Ultrasound-Guided Lumbar Puncture with AI-Based Image Labeling for Intracranial Pressure Assessment in Spaceflight

11/27/2024

As space exploration advances, concerns arise regarding astronauts' health, including Spaceflight-Associated Neuro-Ocular Syndrome (SANS), a condition that affects vision and may be related to increased intracranial pressure during prolonged exposure to microgravity. To mitigate the risks of performing a lumbar puncture in space, a gold-standard procedure for measuring intracranial pressure, an innovative technique that utilizes ultrasound images obtained with a portable probe, a needle-guidance system, and convolutional neural networks (CNNs) to identify anatomical structures is proposed . Images of the lumbar spine from a phantom and volunteers were collected, with 2,231 images manually annotated to train the CNN architectures. U-Net and DeepLab networks, with ResNet-50 and ResNet-101, were trained and tested, and various tuning and data augmentation techniques were applied to refine the models. A Flutter application was developed to connect the model to the probe's images, with plans for direct software integration. The DeepLab model with ResNet-50 achieved the best results, with average and median Dice coefficients, accuracy, and precision of 0.8051, 0.8020, 0.9760, and 0.8274, respectively, in validation, and 0.7072, 0.7286, 0.9768, and 0.7345 in testing. Despite variations in total time due to external factors such as internet connection, some tests recorded a total time of 65.05 ms, indicating that the application is viable for real-time use, helping to measure intracranial pressure safely in space.

Authors in the community:

• 

  B

  Beatriz da Silva Pinheiro Gomes Saragoça

  ist195920

Supervisors of this institution:

• 

  Z

  Zita Martins

  ist31684

industrial-biotechnology - Industrial Biotechnology

eng - English

Instituto Superior Técnico