White paper: Scientific and economic valorisation of biomedical images archives (PACS) in medical research


Authors: Andrea Vaglienti & Karine Seymour

Existing biomedical imaging data are a fuel for the development of image related applications. In this paper we study 4 fields: in-silico clinical trials, imaging biomarkers, radiomics and imaging cognitive tools, which all constitute a major scientific point of interest and have great market expectations. Be prepared to value your existing imaging data, both scientifically and economically!

There is a change of paradigm in the medical R&D process. Up to recently, randomized controlled trials were the gold standard, took years and data were discarded after the trial, especially if the trial failed. We now know that analysing all clinical trial data and using real-world evidence based on novel biomarkers can significantly speed the R&D process and save money.

In-silico models for more efficient and successful clinical trials

For instance, old clinical trials data and real-world data can be used to develop in-silico models based on virtual cohorts of patients.

In-silico trials can significantly reduce costs and time to bring to the market new health products[1]. Not only they can reduce the amount of animal testing, they can also better address the safety and efficiency issues of clinical trials, thanks to better predictions of outcomes and better patient stratification.

We conducted a research on Clarivate Integrity℠, a portal providing information to support drug research and development[2]. We found that the number of publications using in-silico models has significantly increased in last years (Figure 1).

Figure 1 : In-silico models are more and more used

In-silico models often rely on routine medical data to build cohorts of virtual patients which are representative of patient diversity and disease variability.

Among all those data, medical images are increasingly used.

An example is the Living Heart Project[3]  by Dassault Systèmes, which is an accurate model obtained from MRI and CT exams. It aims to test performance and placement of cardiac devices through realistic simulation.

The Living Heart Project obtained a 5-year collaborative research agreement with the FDA.

New imaging techniques drive to imaging biomarkers development

Biomedical imaging is routinely used in many pathologies, for diagnostic and treatment decision. A patient may undergo different exams during his stay in a hospital, such as PET, MRI, CT scans or x-rays. All images are stored in clinical Picture Archiving Communicating Systems (PACS), and according to IBM researchers, medical images represent 90% of all medical data today[4].

Radiologists have access to more and more quantitative information, especially with new imaging techniques such as functional MRI or PET-MRI. However, imaging in clinical routine is still mainly qualitative or “morphologic”, based on image observation and focused on individual, intra-patient changes. The challenge is to obtain reproducible measurements, known as imaging biomarkers, which will give objective indication of the patient’s condition.

Biomarkers are required:

– for routine care,  as it is the basis of “personalized medicine”, that is the identification of the most appropriate treatment for every patient.

– for clinical research, as there is a need for objective measurements to assess the efficiency of a new health product and to target populations of patients who will most benefit from these new products.

Although the development process of an imaging biomarker is still complex[5], especially in its reproducibility part, it has some advantages over this site other biomarkers (e.g. genetic analysis from biopsy):

  • Precise localisation
  • Non-invasive
  • Less expensive
  • Routinely used in diagnostic and follow-up phases in many pathologies


A research on Clarivate Integrity℠ shows an increasing trend on the number of publications related to imaging biomarkers from early 2000 (Figure 2).

Figure 2: imaging biomarkers gain more and more interest

Radiomics: “images are more than pictures, they are data”[6]

Imaging biomarkers discovery is the focus of radiomics, a new discipline which consists in extracting more information from medical images using advanced feature analysis[7]  and finding correlations with genotypes and other phenotypes. Radiomics is at the heart of many research projects, as we can see on Figure 3 showing the number of publications related to radiomics referenced in Pubmed.

Figure 3 : Publications on radiomics have significantly increased over the last years (2017 = projections)

Results of radiomics studies are started to be implemented by private companies (e.g. Aquilab[8], HealthMyne[9], Oncoradiomics).

From big data to deep-learning

At the same time, a number of companies (e.g. IBM[10], Median with his partnership with Microsoft[11], Oncoradiomics) are developing cognitive imaging tools which can “learn as they go”, using deep learning technologies. Early 2017, the U.S. Food and Drug Administration (FDA) approved for the first time a machine learning application to be used in a clinical setting[12]. The application was filled by the company Arterys, whose medical imaging platform will help doctors diagnose heart problems.

iCAD also recently received Premarket Approval (PMA) from the FDA[13] for its digital breast 3D tomosynthesis cancer detection and workflow solution, built on deep learning technology. It had received CE marking in 2016.

These deep-learning algorithms need to be initiated with massive amounts of routine medical images and associated data representative of patient and disease variability


In-silico models, imaging biomarkers, radiomics and cognitive imaging tools have gained increasing interest in last years with many scientific publications and first industrial applications. Market analyses announce a large adoption of these techniques in the near future[14][15], in drug and medical device development first, then in clinical routine.

These applications all have in common the need for analysis of large amount of routine biomedical images currently stored in clinical PACS. These massive amounts of data represent treasures of resources which are still underexploited at this time. Current valorization of these data is essentially scientific, but will likely be also economic in the near future, as this is already the case for other types of medical data and biospecimens in particular. The development of high throughput tools automating the extraction from PACS and anonymization of relevant medical images and associated data render this valorisation possible.

[1] In silico Clinical Trials : how computer simulation will transform the biomedical industry.

[2] http://clarivate.com/?product=integrity

[3] https://www.3ds.com/products-services/simulia/solutions/life-sciences/the-living-heart-project/


[5] Imaging biomarkers roadmap for cancer studies. O’Connor J.P. , Aboagye E.O. ,et al., Nat Rev Clin Oncol. 2017 Mar;14(3):169-186

[6] Gillies RJ, Kinahan PE, Hricak H. Radiomics: Images Are More than Pictures, They Are Data. Radiology. 2016;278(2):563-577. doi:10.1148/radiol.2015151169.

[7] Lambin P, Rios-Velazquez E, et al. Radiomics: Extracting more information from medical images using advanced feature analysis, Eur J Cancer. 2012 March ; 48(4): 441–446.

[8] http://www.aquilab.com/radiomics-available-in-artiview-imaging-software.html?lang=en

[9] https://www.healthmyne.com/platform/

[10] https://www.research.ibm.com/haifa/dept/imt/mia.shtml

[11] http://www.mediantechnologies.com/press-release/collaborative-agreement-between-median-technologies-and-microsoft/

[12]  https://www.forbes.com/sites/bernardmarr/2017/01/20/first-fda-approval-for-clinical-cloud-based-deep-learning-inhealthcare

[13] http://www.icadmed.com/icad-receives-fda-approval-for-powerlook-tomo-detection.html

[14] http://www.prnewswire.com/news-releases/deep-learning-in-medical-imaging-a-300m-market-by-2021-300408645.html

[15] http://www.transparencymarketresearch.com/imaging-biomarkers-market.html