AI-powered tractography is a cutting-edge technique that leverages artificial intelligence to enhance the visualization of nerve pathways in the brain. This method is particularly useful for planning complex neurosurgical procedures.

Tractography is an imaging technique that calculates the course of nerve pathways (also known as nerve fibers or tracts) based on specialized MRI scans. These pathways are crucial for various brain functions, including movement, speech, and thought. Traditional tractography methods rely on mathematical models to infer the location of these pathways from MRI data. However, these methods often involve uncertainties, especially when the brain has been altered due to disease or surgery.

Tractography, the process of reconstructing streamlines that represent neural fiber pathways within the human brain from diffusion MRI (dMRI), has gained significant attention in recent years. It is primarily used for scientific studies and surgical planning. TractSeg is a widely used example of this, automatically reconstructing specific fiber bundles with high precision.

Modern AI methods, such as machine learning, can recognize patterns in MRI data and generate more accurate reconstructions of nerve pathways. One widely used AI method is called TractSeg, which was originally trained on healthy brains. Researchers have tested whether TractSeg can also work for epilepsy patients who have undergone a hemispherotomy – a surgical procedure that disconnects the two hemispheres of the brain.

While TractSeg performed well in many cases, it also produced unexpected errors, such as reconstructing nerve pathways that should no longer exist due to the surgery-a phenomenon known as “hallucination.” Additionally, some remaining pathways were either incompletely captured or entirely missing from the reconstruction2.

To address these issues, a research team from the Lamarr Institute and the University of Bonn has developed the new hybrid method that combines the advantages of AI with the data fidelity of traditional techniques. This approach ensures that only existing nerve connections are reconstructed, eliminating the issue of “hallucinations” where AI might reconstruct pathways that no longer exist due to surgery.

Despite its good overall generalization, TractSeg failed to reconstruct some bundles, due to disconnected bundle masks, or because the start or end regions did not overlap the bundle mask.

Researchers concluded by saying “Although the results are promising, we advise caution and manual quality control when dealing with complex and severely pathological cases. We expect that fully automated and reliable generalization to pathologies that were not seen during training will remain a challenge for the current generation of deep learning based approaches.”