Using open-source software libraries, Joseph and his colleague, physicist Dr Rob Finnegan, have built their own model and platform. He says training a model on the consenting data of RNSH patients that conforms to published national and international guidelines gives better results. “It is more suited to our patients here,” he says. For every 30 to 50 scans, clinicians are batching that data together and retraining the models to learn how edits have been made. “Ideally, it gets closer and closer to how we want it,” Joseph adds. Due to government regulations, they are only able to use this in-house AI as part of a clinical trial via the TGAs clinical trials notification pathway. “To go down the path of using it off trial, we would need to get further TGA approval, which is a long process that costs a lot of money,” he says. He is hoping at the end of the trial, they will release the RNSH model to the wider community for research purposes. “If other centres want to run a trial, they can use it, and it is free.” Joseph sees that AI contouring is a good tool. “We don’t think it should be used to replace the radiation oncologist doing this task, but using it will give us more time to focus on the things that matter, especially the target delineation of the actual tumour itself.”
29
Made with FlippingBook - professional solution for displaying marketing and sales documents online