JMIR Publications

ABSTRACT

Background:

Tinnitus diagnosis poses a challenge in otolaryngology owing to extremely complex pathogenesis, lack of effective objectification methods, and factor-affected diagnosis. There is currrently a lack of explainable auxiliary diagnostic tools for tinnitus in clinical practice.

Objective:

This study aims to develop a diagnostic model using an explainable artificial intelligence method to address the issue of low accuracy in tinnitus diagnosis.

Methods:

In this study, a knowledge-graph-based tinnitus diagnostic method was developed by combining clinical medical knowledge with electronic medical records (EMRs). A method was proposed for integrating patient EMR data with traditional Chinese clinical medical knowledge to construct a tinnitus knowledge graph. Subsequently, weights were introduced that measured patient similarity in the knowledge graph based on mutual information values. Finally, a collaborative neighbor algorithm was proposed that scored patient similarity to obtain the recommended diagnosis. We conducted extensive experiments to explore the effectiveness of our models and compared them with state-of-the-art graph algorithms and other explainable machine learning models.

Results:

The experimental results show that the accuracy, sensitivity, specificity, precision, f1-score and area under curve (AUC) of our proposed method all exceed 98% for five tinnitus subtypes while maintaining excellent interpretability. The topological structure of knowledge graphs provides a transparency that can explain why certain patients are similar.

Conclusions:

This method provides doctors with a reliable and explainable diagnostic tool that is expected to improve tinnitus diagnosis accuracy.