DoMore! : In silico Pathology - Improving diagnosis by utilizing Big Data and software-driven automation of pathology

Cancers are heterogeneous; one region of a tumor is therefore not representative of the tumor as a whole. To account for this heterogeneity, most studies would have to increase their workload by a factor of 5-10. With limited resources and strongly limited access to pathologists, this can only be achieved by digitalizing and largely automating the analysis required to render a diagnosis and identify prognostic biomarkers. This is what the DoMore! project was designed to do. Centering on researching and developing ICT solutions to supplement or replace methods in pathology, DoMore! aimed to completely transfer the very complex thinking and decision-making from its current basis in visual observation to a computer basis with objective, reproducible algorithms. The concepts involved were based on image analysis and more specifically: deep learning, texture analysis, and quantification of DNA. We focused on three major cancer forms, i.e. lung, colorectal and prostate cancer, and work with large retrospective clinical materials with known outcomes. The extensive research within DoMore! have been done in close collaboration with researchers and clinicians from the University of Oxford?s Institute for Cancer Medicine and Cancer Hospital, University College London (UCL), University of Oslo (UiO), Vestfold Hospital Trust (SiV), Cheltenham General Hospital, University of Liverpool, University of Glasgow, the Arctic University Norway, Stavanger University Hospital, Akershus University Hospital and several departments at our own Oslo University Hospital, providing material from 11.454 patients with prostate, lung, colorectal, bladder, mammae or endometria cancer. We have analysed 57.326 tumour samples, producing a total of 2500TB data. This extensive material has allowed us to design and develop new AI solutions focusing on cancer prognosis. During this five-year course we have written code for more than 100 applications and scripts, resulting in the development of 7 finished products, and at least another 4 is expected over the next 2 years. Using convolutional neural networks, we have developed solutions such as Histotyping, tumour segmentation, automatic mitotic figure counting and focus detecting application. In addition to adapting methods such as DNA ploidy to work on scanners, and developed a framework to characterize the chromatin structure. DoMore! has sparked a great deal of interest from both national and international parties, due to several high impact academic publications. During the project course, 19 papers have been published with the mean / median IF = 22 / 8. In addition to our academic progress, we have been spreading the word via other pieces of multimedia such as news articles, illustrative videos, and podcasts. From integration of software in DIPS Arena to implementation of Histotyping at Oxford, OUS and Vestfold Hospital and the spin-off company DoMore Diagnostics AS, our artificial intelligence methods will certainly reach the patients and allow us to do more to find the optimal treatment for each patient.

By developing generic and objective digital prognostic markers for cancer, we have established robust systems and ICT solutions to supplement methods in pathology to increase productivity and quality, and hence treatment of cancer. With automated methods, we have been able to DoMore, and to properly address the heterogeneity with less resources than used in todays practice. We have contributed to the digitalization of pathology and paved the way for the transition from digital pathology to in silico pathology, by introducing AI/deep learning into tissue diagnostics. This transition will change pathology as we know it, and compensate for the shortage of pathologists and the uneven distribution of best practice. To bring our research to the clinic, thus the patient, we have provided DIPS with the necessary tools for integration of our products and also through the establishment of the company DoMore Diagnostics AS that will commercialize the products within the project.

Preproject number 255241 The main project is all about researching and developing ICT solutions to supplement or replace methods in pathology to increase productivity and quality and hence treatment of cancer, based on analysis of Big Data produced by digital pathology. Most prognostic studies suffer from undersampling. To account for heterogeneity most studies would have to increase the workload by a factor of 5-10. With limited resources and strongly limited access to pathologists, this cannot be achieved unless we manage to digitalize and largely automate both the preparation and analysis required to render a diagnosis and identify prognostic biomarkers. This is what this project is designed to do, and we have put together a group of international leaders in the different fields involved, from robotics and machine learning to digital image analysis and from tumor pathology to cancer surgery and oncology. Our ambition is to completely transfer the very complex thinking and decision-making from its current basis in visual observation to a computer basis with objective, reproducible algorithms. The concepts involved are based on image analysis and more specifically: deep learning, texture analysis, and quantification of DNA. We will focus on three major cancer forms, i.e. lung cancer, colorectal cancer and prostate cancer and work with large retrospective clinical materials with known clinical outcomes and the methods will be applied to routine paraffin-imbedded material. The DoMore! project will facilitate the long awaited digitalization of pathology and establish more efficient and objective cancer prognostication that can be made equally available to all patients. We expect a number of different project results; increased efficiency in pathology, methods and markers to aid the clinician to give better and more personalized treatment to cancer patients, patents and publications, products (algorithms, applications, services, data) and spin-off companies.