Doctors need precise maps of brain tumors before surgery to plan safe procedures. A large review looked at 36 different computer models designed to outline gliomas, a type of brain tumor, using MRI scans. These AI tools showed high accuracy for mapping the whole tumor. The average score was 0.860 on a standard measure of precision. This means the computer outlines matched the human expert outlines very well in most cases.
However, the tools struggled with specific parts of the tumor. They were significantly less accurate at defining the enhancing and non-enhancing areas. Using advanced MRI inputs like 3D images or multiple types of scans helped the models perform better. Training the AI on a specific public dataset also led to higher scores. Yet, simply having more training data did not automatically make the tools better.
The review found that the year a study was published predicted better accuracy. This suggests technology is improving over time. However, researchers could not find one single reason why some models worked better than others. The reasons for these differences are still unclear. While the tools show promise, more research is needed to understand exactly what makes them succeed or fail before they become standard in every hospital.