Lumbar Puncture Project
Overview
In febrile infants younger than 30 days, lumbar puncture (LP) is a procedure routinely performed to evaluate for meningitis. LPs are mainly performed in the emergency setting by clinicians and trainees. However, novice success rates are historically poor with over 60% failure rates that can lead to diagnostic uncertainty, prolonged pain, and unnecessary resource utilization. Reduction of unsuccessful and traumatic LPs in infants can improve diagnostic ability and reduce patient harm. Ultrasound performed at the point-of-care has the potential to increase LP success rates through improved visualization of the anatomy, however it is dependent on the skill of the operator to interpret findings accurately thereby limiting it’s efficacy in the population of providers that most needs it.
The main purpose of this project is to use a pre-existing ultrasound database of ultrasound spinal anatomy videos to develop an artificially intelligent algorithm that can identify the important anatomic structures for planning an infant lumbar puncture procedure. The specific aim is to annotate spinal anatomy in a corpus of ultrasound data to train a deep learning algorithm and test accuracy of algorithmic feature recognition against expert labels in a hold-out set.