**3.1 AI in achalasia diagnostics**

AI has not explored much of achalasia diagnosis. One of the scarce examples is the work of Carlson et al. where they used functional luminal imaging probe panometry as a method to detect achalasia subtypes using ML. Manometry was performed on 180 patients with achalasia's 3 subtypes. FLIP is a technique that is used to measure distensive pressures and distension-induced esophageal contractions. Correlation analysis, single tree, and random forest were adopted to develop classification trees to identify achalasia subtypes. Their decision tree model accurately identified spastic (type III) versus nonspastic (types I and II) achalasia with 90% and 78% accuracy, respectively. The train and test cohorts correctly identified achalasia subtypes I, II, and III with 71% and 55% accuracy, respectively [78]. In a recent conference proceeding, Jiang et al. reported an automated real-time esophagus achalasia detection method for esophagoscopy assistance through the use of convolutional neural network (CNN) to detect all achalasia frames in esophagoscopy videos. Since it is hard to distinguish achalasia features, they further introduced dense pooling connections and dilated convolutions in the CNN to better extract features from esophagoscopy frames. They reported a real-time achalasia detection system that achieved 0.872 accuracy and 0.943 AUC score on their dataset [79].
