A Deep Learning-based Assessment Pipeline for Peripherally Inserted Central Catheters (PICCs) on Pediatric Radiographs - Project summary: Peripherally inserted central catheters (PICCs) have become an integral part of the complex intensive care of critically ill patients in pediatric intensive care units (ICUs). Unfortunately, PICC malposition occurs in approximately 3.6%-5.6% of pediatric patients. It is critically important to frequently assess the position of PICCs, because malpositioned PICCs can lead to a variety of clinical complications, such as thrombosis, cardiac arrhythmia, and inadequate therapy. Chest X-rays (CXRs) are widely used radiographic images for the periodic assessment of PICCs in children. However, assessing PICCs is a labor-intensive, time-intensive task for radiologists. Prior studies have demonstrated the promise of deep learning for automated PICC assessment in adult patients. To date, there are no available methods for pediatric PICC assessment. An accurate, robust, and clinically available tool that can automatically assess PICCs in pediatric CXRs is a critical unmet need. The objective of this proposal is to develop an automated PICC assessment pipeline for pediatric CXRs using deep learning algorithms. To accomplish our objective, we will leverage ~4,000 retrospectively collected CXRs from pediatric ICUs. In the current clinical assessment of PICCs using CXRs, radiologists would typically (1) identify each PICC line, (2) locate the ideal PICC landing zone, and (3) determine whether the distal tip of the PICC is appropriately positioned within the landing zone. By closely mimicking radiologists, we design a novel, transparent, and interpretable three-component PICC assessment pipeline. Specifically, we propose to (Aim 1) develop a topology-preserving deep learning method to segment PICCs, (Aim 2) develop a landmark detection method to locate the ideal PICC landing zone, and (Aim 3) assemble the pipeline and evaluate its performance in detecting PICC malposition using CXRs in pediatric patients. This project is highly impactful to clinical practice. Assessment of PICCs commonly occurs within a procedure called Lines, Drains, and Airways (LDAs) reconciliation, in which radiologists review LDA intensive care devices on CXRs and communicate with intensive care physicians in ICUs to ensure safe patient care. Our proposed pipeline will promote LDA reconciliation between radiologists and intensive care physicians. By taking timely remedies, physicians in ICUs can quickly reposition those malpositioned PICCs back to the ideal landing zone, thereby mitigating potential risks of major clinical complications. Once validated, our pipeline can also be extended to other LDA devices, such as endotracheal tubes or chest tubes. The resulting models will be made available to the community for both research and clinical purposes.
