The customer, a bioengineering company, is developing a special tool intended to fix broken bones of patients that are in critical conditions, e.g. are in coma or are heavily injured. The primary use of the tool is intended for is tibia bone. The tool is operated by a medical staff, such as doctor or medic, who injects titanium pins into the bone and connect the pins with the tool to a static position. The pins have various shapes for different uses. The tool is to be used during transportation, surgery or other actions necessary to preserve patient’s life while in critical condition. During this time, doctors usually have limited space to take care of broken bones.

One of the main challenges is to avoid further destruction of the already broken bone during injection of the titanium pins. The destruction of the bone can be in practice determined by shear and contact stresses on the surface of the bone caused by injection of the pin. Usually this stress is determined by running finite element method (FEM) analysis in specialized engineering simulation software. FEM analysis relies on a good quality mesh of the bone geometry, for which a domain expert is needed, and is usually time consuming (the simulation can take long time). In critical situations where the tool should be used, this is hard to achieve. The challenge then becomes to enable medical staff to use the tool and know how deep they can inject into the bone without causing unwanted damage. Also if the pin geometry changes, the simulation has to be re-run. In case that doctor needs to choose from different pin geometries, this can be time consuming.

Our solution provides a bridge between the fixator tool and the medical staff. The FEM analysis can be replaced by a physically informed neural network. The network can be run on a remote server equipped with a powerful GPU to wait for data from fixator tool. Fixator pins are now equipped with a sensor, that sends data to a remotely hosted network and if the pressure caused by the tool is too high, the tool will beep to let the staff know that the pin is deep enough. Thanks to this solution, the staff does not have to know anything about physics. No domain expert is needed. Also no mesh is necessary, which would be hard to receive. Thanks to the parametrization of geometry, only one model is necessary for various pin geometries, saving the simulation time. Last, but not least, the inference of the network runs in milliseconds, instead of minutes as in case of FEM analysis. Thanks to this the tool is applicable in real life scenarios.

Outcomes Compared to the Solution using FEM Analysis:

• The solution is aproximatelly 1000x times faster on inference

• No domain expert is needed

• No bone mesh is needed for simulation

• Cost of the solution in inference reduced by 98% (inference in milliseconds for NN)

• Single model for different parameters, such as pin types, bone shape and physical parameters

Want to use Siml.ai for your next project, but don't know where to start? Make sure to check out our documentation, get up-to-speed by going through videos in Siml.ai University, and make sure to try the app for free. No installation, special hardware, or credit card is required.