Sterile supplies: AI-based recognition of surgical instruments with Cir.Log

The Cir.Log project between Charité Facility Management GmbH (CFM) and Fraunhofer IPK aims to increase the efficiency of sterile processing in hospitals and improve patient safety. Essentially, the aim is to create a smooth transition from current workflows to a more efficient, technology-supported system.

The project focusses on the reprocessing unit for medical devices to significantly increase patient safety.

How can processes be optimized and sources of human error minimized? Two questions that seem to be omnipresent, and not just in a medical context. With the advent of artificial intelligence in multiple fields, there are new ways to answer these questions. The idea: to integrate a technology that makes it possible to recognize and track instruments using image processing.

"CFM became aware of Fraunhofer IPK via TU Berlin," explains Ümit Ejder from Charité Facility Management GmbH. "At the beginning, it was important to get to know each other and discuss all the possibilities. On the one hand, what the technology can do, and on the other, how the developers can optimally integrate the technology into the process."

"This was undoubtedly one of the biggest challenges," explains Jan Lehr from Fraunhofer IPK. "Our institute, the Fraunhofer Institute for Production Systems and Design Technology IPK, had mainly gained its experience in an industrial context. The main use case is spare parts logistics. If a machine breaks down, the defective part can be photographed. AI can then determine the exact item in question in a matter of seconds and initiate an ordering process. The question of whether this machine learning method could also work for surgical instruments led to the decision to simply try it out, starting with a feasibility study."

Compared to the conventional method of instrument tracking, the new technology is said to have several advantages. "You have to imagine it like this," explains Ümit Ejder, "the instruments used during surgery have to be cleaned, disinfected, sterilized and then made available to the users again. That roughly describes our task. After cleaning and disinfection, the medical devices are returned to the clean area, separate from the unclean area. In the clean area, the medical devices are sorted into so-called sieve baskets according to specific requirements and packing lists and prepared for surgery again. Our employees use a PC monitor at their workstation to call up the packing lists for each set and then check each instrument individually. Although this process is computerized, it is still carried out by people. Complaints cannot be ruled out in this area, even though we have already optimized the process."

This is where Cir.Log, the Fraunhofer Institute's system, comes into play. It acts like an extra pair of eyes, checking to see if something has been forgotten or if there is an item that does not belong there. "Initially, the process of holding the instrument under the camera and triggering the detection was extremely time-consuming and anything but practical, taking around five seconds per instrument," Jan Lehr reports. "To speed up this process, we have further developed our AI so that it does not only recognize a single instrument per image, but can process several at the same time. This increases the speed many times over, as more than a dozen instruments can now be detected at once. Feedback from users onsite showed us that there is a desire to be able to capture instruments directly, without having to lay them out. Our aim now is for the AI to be able to track the instruments seamlessly right through to the sieve."

Ümit Ejder adds: "At the moment, it is not yet possible to hold the already packed sieve under the camera and use a photo to recognize all the instruments. The challenge lies primarily with the basic sieves, as the instruments are packed close together and sometimes overlap. Accordingly, the camera would only receive a small section of the instruments – currently too little for recognition."

The development and reliability testing of the system are crucial steps in the process. "Threshold values are set and empirical values are used to ensure the safety of the system," explains Jan Lehr. "Above a certain safety level, the system appears very reliable to users." To this end, a field test was started with a prototype in the reprocessing unit at the Benjamin Franklin Campus of Charité. Employees use it for their daily work and provide feedback on its reliability.

The system is also continuously learning. Every photo and every instrument recognition contributes to improvement. All data is stored and the system is regularly trained with images from the packing process. This is an essential step in further increasing the precision of the system.

Continuous data collection and analysis has enabled the AI to constantly learn and improve its capabilities. In addition, special attention was paid to user-friendliness to ensure that users on site could use the system smoothly and effectively.

"It is important to understand that this is an assistance system that is not intended to replace employees or make their work superfluous," emphasizes Ümit Ejder. "Ultimately, our employees always retain the final view and decision-making power. We do not simply leave the decision to an AI."

The future development of the system aims to further refine and make the technology more compact to enable a broader market launch. A beta version is planned, which will be tested in pilot phases in the near future in order to gather feedback from users and optimize the system accordingly.

In spring 2025, Cir.Log will then be offered on the market by a Fraunhofer IPK spin-off. In the long term, further areas of application such as quality control and automation at the incoming goods department of the preparation unit could be considered in order to further improve the efficiency and safety of the preparation process.