An interpretable deep learning based approach for chronic obstructive pulmonary disease using explainable artificial intelligence

Artificial intelligence has become like-humans in thinking and interpretations. But its uses are still limited and are viewed as black boxes, and this is the most important factor underlying the limited applications, especially in the field of health care. Therefore, there is a need for interpretable prediction that provide better predicts and also explain their prediction. This paper proposed an approach for identification and interpretation of chronic obstructive pulmonary disease (COPD) using exhaled breath data and providing explanations of the predictions results based on explainable artificial intelligence. This paper examined a total of 78 patients, using data collected from 8 sensors that analyzed exhaled air. Diagnostic reasons often involve the utilization of transfer learning techniques based on Deep Neural Networks (DNNs). Five pre-train CNNs were tested for recognizing COPD with global average pooling and flattening layers. For interpreting the results, the given results of the DNN are explained by a twin system that uses Case Based Reasoning. The mission of the twinning system is to convert a black box into a white box to be easier to explain. Compared to the other pre-train CNNs models, GoogleNet deep learning model produces promising results, with up to 100% test accuracy with small number of parameters.