At a time when everyone in medical imaging only talks about artificial intelligence, our only goal is to make it invisible.
The convergence of reliable artificial intelligence (AI) algorithms and the growing body of image data have created fertile ground for its use in solving the problems facing radiologists and clinicians. Press articles and communication around this new technology often focus on the technical side, highlighting the accuracy and reliability of the algorithms. But in their daily practice, radiologists are actually much more concerned with the quality of care they provide to patients and the clinical value of the tools they use. AI is likely to have a significant impact on these goals, but achieving these goals requires a different approach centered around integration into clinical workflow.
Artificial intelligence in the medical imaging sector remains at an early stage, in the stammering of its technical development and practical application, there are still many regulatory and commercial obstacles that prevent the widespread adoption of this technology.
Our current thinking is focused on the next step that will be decisive in the adoption of AI by the world of medical imaging, integration into a clinical workflow that adapts to the needs of users. It's not just about using this technology and displaying results in interfaces; it's about rethinking and adapting workflows to the ways in which AI interacts with exam data and provides information necessary for diagnosis. Bringing data from algorithms in a non-intrusive way at the right time, displaying them in the right way, without leaving the interpretation workflow requires expertise and consideration of the clinical context but also of the clinicians themselves. Due to our close collaboration with the clinical world, we can better understand these needs. The development of our workflows is the result of years of reflection and improvements stemming from feedback from the field and from the daily use of our tools by radiologists.
AI algorithms in medical imaging, are created to target certain specific organs such as the brain, heart, lungs, liver ... or pathologies such as neurodegenerative diseases, cancers ... or according to the acquisition methods (CT, MRI, X-rays…). While a cardiac CT exam will require access to a set of hyper-specialized algorithms, a more general exam will require fewer specific tools. This difference in needs places AI in the field of computer system interoperability. The results must be made available to users in a unified manner but with an understanding and adaptation to the requirements and needs of the clinician. This is in line with the vision of Intrasense which develops specific workflows adapted to the anatomical regions studied or pathologies, so our main objective is to provide the radiologist and the clinician with the right image, at the right time, accompanied by the most relevant tools.
The other essential element for clinical integration of AI concerns the automatic learning of algorithms which become more efficient, competent and precise thanks to the data they integrate but also thanks to expert comments and tests during the time of their development. Transparency in this process must be important and the clinician must be able to remain in control of the results through the functions of accepting, rejecting or even correcting diagnostic information. This ability to act directly on the diagnostic capabilities of AI allows the clinician to control the process, which builds confidence and acceptance of AI as a diagnostic tool.
Intrasense has worked for many years in partnership with clinicians, radiologists and many imaging departments in leading hospitals. The arrival of AI encourages us to strengthen these partnerships and to work in close collaboration with the clinical world. We are very attentive to the specific needs that are expressed within the framework of the daily clinical workflow of our partners. We apply this principle to AI which becomes completely transparent and integrated through standardized interfaces which do not require any adaptation in the way of working of users, it is up to the technology to adapt to the needs of radiologists and clinicians. In addition, AI tools are not always useful at work of the radiologist, which is why it is essential to maintain flexibility in its use with standardized processes.
Once the information has been consulted and examined, the report stage must also leave flexibility to the user with the possibility of including or not the results from the algorithms according to their relevance or according to the recipient of the report. Our structured reports are an integral part of our workflow and are perfectly adapted according to the examinations performed or the organs examined, they are automatically generated according to the choices made by the radiologist or the clinician.
Our choice is to integrate AI transparently for the user, in clinical tools and in their workflow, to put AI into clinical routine. For us, integration is an essential step on the way to the adoption of this new technology which allows real productivity gains, significant clinical benefits as well as increased performance, without this integration, AI will not be able to deliver all its potential.