Machine learning prediction of emesis and gastrointestinal state in ferrets

Abstract

Although electrogastrography (EGG) could be a critical tool in the diagnosis of patients with gastrointestinal (GI) disease, it remains under-utilized. The lack of spatial and temporal resolution using current EGG methods presents a significant roadblock to more widespread usage. Human and preclinical studies have shown that GI myoelectric electrodes can record signals containing significantly more information than can be derived from abdominal surface electrodes. The current study sought to assess the efficacy of multi-electrode arrays, surgically implanted on the serosal surface of the GI tract, from gastric fundus-to-duodenum, in recording myoelectric signals. It also examines the potential for machine learning algorithms to predict functional states, such as retching and emesis, from GI signal features. Studies were performed using ferrets, a gold standard model for emesis testing. Our results include simultaneous recordings from up to six GI recording sites in both anesthetized and chronically implanted free-moving ferrets. Testing conditions to produce different gastric states included gastric distension, intragastric infusion of emetine (a prototypical emetic agent), and feeding. Despite the observed variability in GI signals, machine learning algorithms, including k-nearest neighbors and support vector machines, were able to detect the state of the stomach with high overall accuracy (>75%). The present study is the first demonstration of machine learning algorithms to detect the physiological state of the stomach and onset of retching, which could provide a methodology to diagnose GI diseases and symptoms such as nausea and vomiting.

Conflict of interest statement

BM, JO, LW, GC are employees of Micro-Leads Inc. This does not alter our adherence to PLOS ONE policies on sharing data and materials.

Figures

Fig 1. Placement of gastrointestinal recording electrodes.

A) Micro-Leads planar electrodes, 1 to 6, were sutured to the stomach and duodenum of adult male ferrets. Planar electrodes are shown in inset images, which contain four contacts. B) A representative surgical placement of gastric electrodes 1 to 4 (top right), and the diagram shows how electrode position was determined (results in Table 1). C) GI myoelectric signals displaying dominant frequency are highlighted for paddle averaged signals (green) and bipolar differenced (blue).

Fig 2. Summary of gastric perturbation trials applied across all acute preparations.

Fig 3. Baseline electrogastrogram (GI myoelectric) in anesthetized ferret.

A) Example of a 20 min filtered and downsampled GI myoelectric recorded from segment 1 of an anesthetized ferret at baseline. B) Waterfall plot of the power spectral density for the waveform shown in A. Each time window corresponds to the FFT of 1 min of GI myoelectric data. C) Percentage of total power in the 6–15 cpm range partitioned by bradygastric (6.4–8.4 cpm), normogastric (8.4–10.4 cpm) and tachygastric (10.4–12.4 cpm) ranges for a signal with a DF of 9.4 cpm at baseline.

Fig 4. Effect of gastric distension on GI myoelectric.

A) Example of a 19 min filtered and downsampled GI myoelectric recorded from segment 3 of an anesthetized ferret at during gastric distension at 20 ml. Distension is maintained for 5 min (green shaded area) between infusion start and end (solid and dashed line). B) Waterfall plot of the power spectral density for the waveform shown in A. Each time window corresponds to the FFT (fast fourier transform) of 1 min of GI myoelectric data. C) Percentage of total power in the 6–15 cpm range partitioned by bradygastric (6.1–8.1 cpm), normogastric (8.1–10.1 cpm) and tachygastric (10.1–12.1 cpm) ranges for a signal with a DF of 9.1 cpm at baseline versus during distension.

Fig 5. Effect of gastric volume on DF and Pnorm.

A) Heatmap of the change in DF and B) percentage of total power in the normogastric range in response to gastric distension at 5, 10 and 20 ml (columns) across subjects and gastric segments for animals B, C, D, E, H. Grating represents gastric segments that did not display a DF at baseline or trials that were not administered.

Fig 6. Change in DF and Pnorm after intra-gastric emetine.

A) Heatmap of the change in DF and B) percentage of total power in the normogastric range after emetine infusion across subjects and gastric segments for animals 14–18, 16–18, 13–18 and 15–18. Grating represents gastric segments that did not display a DF at baseline or trials that were not administered.

Fig 7. GI myoelectric activity recorded from segment 3 during emetine infusion.

A) Example of a 70 min filtered and downsampled GI myoelectric recorded from segment 3 of an anesthetized ferret during emetine infusion trials. Emetine induced retch (yellow line) is observed approximately 28 min post emetine infusion (blue line). B) Waterfall plot of the power spectral density for the waveform shown in A. Each time window corresponds to the FFT of 1 min of GI myoelectric data. C) Percentage of total power in the 6–15 cpm range partitioned by bradygastric (6–8 cpm), normogastric (8–10 cpm) and tachygastric (10–12 cpm) ranges for a signal with a DF of 9 cpm at baseline versus 20 min prior to the first emetine induced retch.

Fig 8. Gastric myoelectric activity during food consumption.

A) Example of a 60 min filtered and downsampled GI myoelectric recorded from segment 4 of an awake behaving ferret during a feeding trial. Solid and dashed red lines denote when food was presented and withdrawn. B) Waterfall plot of the power spectral density for the waveform shown in A. Each time window corresponds to the FFT of 1 min of GI myoelectric data. C) Percentage of total power in the 6–15 cpm range partitioned by bradygastric (6.6–8.6 cpm), normogastric (8.6–10.6 cpm) and tachygastric (10.6–12.6 cpm) ranges for a signal with a DF of 9.6 cpm at baseline versus during food presentation and consumption.

Fig 9. GI myoelectric activity recorded during emetine infusion in an awake behaving ferret.

A) Example of a 90 min filtered and downsampled GI myoelectric recorded from segment 1 of an awake behaving ferret during an emetine infusion trial. Emetine induced retch (yellow line) was observed approximately 27 min post infusion (blue line). B) Waterfall plot of the power spectral density for the waveform shown in A leading up to the first retch. Each time window corresponds to the FFT of 1 min of GI myoelectric data. C) Percentage of total power in the 6–15 cpm range partitioned by bradygastric (7.3–9.3 cpm), normogastric (9.3–11.3 cpm) and tachygastric (11.3–13.3 cpm) ranges for a signal with a DF of 10.3 cpm at baseline versus pre-retch.

Fig 10. Change in DF and Pnorm during gastric distension.

A) Heatmap of the change in DF and B) percentage of total power in the normogastric range across multiple days of testing for subjects 37–18, 40–18 and 48–18 (columns).