AI Model Learns to Pinpoint Tumours in Genetic Disorder

Patients with the genetic condition neurofibromatosis 2 (NF2) develop tumours, known as vestibular schwannomas, on the nerves affecting hearing and balance, often on both sides of the head. Constant monitoring requires numerous MRI scans, making manual measurement a significant task.

Addressing this challenge, a new computer vision model has been developed to automate this process. Researchers trained the AI using a combination of institutional and public MRI data to teach it how to precisely outline and measure the volume of these tumours.

The model achieved high accuracy, with a mean Dice score of 0.87 on the complex bilateral scans. It proved particularly adept at distinguishing the target tumours from other adjacent lesions, a frequent complication for NF2 patients. While it successfully detected even small tumours, the model struggled when a tumour consisted of separate, non-contiguous segments. This study marks a crucial first step toward faster, automated analysis for this condition.