CoAD: Automatic Diagnosis through Symptom and Disease Collaborative Generation

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 6348–6361 July 9-14, 2023 ©2023 Association for Computational Linguistics CoAD: Automatic Diagnosis through Symptom and Disease Collaborative Generation Huimin Wang1∗, Wai-Chung Kwan2,3∗, Kam-Fai Wong2,3, Yefeng Zheng1 1Jarvis Lab, Tencent, Shenzhen, China 2The Chinese University of Hong Kong, Hong Kong, China 3MoE Key Laboratory of High Confidence Software Technologies, China {hmmmwang,yefengzheng}@tencent.com {wckwan,kfwong}@se.cuhk.edu.hk Abstract Automatic diagnosis (AD), a critical applica- tion of AI in healthcare, employs machine learning techniques to assist doctors in gath- ering patient symptom information for pre- cise disease diagnosis. The Transformer-based method utilizes an input symptom sequence, predicts itself through auto-regression, and em- ploys the hidden state of the final symptom to determine the disease. Despite its simplicity and superior performance demonstrated, a de- cline in disease diagnosis accuracy is observed caused by 1) a mismatch between symptoms observed during training and generation, and 2) the effect of different symptom orders on disease prediction. To address the above obsta- cles, we introduce the CoAD, a novel disease and symptom collaborative generation frame- work, which incorporates several key innova- tions to improve AD: 1) aligning sentence-level disease labels with multiple possible symptom inquiry steps to bridge the gap between train- ing and generation; 2) expanding symptom la- bels for each sub-sequence of symptoms to en- hance annotation and eliminate the effect of symptom order; 3) developing a repeated symp- tom input schema to effectively and efficiently learn the expanded disease and symptom labels. We evaluate the CoAD framework using four datasets, including three public and one private, and demonstrate that it achieves an average 2.3% improvement over previous state-of-the- art results in automatic disease diagnosis. For reproducibility, we release the code and data at https://github.com/KwanWaiChung/coad. 1 Introduction The healthcare industry worldwide is facing an acute shortage of healthcare professionals such as doctors, nurses, and other staff, which results in millions of people not receiving the care they need, particularly in low-income countries (World Health Organization, 2016). Artificial intelligence (AI) ∗Equal Contribution Figure 1: An example of automatic diagnosis procedure. has the potential to revolutionize medicine by au- tomating tasks traditionally done by humans, re- ducing the time and cost of such tasks. Automatic diagnosis (AD) is a valuable application of AI in healthcare that aims to improve patient outcomes. When deployed on mobile devices, the AD agent functions as a chatbot, querying patients about their symptoms and health concerns, and directing them to the appropriate care based on the diagnosis. This allows for faster treatment decisions, prompt notifi- cation of care teams, and increased communication between providers, ultimately leading to improved patient outcomes and potentially saving lives. The process of AD can be conceptualized as a series of questions and answers. As depicted in Figure 1, the diagnosis begins with the patient reporting ini- tial symptoms (in this example, only a headache). The AD agent then determines whether to ask for additional symptoms or provide a disease diagno- sis. The agent carefully …

Content truncated to 500 words