Esophageal virtual disease landscape for disease pathogenesis and diagnostics using mechanics-informed machine learning

The mechanical behavior of the esophagus plays a major role in its physiology. But most diagnostic devices cannot estimate this mechanical behavior directly. We present a method that combines fluid mechanics and machine learning to identify and differentiate different esophageal disorders and maps them onto a parameter space which we call the virtual disease landscape (VDL). We have developed a one-dimensional mechanics-based inverse model that predicts the mechanical "health" of the esophagus through mechanics-based parameters such as esophageal wall properties and contractile behavior using the output from an esophageal diagnostic device called EndoFLIP. These predicted mechanical parameters are then used to train a variational autoencoder (VAE) that generates a latent space where different esophageal disorders get clustered into different regions and forms the VDL. This network also has a side network to predict the work done by the esophagus during bolus transport which also acts as a metric for differentiating various esophageal disorders. The VDL helps to quantify the extent of an esophageal disorder as well as disease progression in time. Thus, the VDL can potentially be used to guide treatment procedures and quantify their effectiveness.