Every day, hospitals generate a lot of medical data: admission records, hospitalization reports, analysis results, imaging reports, etc. This data is a very rich database for research and can be used to develop digital tools for optimizing patient flows, robotic surgery and diagnostic tools. They constitute a very rich database for research and can be used as a basis for the development of digital tools, from the optimization of patient flows to robotic surgery and diagnostic tools. However, this database is not yet widely exploited because, to make their use possible, the data must be recognizable and annotated in a non-noisy way with standardized keywords. This is even more complex for medical imaging data, which today requires annotation by specialists who can understand and interpret the images. Alcimed comes back on the current difficulties encountered for the valorisation of medical images as well as on the conditions to respect for the implementation of medical image annotation tools by artificial intelligence (AI).
The current difficulties encountered for the valorisation of medical images
The significant mobilization of physician time required
Current hospital tools (PACS and DICOMS) do not allow instant identification of desired image types. Thus, when setting up a cohort requiring medical images, the selection is done manually by physicians. This is a tedious step that is essential for the launch of research projects after the fact and takes on average between 3 and 5 months. The time spent by doctors on this stage has a direct economic impact on the hospital.
Heterogeneity of medical image annotation practices and errors that limit the a posteriori and a priori use of these data for medical research
Today, there is no standard medcial image annotation. The result is a very heterogeneous treatment depending on the keywords and the level of detail filled in by the doctors. It is estimated that the rate of data errors and omissions in the cohorts constructed is 30% and, beyond the associated difficulties of valuation, each error or omission impacts the quality of care by preventing the integration of a patient into a research program.
Automating this selection and medical image annotation process with an artificial intelligence solution would accelerate the creation of cohorts and promote the development of research projects based on the use of medical images.
Learn more about the integration of artificial intelligence in healthcare >
The need to implement digitized annotation tools that must meet several key conditions
Storage adapted to the characteristics of medical images
Medical images are heavy files since an examination can produce up to 80 images whereas a blood analysis will generate a document of less than 5 pages, this can be accentuated by a factor of 10 to 100 depending on the type of images and their resolution. The storage structure must therefore have an appropriate capacity while respecting the cybersecurity criteria inherent in the storage of health data.
Integration with existing tools
To limit the number of interfaces and facilitate integration with the practices of administrative and medical teams, the annotation tool must be able to be integrated with existing hospital tools such as health data warehouse projects or PACS/DICOM of the institutions.
A flexible medical image annotation tool
The tool and the proposed annotations will have to be able to adapt to the specialties concerned, to the habits of the institutions and even to the specific research project for which the images are annotated.
The potential of medical image enhancement is currently hampered by the doctor time required and by the errors that result from the heterogeneity of annotation practices. In order to optimize the valorisation of medical images, AI is a solution for the automation and standardization of their processing. Our team at Alcimed can help you to explore the ways of valorization of your health data, in particular through innovation strategy and the realization of commercial strategy!
About the author,
Elody, Consultant in Alcimed’s Innovation and Public Policy team in France