborations with academia and pathology departments throughout the world, additional data sources are available, opening up powerful opportunities for future product development.
Top placements in challenges
Our technological advances have landed us top positions in prestigious competitions using deep learning in digital pathology.
In Tumor Proliferation Assessment Challenge (TUPAC16), we were awarded 2nd place for our breast cancer mitosis detection algorithms and their ability to correctly predict tumor proliferation in whole-slide images. To reflect the actual conditions of the clinical diagnostic process, the challenged required algorithms to be applied to entire H&E stained tissue samples, and the results were summarized in a proliferation score that can be integrated in current prognostic grading systems.
Our results outshone those of many highly reputable academic teams as well as research groups from companies such as IBM and Microsoft.
In Tumor Proliferation Assessment Challenge (TUPAC16), we were awarded 2nd place for our breast cancer mitosis detection algorithms and their ability to correctly predict tumor proliferation in whole-slide images. To reflect the actual conditions of the clinical diagnostic process, the challenged required algorithms to be applied to entire H&E stained tissue samples, and the results were summarized in a proliferation score that can be integrated in current prognostic grading systems.
Our results outshone those of many highly reputable academic teams as well as research groups from companies such as IBM and Microsoft.
In CAMELYON17, deep-learning algorithms were used to analyze 500 whole-slide H&E lymph node images and calculate TNM staging of 100 breast cancer patients. The goal of this challenge was to evaluate new and existing algorithms for automated detection and classification of breast cancer metastases in whole-slide images – a task with high clinical relevance, and which normally requires extensive microscopic assessment by pathologists. ContextVision latest algorithms scored a very awesome 2nd place among all the accepted teams, now 60 in total, see Camelyon Challenge website.
Publications
Nikolay Burlutskiy, Feng Gu, Lena Kajland Wilen, Max Backman, Patrick Micke.
A Deep Learning Framework for Automatic Diagnosis in Lung Cancer
International Conference on Medical Imaging with Deep Learning (MIDL2018 in Amsterdam).
Lars Björk, Jonas Gustafsson, Feria Hikmet Noraddin, Kristian Eurén, Cecilia Lindskog.
A new high-throughput auto-annotation method to detect and outline cancer areas in prostate biopsies
Presentation at ECDP 2018 Helsinki, Finland.
Peter Bandi, Oscar Geesink, Ludwig Jacobsson, Martin Hedlund et al.
From detection of individual metastases to classification of lymph node status at the patient level: the CAMELYON17 challenge
IEEE Transaction on Medical Imaging PP(99):1-1, To be released.
Feng Gu, Nikolay Burlutskiy, Mats Andersson, and Lena Kajland Wilén.
Multi-Resolution Networks for Semantic Segmentation in Whole Slide Images
MICCAI 2018 – COMPAY Workshop on computational pathology, Granada, Spain.
Mikael Rousson, Martin Hedlund, Mats Andersson, Ludwig Jacobsson, Gunnar Lathen, Bjorn Norell, Oscar Jimenez-del-Toro, Henning Mueller, Manfredo Atzori.
Tumor proliferation grading from whole slide images
SPIE Medical Imaging, 2018, Houston, Texas, United States.
Sebastian Otalora, Manfredo Atzori, Vincent Andrearczyk, and Henning Muller.
Image Magnification Regression Using DenseNet for Exploiting Histopathology Open Access Content
MICCAI 2018 – COMPAY Workshop on computational pathology, Granada, Spain.
Sebastian Otálora, et. al.
Determining the scale of image patches using a Deep Learning Approach
ISBI 2018. Washington D.C., USA, 2018.
Otálora s., Andrearczyk V., Atzori M., Müller H.
BIWGAN: Learning stable adversarial representations for prostate histopathology images
Medical Imaging Summer School 2018: Medical Imaging meets Deep Learning. Favignana, Italy.
Oscar Jimenez del Toro, Manfredo Atzori, Sebastian Otálora, Mats Andersson, Kristian Eurén, Martin Hedlund, Peter Rönnquist and Henning Müller.
Convolutional neural networks for an automatic classification of prostate tissue slides with high-grade Gleason score
SPIE Medical Imaging, pages 101400O-101400O-9, 2017.
Oscar Jimenez del Toro, Sebastian Otálora, Mats Andersson, Kristian Eurén, Martin Hedlund, Mikael Rousson, Henning Müller and Manfredo Atzori.
Elsevier book on Texture Analysis, chapter Analysis of Histopathology Images
From Traditional Machine Learning to Deep Learning, 2017.
Oscar Jimenez del Toro, Sebastian Otálora, Manfredo Atzori and Henning Müller.
Deep Multimodal Case-Based Retrieval for Large Histopathology Dataset
MICCAI 2017 workshop on Patch-based image analysis, Quebec City, Canada, 2017.
(The paper will be published in Springer LNCS)
(The paper will be published in Springer LNCS)