Software Prototyping in Java: Examples on Automatic Annotation and AI-based Instrument Tracking for Stroke Treatment

Kurzfassung

Software prototypes for image-guided therapy (IGT) are complex to implement. In this paper, we present an easy-to-compile Java software that communicates with existing toolkits to process tracking and image data in real time. As an application example, we used an electromagnetic tracking system to automatically annotate 815 fluoroscopic images with catheters and guidewires for stroke treatment. Building on this, we trained a you only look once (YOLO) model for image-based tracking of these instruments. In our test data, the instrument was correctly detected in 87% of the frames. We further demonstrated usage of the model by exporting it to the open neural network exchange (ONNX) format and loading it into our Java software prototype for live instrument detection.