An Ensemble Net of Convolutional Auto-Encoder and Graph Auto-Encoder for Auto-Diagnosis

Effective auto-diagnosis assistants can benefit our healthcare system in various aspects, such as, saving labor cost, sharing knowledge among the crowd, and timely supporting the patients. However, the existing auto-diagnosis models are ineffective due to issues caused by information island, poor information coding, and inefficient informative retrieval. To address these issues, this article presents a diagnosis assistant that is designed and implemented to manage abundant historical inquiries between patients and doctors. The core of the auto-diagnosis system is a novel model called ensemble net of convolutional auto-encoder and graph auto-encoder (EN-C+GAE) which can be trained using historical data and generate a list of candidate diagnoses for a doctor to select. The experimental results show that the proposed approach outperforms the counterparts in generating more fluent and relevant diagnoses. The proposed system also shows its potential in real-world deployment in healthcare scenarios.