Deep learning facilitates the diagnosis of adult asthma
Katsuyuki Tomita, Hirokazu Touge, Hiromitsu Sakai, Hiroyuki Sano, Yuji Tohda · 2018
We explored whether the use of deep learning to model combinations of symptoms, physical findings, and diagnostic tests, such as spirometry and the bronchial challenge test, would improve model performance in predicting the initial diagnosis of adult asthma compared to the conventional method. The data were obtained from the clinical records on 566 adult out-patients who visited Kinki University Hospital for the first time with complaints of non-specific respiratory symptoms. Asthma was comprehensively diagnosed by specialists based on symptoms, physical findings, and diagnostic tests. Model performance metrics were compared to neural network and support vector machine (SVM) learning. For the diagnosis of adult asthma based on symptoms and physical findings, the accuracy for the RNN model was 0.68, compared to that for the SVM (0.60). When adult asthma was diagnosed based on symptoms, physical findings, and diagnostic tests, the accuracy of the RNN model increased to 0.89 and was significantly higher than the 0.7 accuracy of the SVM. Deep learning models based on symptoms, physical findings, and diagnostic tests appear to improve the performance of models for diagnosing adult asthma.