Innovative AI Model Developed at Columbia University Enhances Infant Lumbar Puncture Procedures

Columbia University Professor of Pediatrics in Emergency Medicine, David Kessler, with the support of the CUIT Emerging Technology's AI Community of Practice grant, has spearheaded a groundbreaking project focuses on developing an artificial intelligence (AI) algorithm model to enhance the accuracy and safety of infant lumbar punctures (LP) by recognizing key anatomical features on spinal ultrasounds. This innovative work is set to transform pediatric emergency medicine and improve outcomes for some of the youngest and most vulnerable patients.

Addressing a Crucial Medical Challenge

Infant LPs are routinely performed to diagnose conditions such as meningitis in febrile infants under 30 days old. However, the procedure poses challenges due to infant spines' small size and subtle anatomical features, often resulting in high failure rates among novice practitioners. The use of ultrasound imaging at the point of care has shown the potential to increase the success rate of LPs, but its effectiveness heavily relies on the operator's skill in interpreting the images.

Kessler's project aims to mitigate these challenges by developing an AI model that can autonomously identify critical spinal features in ultrasound images, thereby aiding clinicians in performing more accurate and safer lumbar punctures.

Key Achievements and Innovations

Over the year, Kessler's team has made notable progress, achieving several milestones that underscore the potential of AI in medical procedures:

• Binary Feature Classifier: The team developed a classifier capable of identifying three key spinal features: spinal cord, spinal fluid, and image quality.
• Novel Annotation Approach: They demonstrated the feasibility of an automated feature annotation method, expediting the process of creating annotated datasets.
• Extensive Dataset Creation: A frame-annotated dataset of cross-sectional spinal ultrasound images was created, consisting of 1,515 original frames and 11,224 augmented frames.
• Real-time Deployment Strategy: A device-agnostic strategy was designed for real-time model testing in clinical environments, ensuring the model's practical applicability. 

Image: Identified AI model capabilities and rapid image annotation techniques - histogram equalization & contrast enhancement.

Future Directions

The project is poised to enter its next phase, focusing on refining the model to include object localization for specific features and testing its real-time deployment in clinical settings. Kessler's innovative approach to using AI in medical procedures exemplifies the transformative potential of emerging technologies in healthcare. The successful development of the AI model for infant lumbar punctures promises to improve procedural accuracy and patient outcomes, setting a new standard in pediatric emergency medicine.

Emerging Technology AICoP Grant Awardee

David Kessler, Professor of Pediatrics in Emergency Medicine, Vagelos College of Physicians and Surgeons

David Kessler is the Vice Chair of Innovation & Strategic Initiatives in the Department of Emergency Medicine at Columbia University. As Vice Chair of Innovation, Dr. Kessler fosters a culture of innovation and collaboration within the department to help implement novel programs and innovative solutions to complex adaptive systems problems. He is a graduate of Princeton University and Mount Sinai School of Medicine, where he completed his pediatric residency on the global health track. Dr. Kessler completed his Pediatric Emergency Medicine Fellowship and Emergency Ultrasound training at Bellevue Hospital in Manhattan, NY, and his Master of Science in Clinical Investigation at New York University. In 2011, he joined the faculty at Columbia University Irving Medical Center as the Director of Operations of Pediatric Emergency Medicine. Since then, he has helped create several novel programs, such as a curriculum for interprofessional clinical simulation and the fellowship in pediatric emergency ultrasound. 

Emerging Technologies' mission is to empower faculty, researchers, and administrative members to utilize next-generation technologies to explore, adopt and integrate new applications and strategies. Please visit our AI: Community of Practice for discussion highlights on AI topics and an interest intake form to join the monthly gatherings.