Towards Interpretable Clinical Diagnosis with Bayesian Network Ensembles Stacked on Entity-Aware CNNs

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 3143–3153 July 5 - 10, 2020. c⃝2020 Association for Computational Linguistics 3143 Towards Interpretable Clinical Diagnosis with Bayesian Network Ensembles Stacked on Entity-Aware CNNs Jun Chen, Xiaoya Dai, Quan Yuan, Chao Lu and Haifeng Huang Baidu Inc, Beijing, China {chenjun22,daixiaoya,yuanquan02,luchao,huanghaifeng}@baidu.com Abstract The automatic text-based diagnosis remains a challenging task for clinical use because it requires appropriate balance between accu- racy and interpretability. In this paper, we attempt to propose a solution by introducing a novel framework that stacks Bayesian Net- work Ensembles on top of Entity-Aware Con- volutional Neural Networks (CNN) towards building an accurate yet interpretable diagno- sis system. The proposed framework takes advantage of the high accuracy and general- ity of deep neural networks as well as the interpretability of Bayesian Networks, which is critical for AI-empowered healthcare. The evaluation conducted on the real Electronic Medical Record (EMR) documents from hos- pitals and annotated by professional doctors proves that, the proposed framework outper- forms the previous automatic diagnosis meth- ods in accuracy performance and the diagnosis explanation of the framework is reasonable. 1 Introduction The automatic diagnosis of diseases has drawn the increasing attention from both research com- munities and industrial companies in the recent years due to the advancement of artificial intelli- gence (AI) (Liang et al., 2019; Esteva et al., 2019; Liu et al., 2018). As reported in (Anandan et al., 2019), “AI-enabled analysis software is helping to guide doctors and other health-care workers through diagnostic processes and questioning to arrive at treatment decisions with greater speed and accuracy.” Although the image-based diag- nosis has been well studied using PACS (Picture Archiving and Communication Systems) data (Lit- jens et al., 2017), the text-based diagnosis for Clin- ical Decision Support (CDS) (Berner, 2007) re- mains difficult due to the rare access to reliable clinical corpus and the difficulty in balancing be- tween accuracy and interpretability. Table 1: A real outpatient EMR from hospital. Section Content Basic 男, 30岁 (Male, 30 years old) CC 咽部不适3天 (Pharyngeal discomfort for 3 days) HPI 患者于3日前起咽痛伴发热, 无呼吸困难、咳嗽、 咳痰、嗳气或反酸 (The patient developed pharyngalgia and fever 3 days ago, without dyspnea, cough, sputum, belching or acid reflux) PE 咽峡稍充血, 双侧扁桃体Ⅰ度肿大, 无栓塞物及瘢痕 (The hypopharyngeal isthmus is slightly congested. The bilateral tonsils are first-degree enlarged. There is no embolism or scar in the pharynx.) TR 血常规示白细胞计数升高, WBC12.5 ∗ 109/L. C反应 蛋白正常. ( The blood test showed elevated white blood cell count, WBC12.5 ∗ 109/L. The C-reactive protein is normal.) Diagnosis 急性扁桃体炎 (Acute tonsillitis) There have been attempts to study automatic text-based diagnosis with Electronic Medical Record (EMR) documents integrated in the Hospi- tal Information System (Mullenbach et al., 2018; Yang et al., 2018; Girardi et al., 2018). Basically, an EMR document is written by a doctor and con- sists of several sections that describe the illness of the patient. Besides the patient’s basic informa- tion like name, age and gender, an EMR document contains Chief Complaint (CC), History of Present Illness (HPI), Physical …

Content truncated to 500 words