AI-Assisted Diagnosis of Dyssynergic Defecation Using Deep Learning Approach on Abdominal Radiography and Symptom Questionnaire

Abstract

Patients are required to undergo specialized tests for dyssynergic defecation diagnosis. However, these tests are limited to tertiary healthcare centers. The aim of this paper is to prescreen potential patients from primary and secondary healthcare centers for further diagnostic tests by using easily obtainable data. We proposed an integrated model which utilizes symptom questionnaire and abdominal radiograph. First, we applied some of the most popular tree-based machine learning algorithms on symptom questionnaire. The best set of features was selected through feature selection. Second, a state-of-the-art image classification model, EfficientNet, was applied on abdominal radiograph with several image augmentation techniques for data preprocessing. Third, we combined the selected input features from symptom questionnaire with the image features extracted from the abdominal radiograph using a concatenate layer to imitate how human experts diagnose in real life. The combined data was used as input to the integrated model. The results demonstrate that our model outperforms the baseline models with a sensitivity of 73.08%, specificity of 57.33%, f1-score of 65.07%, and accuracy of 65.36%.