Improved Detection of Gastroesophageal Reflux Disease Through Advanced Machine Learning Architecture

Strategic Insights -

Researchers at the Isfahan University of Medical Sciences have made significant strides in the diagnosis and treatment of gastroesophageal reflux disease (GERD) using a novel machine learning architecture. By leveraging a U-Net-inspired design, which combines both 2D and 1D convolutional neural networks (CNNs), the team has developed an efficient and accurate method for detecting GER events in patients.

Key Takeaways:

  • The proposed architecture consists of a semi-U-Net structure, where the 2D CNN serves as the first encoder to capture features across all channels, followed by 1D CNNs to preserve the 1D nature of the signal while minimizing the number of parameters.
  • The model achieves a sensitivity of 95.24% and a positive predictive value of 100%, outperforming existing methods in GER event segmentation.
  • The average duration of the detected GER events is 17.52 ± 6.39 s, and the model demonstrates strong generalizability across diverse GER events.
  • The proposed architecture is compact, efficiently utilizing parameters, and offers robust robustness and adaptability to varying input durations.
  • The research aims to enhance the utility of 24-h Multichannel Intraluminal Impedance (MII) pH monitoring, enabling clinicians to make better-informed decisions for patient selection in invasive surgical procedures.
  • The study used a dataset of 202 episodes containing 208 GER events collected from 26 patients who underwent 24-h MII pH monitoring.

Statistics:

  • Sensitivity: 95.24%
  • Positive predictive value: 100%
  • Average duration of detected GER events: 17.52 ± 6.39 s
  • Number of episodes: 202
  • Number of GER events: 208
  • Number of patients: 26
  • Parameter efficiency: Low parameter count offers robust generalizability

Sources:

  • Segmentation of gastroesophageal reflux events using a semi-U-Net architecture with 1D/2D CNNs. Scientific Reports, 2025;15(1):37152. Nature Portfolio, Heidelberger Platz 3, Berlin, 14197, Germany.
  • Isfahan University of Medical Sciences
  • Medical Image and Signal Processing Research Center
  • School of Advanced Technologies in Medicine
  • Isfahan University of Medical Sciences, Isfahan, Iran