Artificial Intelligence Model for Time Series Classification: Prediction of Delayed Balloon Expulsion Test Using High-Resolution Anorectal Manometry Data and Time-Series Integrated Pressurized Volume.

Background: We previously demonstrated the novel concept of using the integrated pressurized volume (IPV) with high-resolution anorectal manometry (HRAM) and found that it was predictive of delayed balloon expulsion (BE) test results. However, previous IPV methods did not account for chronological changes in anorectal force. To fully utilize the temporal dynamics of HRAM data and enhance BE test prediction, we introduced time-series IPVs (TS-IPVs) and developed an artificial intelligence (AI)-based diagnostic model.

Methods: A total of 300 patients with constipation (130 male and 170 female patients) were enrolled and underwent HRAM and BE tests from September 2020 to May 2021. The TS-IPVs were calculated within a particular time interval during the push maneuver. Convolutional neural networks (CNNs) and a long short-term memory (LSTM) network were applied to predict BE test results.

Key results: Delayed BE was observed in 69 (53.1%) male and 49 (28.8%) female patients. According to the receiver operating characteristic curve analysis, the TS-IPV ratio between the upper 1 cm and lower 3 cm of the anal canal (TS-IPV₁₃ ratio) was the best parameter for predicting BE test results in all patients. Using the TS-IPV₁₃ ratio, the proposed model achieved area under the curve (AUC) values of 0.988 and 0.996 for female and male patients, respectively.

Conclusions and inferences: Our AI model accurately classified delayed BE test results using raw HRAM data and TS-IPVs of male and female patients with an AUC of 0.99. Furthermore, the model used time-variant HRAM pressure data and TS-IPVs throughout the push maneuver without any data loss; therefore, TS-IPV could be used as a more reliable marker than conventional parameters for classifying delayed BE test results.