ASN's Mission

ASN leads the fight to prevent, treat, and cure kidney diseases throughout the world by educating health professionals and scientists, advancing research and innovation, communicating new knowledge, and advocating for the highest quality care for patients.

learn more

Contact ASN

1401 H St, NW, Ste 900, Washington, DC 20005

email@asn-online.org

202-640-4660

The Latest on Twitter

Kidney Week

Abstract: PO0908

Artificial Intelligence-Driven System to Automatically Identify Arterial Oxygen Saturation Saw-Tooth Pattern in Hemodialysis

Session Information

Category: Dialysis

  • 701 Dialysis: Hemodialysis and Frequent Dialysis

Authors

  • Zhang, Hanjie, Renal Research Institute, New York, New York, United States
  • Kotanko, Peter, Renal Research Institute, New York, New York, United States
Background

Sleep apnea (SA) is a condition where normal respiration is disrupted by episodes of apnea because of disturbed respiratory control (central SA) or upper airway obstruction (obstructive SA). Arterial oxygen saturation (SaO2) saw-tooth pattern indicate respiratory instability. We aimed to automatically identify patients with repetitive episodes of intermittent SaO2 saw-tooth pattern.

Methods

The analysis was based on SaO2 recordings taken at a frequency of 0.1 by the Crit-Line® device (Fresenius Medical Care, Waltham, MA). Segments of 30 consecutive SaO2 recordings (i.e., 5 minutes of SaO2 time series) were adjudicated and categorized as (a) no saw-tooth pattern; (b) mild saw-tooth pattern; and (c) severe saw-tooth pattern (examples shown in Figure 1). We used one-dimensional convolutional neural networks (1D-CNN) for time series classification. We randomly assigned SaO2 time series segments to training (80%) and validation (20%) sets, respectively.

Results

We analyzed 89 hemodialysis (HD) treatments with 4,075 adjudicated SaO2 time series segments. Their distribution across the 3 categories was 78% (a), 11% (b), and 11% (c), respectively. In the validation data set of 815 SaO2 time series segments, we achieved an accuracy of 93.9%, 95.8% of category (a), 91.2% of category (b) and 82.8% of category (c) pattern were classified correctly by our 1D-CNN.

Conclusion

Our 1D-CNN algorithm accurately classifies saw-tooth pattern in SaO2 time series recorded in HD patients. The SaO2 pattern classification could be performed in real time during an ongoing HD treatment and provide timely alert in the event of respiratory instability.

Panel A: Intradialytic SaO2 saw-tooth pattern. Panel B to D: SaO2 time series with no saw-tooth pattern (B); mild saw-tooth pattern (C); severe saw-tooth pattern (D)

Funding

  • Commercial Support