DL: PROVIZ - Prostate cancer visualization by MRI - Improved diagnostics using artificial intelligence

Prostate cancer is the most common form of cancer among men. MRI is currently the first examination performed when men are referred to the specialist health service due to suspicion of prostate cancer. Despite recent improvements in the imaging strategy, it is still difficult to distinguish between indolent and aggressive prostate cancer, and the final diagnosis is a procedure largely dependent on invasive and often unnecessary biopsies. A major bottleneck in current clinical practice is that the processing and evaluation of the MRI data is entirely performed by human experts (radiologists). These experts are competent but also time-limited and cost-intensive resources that cannot be scaled to the imaging demands expected in near future. Artificial intelligence (AI) has recruited tremendous interest in the field of radiology by its potential to improve diagnostic processes. By automating and supporting radiological workflow, the promise of AI is to substantially alleviate the workload on healthcare personnel. Moreover, AI exploits quantitative image information, thus potentially increasing diagnostic accuracy compared to what is possible to achieve with the current qualitative expert interpretation. The aim of the current project was to develop a transparent and interpretable artificial intelligence system to support prostate cancer diagnosis, thereby substantially reducing health care costs, alleviating the demand on medical personnel, and improving the diagnostic accuracy, treatment and quality of life for the patients. Development of decision support systems based on artificial intelligence requires a large amount of representative data, preferably generated at different sites across the world to ensure generalizability. To this end, PROVIZ has generated an extensive database of MRI scans and clinical information from Norway, the Netherlands and Taiwan, where data from more than 5000 patients are available. This data set will be a unique source for further research and development, also after the end of PROVIZ. In PROVIZ, we have exploited these data to develop an end-to-end automated system for detection of prostate cancer. The final decision-support tool has been patented, and the research has led to several spin-off innovations projects. We are currently performing a clinical trial for validation of the decision-support in clinical flow at St. Olavs University Hospital. This is the first own-developed AI product intended for use in a specific MRI application being tested under the new EU regulative for medical device software in Norway. Transdisciplinary collaboration has been a crucial success factor for achieving the goals.

Resultatene fra prosjektet har nytteverdi for ulike sektorer, og de har brakt forskningsfeltet videre. Vår kliniske studie er den første som gjennomføres i Norge hvor en egenutviklet KI-algoritme til bruk i MR-diagnostikk valideres i klinisk flyt (med godkjenning i hht EUs nye regulativ om utprøving av medisinsk utstyr). Prosjektet har dessuten samlet store mengder annoterte data, noe som vil være viktig grunnlag for ny forskning i et felt som er i rivende utvikling. Prosjektet har hatt stor nytteverdi for kompetanseutvikling. Det tette samarbeidet mellom klinikk og teknologi har etablert et solid grunnlag for ny/videre forskning basert på større forståelse av kliniske problemstillinger og hvordan KI kan være et hjelpemiddel. Det etablerte transdisiplinære samarbeidet og rammeverket for RRI vil også være nyttig når vi skal være med på KI-satsning i årene som kommer (EU, NFR). RRI fokuset og samarbeidet med Kreftforeningen har dessuten gitt oss erfaring i toveis-kommunikasjon med relevante sluttbrukere. Innovasjonsprosessen for vårt produkt pågår fortsatt. Patentsøknaden blir ferdigbehandlet neste år. Vi har hatt dialog med relevante stakeholders om lisensiering, og samtidig vurderes det om det kan være mulig å etablere et spin-off selskap for å håndtere kommersialiseringen av produktet. Det endelige svaret på om vårt produkt kan bidra til å redusere helsetjenestekostnader og etterspørsel på medisinsk personell, samt forbedre diagnostisk nøyaktighet, behandling og livskvalitet for pasientene, ligger derfor litt frem i tid.

Prostate cancer affects approximately 1 in 8 men during their lifetime. This number is expected to increase substantially due to the aging population, and new clinical diagnostic tools are urgently needed. Magnetic resonance imaging (MRI) has become a key component in the diagnostic workup. However, the interpretation of MRI images relies on the manual reading by experienced radiologists, which is a time and cost-intensive resource. Moreover, this process underuses the quantitative nature of the data. We hypothesize that better diagnostic performance is achievable by providing the radiologist with a decision support system based on artificial intelligence (AI). For such a system to work in clinical practice, it needs to be accurate as well as transparent and interpretable. We propose to develop a decision support system that combines transparent AI methods, deep learning and model-based imaging features, and clinical information to provide the radiologist with a new set of interpretable tools to more accurately and efficiently detect prostate cancer, differentiate between high-risk and low-risk disease, and target prostate biopsies. The foundation of this project is formed by a unique Norwegian dataset of >1600 patients with MRI examinations and clinical variables, and an interdisciplinary project team with dedicated experience in MRI, AI, urology and radiology. Our collaboration with international experts in the field ensures access to similar data from The Netherlands and Taiwan, enabling solutions that also cover challenges related to demographic and multi-center variance. The project has the potential to substantially reduce health care costs, alleviate the demand on medical personnel, and obtain better treatment stratification, less side-effects, and improved quality of life in a relatively short timeframe. Development of transparent and interactive processes for responsible research and innovation will be an integrated part of the project.