Abstract: TH-PO020
Automatic Noninvasive Peritonitis Detection and Monitoring in Patients on Peritoneal Dialysis Using High-Resolution Ultrasonography and Artificial Intelligence
Session Information
- Augmented Intelligence for Prediction and Image Analysis
October 24, 2024 | Location: Exhibit Hall, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Augmented Intelligence, Digital Health, and Data Science
- 300 Augmented Intelligence, Digital Health, and Data Science
Authors
- Quero, Maria, Hospital Universitari de Bellvitge, L'Hospitalet de Llobregat, Catalunya, Spain
- Jobst, Beatrice Maria, Kriba, Barcelona, Catalunya, Spain
- Guillén Fernández-Micheltorena, Sara, Kriba, Barcelona, Catalunya, Spain
- Carandell Verdaguer, Francesc, Kriba, Barcelona, Catalunya, Spain
- Santos Abreu, Roberto Fabião, Kriba, Barcelona, Catalunya, Spain
- Jimenez, Javier, Kriba, Barcelona, Catalunya, Spain
- Quesada, Rita, Kriba, Barcelona, Catalunya, Spain
- Andujar Asensio, Alex, Hospital Universitari de Bellvitge, L'Hospitalet de Llobregat, Catalunya, Spain
- Rau, Ana Melissa, Hospital Universitari de Bellvitge, L'Hospitalet de Llobregat, Catalunya, Spain
- Molina, Yolanda, Consorci Sanitari de Terrassa, Terrassa, Catalunya, Spain
- Moreno Guzmán, Fátima, Consorci Sanitari de Terrassa, Terrassa, Catalunya, Spain
- Broseta Monzo, Jose Jesus, Hospital Clinic de Barcelona, Barcelona, Catalunya, Spain
- Sanchez Escuredo, Ana, Hospital de Sant Joan Despi, Sant Joan Despi, Catalunya, Spain
- Slon Roblero, Maria Fernanda, Complejo Hospitalario Navarra, Pamplona, Navarra, Spain
- Rama, Inés, Hospital Universitari de Bellvitge, L'Hospitalet de Llobregat, Catalunya, Spain
Background
Peritonitis is a significant complication in peritoneal dialysis (PD) due to its high incidence and associated consequences, such as technique failure and mortality. Early diagnosis and treatment are crucial to reduce complications.
We developed an automatic, quick, non-invasive leukocyte counting and differential method based on high-resolution ultrasound (HRUS) and artificial intelligence (AI).
This method could be useful for early peritonitis detection and home monitoring in PD patients to extend technique’s survival and reduce mortality.
Methods
We acquired HRUS in vitro data from polystyrene particles suspended in distilled water in drainage bags (Baxter’s and Fresenius’) at concentrations 0-4000 cells/µl. Four AI models were trained on these data: one to classify concentrations below and above 200 cells/µl; two models to predict leukocyte numbers in the respective range, and a fourth to classify between polymorphonuclear cells and lymphocytes. Patient effluent samples with >4000 cells/µl were automatically filtered by intensity. Each model was tested with 3 measurements on each of the 62 samples from 6 patients with confirmed peritonitis. Results were compared to laboratory measurements (LM).
Results
The counting models yielded an R2 score of 0.94 (0-4000 cells/µl). For LM<200, containing the diagnostic threshold of 100 cells/µl, the mean error over patient samples was 18.1±22.0 cells/µl. 18/20 measurements >4000 cells/µl were correctly filtered. The remaining 2 were predicted with >2700 cells/µl. For classification below and above 100 cells/µl, we obtained a sensitivity/specificity (SE/SP) of 95.1/94.7%, respectively. Additionally, the cell-type classification model achieved an accuracy of 90.2% (SE/SP: 86.5/92.5%).
Conclusion
These results demonstrate the potential of this non-invasive AI and HRUS based leukocyte counting and differential system for early detection and objective home monitoring of peritonitis in PD patients.
Funding
- Commercial Support – Kriba