Abstract: SA-PO0072
Label-Free Holographic Imaging and Deep Learning for Rapid Monitoring of Peritoneal Dialysis-Associated Peritonitis in Pediatric Patients
Session Information
- Artificial Intelligence: Kidney Imaging, Pathology, and Precision Phenotyping
October 24, 2026 | Location: Exhibit Hall A, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Artificial Intelligence, Digital Health, and Data Science
- 300 Artificial Intelligence, Digital Health, and Data Science
Authors
- Pathak, Sanskar, University of Minnesota Twin Cities, Minneapolis, Minnesota, United States
- Bravo-Frank, Nicholas Gabriel, University of Minnesota Twin Cities, Minneapolis, Minnesota, United States
- Kouri, Anne, University of Minnesota Medical School, Minneapolis, Minnesota, United States
- Hong, Jiarong, University of Minnesota Twin Cities, Minneapolis, Minnesota, United States
Background
Peritoneal dialysis (PD)-associated peritonitis remains a major barrier to safe long-term PD, particularly in pediatric patients on home-based care. Current diagnostic workflows rely on clinic- or laboratory-based effluent testing, including white blood cell (WBC) count with differential and culture, delaying early inflammatory assessment. We developed a compact, automated, and low-cost digital inline holography (DIH) and deep learning platform for label-free WBC detection, classification, and longitudinal inflammatory monitoring in human PD effluent.
Methods
PD effluent samples were imaged using a compact DIH system with microfluidic sample delivery and 10× optical magnification. DIH enables label-free imaging with an extended depth of field for volumetric cell detection in flowing effluent. A deep learning pipeline was trained using blood-derived human WBC holograms and high-confidence PD-effluent labels, with hard-negative particles reducing false detections from effluent debris. Detected cells were converted to WBC concentration, and reconstructed cell images were used for polymorphonuclear leukocyte classification. Longitudinal pediatric PD effluent samples from three patients were analyzed against clinical status using DIH-derived WBC trends.
Results
The DIH and deep learning pipeline achieved >90% WBC detection recall at a particle-level false-positive rate of ~0.1% for automated detection in complex PD effluent. Unlike conventional WBC counting workflows that rely on trained personnel or clinical laboratory analyzers, the DIH platform provides automated, label-free analysis in a mobile format designed for home-based monitoring. In longitudinal pediatric PD samples collected over 9 months, DIH-derived WBC trends distinguished clinically stable periods from episodes with elevated inflammatory burden. The platform detected WBC elevation above the clinical peritonitis threshold during a documented peritonitis episode and tracked decreasing WBC levels during clinical improvement in a symptomatic episode.
Conclusion
DIH combined with deep learning enables rapid automated WBC analysis and longitudinal inflammatory monitoring in PD effluent. This compact, low-cost approach may complement standard lab test and future point-of-care monitoring of peritonitis. Larger clinical studies will further evaluate diagnostic performance and clinical utility.
Funding
- Private Foundation Support