Here is a tool, which is not special to brain surgery.

https://www.impactlab.com/2023/07/27/ai-tool-revolutionizes-brain-lesion-quantification-on-mri-scans/

Regards,

Joachim

Dear Joachim, Thank you very much for your quick response. The tool is very interesting for our stroke project. For patients who have undergone brain tumor surgery, I am still looking for a tool that has been trained with data from this specific patient group and can detect the specific lesions they have. Best, Jale

The most effective AI solutions for post-operative brain lesion quantification are generally found within flexible, open-source platforms that allow for the deployment of custom-trained AI models. These are often used by researchers and specialized clinical groups to develop and validate their own algorithms.

• 3D Slicer (and MONAI): This is one of the most prominent platforms. 3D Slicer is a free, open-source software for medical image analysis. Its true power for your question lies in its extensions, particularly those based on the Medical Open Network for AI (MONAI). Researchers can build and integrate deep learning models in MONAI to perform highly specific segmentation tasks, such as differentiating between a resection cavity, residual tumor, and surrounding edema on post-operative MRI scans.
• ITK-SNAP: Similar to 3D Slicer, ITK-SNAP is an open-source tool for image segmentation and visualization. While it's more focused on manual or semi-automated segmentation, its scripting capabilities and ability to integrate external algorithms make it a popular choice for custom AI model deployment.
• BraTS Challenge Models: The Brain Tumor Segmentation (BraTS) Challenge is an annual international competition that has been a major driver of innovation in this field. Many of the winning models and frameworks are open-source and specifically designed to segment tumors and other abnormalities on pre- and post-operative MRI images. While the primary goal is often tumor segmentation, the methods can be adapted to quantify the entire lesion volume, including the resection cavity. You can find these models and their code repositories on platforms like GitHub.

The Challenge for AI

The primary reason for the lack of a simple, all-in-one commercial tool is the immense variability of the post-operative brain. The appearance of the resection cavity, the amount of surrounding edema, and the presence of hemorrhage can vary wildly between patients and even within a single patient over time. A robust AI model must be able to distinguish all these different components from the residual tumor, which requires extensive and carefully annotated training data that is not always readily available.

Ultimately, the most successful approaches use custom-trained AI models built on a foundation of open-source frameworks to address the precise nuances of a specific type of post-operative brain lesion. The future lies in making these powerful, specialized models more accessible to the wider medical community.