AI with a purpose – efficient PACS workflows

Artificial intelligence has arrived in radiological diagnostics. However, it will only become truly productive when the old and new worlds of radiology agree on the language of the standards. This is the prerequisite for mapping the entire examination workflow in PACS in the future.

After some two decades in the PACS business, Mark Rawanschad can definitely be described as an old hand. When a young AI company from Vienna asked him if he could imagine switching from a large corporation to a start-up, he immediately posed the question that was most important to him: “How do you envision integration into a PACS?” Five years on, contextflow is one of the leading providers of AI analysis of thoracic CT scans and, in 2023, was the only Austrian company to receive funding from the European Innovation Council Accelerator Program. And Mark Rawanschad, in his role as Business Development Manager, is responsible for further expanding the customer base, which currently comprises just over 40 institutions.

One in five hospitals uses AI for diagnosis

The AI market as a whole is growing incredibly fast. A study published in the summer in the German medical journal, the “Ärzteblatt”, considered the beginning of a “new phase from experimental development to productive application” to be a distinct possibility. The numbers back up this assessment: In July 2025, the US Food and Drug Administration (FDA) listed 1,247 approved AI-based medical devices on its website - over 1,000 of them in the field of radiology. And according to a recent survey conducted by the digital association Bitkom in collaboration with the Hartmannbund among more than 600 medical professionals in Germany, AI is used to support diagnosis in almost one in seven practices and almost one in five hospitals.

The fact that ADVANCE Chest CT employs the contextflow AI solution, often in combination with our JiveX Enterprise PACS is due to the aforementioned integration capability of the systems. As one of the few PACS systems on the market, JiveX can display the AI results as DICOM SR objects, which are sent by contextflow, in the images to be evaluated. Radiologists can adjust the display to suit their needs and navigate directly to abnormalities. “In addition, JiveX PACS is able to send us the appropriate CT series - which is very useful because it filters out "noise" on the PACS side and reduces data traffic,” explains Mark Rawanschad.

Mark Rawanschad is also well aware that it can be a mental challenge for some start-ups to embrace the world of DICOM or IHE. "However, as AI providers, we should not forget that radiology has already been using digital workflows for decades. Reinventing the wheel would be the wrong approach."

Fast and seamless integration into JiveX Enterprise PACS

In addition to the quality of AI analysis, experts believe that there are three key criteria that make up an optimal AI workflow. Firstly, there is the question of speed, as Mark Rawanschad explains: “If it took us 20 minutes to evaluate a lung CT scan, we would be useless.” Secondly, user acceptance increases significantly when users can evaluate the AI results in their familiar user interface.
Simon Andrzejewski, Product Manager for JiveX Enterprise PACS, emphasizes: “Breaks in media cost time and nerves, and our mission is to make the workflow as seamless as possible for physicians.” The fact that the latest version of JiveX also validates AI in JiveX and ensures that the radiologist retains his/her decision-making role is a logical further development of this physician-centered approach.

VISUS is about to implement the third criterion: The implementation of a feedback loop which allows the results from the validation process of the AI results to be communicated in a standardized, interoperable manner. This allows providers to use them to check whether the quality of results changes contrary to the expected outcome - an important part of the EU AI Act's requirements for post-market surveillance. The integration of the so-called AIRA profile in JiveX PACS provides the basis for this

The before and after: en route to a “PACS-driven workflow”

However, the truth also includes: The potential of AI for the radiological workflow can only be fully exploited if the stages before and after the examination are also mapped. Mark Rawanschad has a clear personal opinion on this: “For me, the findings belong in PACS – namely in a structured form that automatically integrates AI results.” Our Head of Innovation, Dr. Marc Kämmerer, goes one step further: “Ideally, a ‘PACS-driven workflow’ should also provide the physician with relevant contextual information about a patient, e.g., aggregated by AI.”

However, the biggest structural lever for bringing AI to the mainstream concerns financing instruments. One alternative is to use subscription models, such as offered by the AI market place from connectMT which currently includes some 30 validated AI solutions and saves on scarce IT resources. “However, with pay-per-use procedures, one should not be too restrictive in the number of images transmitted so that the AI can play out its decisive advantage as a safety net,” emphasizes Mark Rawanschad.

AI as a health insurance benefit?

One option that has not yet been widely tested is selective contracts, which are currently being tested by Healthy Hub, an association of smaller German health insurance companies. For example, participating practices receive 25 euros per examination for using contextflow's Malignancy Score, an additional feature that indicates how likely a pulmonary nodule is to be benign or malignant. In the best case scenario, this results in a win-win-win situation: The investment pays for itself within a short period of time for the practice, the health insurance companies save money due to optimally adjusted examination intervals, and patients receive certainty sooner.

Of course, it would be even better if the costs were covered by health insurance. Annual lung cancer screening for heavy smokers, which Germany plans to introduce in April 2026, could become an exciting pioneering application for this. The high customer demand recorded by contextflow suggests that Germany's radiologists expect the system to be a success. After all, the more people who use the service, the more they could benefit from support - especially if it is seamlessly integrated into their examination process.
 

About contextflow

contextflow is a spin-off of the Medical University of Vienna, the Vienna University of Technology, and the European research project Khresmoi. They have been developing algorithms that use deep learning models to support diagnosis and quantification in radiological images since 2016.