Towards Causality-Aware Inferring: A Sequential Discriminative Approach for Medical Automatic Diagnosis.
2021
Through learning from the patient simulator built on the collected patient-doctor dialogues records, medical automatic diagnosis (MAD) aims to build an interactive diagnostic agent to sequentially inquire about symptoms for discriminating diseases. However, due to some task-unrelated and non-causal associations in these collected data, e.g., the preference of the collectors, the simulator is probably biased against the disease-symptom causality and the diagnostic agent might be hindered from capturing the transportable knowledge. This work attempts to address these critical issues in MAD by taking advantage of the structural causal model (SCM) to identify and resolve two representative non-causal biases, i.e., (i) default-answer bias and (ii) distributional inquiry bias, from the aspects of the data usage and the agent design, respectively. Specifically, Bias (i) originates from that the patient simulator tries to answer unrecorded inquiries with default answers, which cannot be resolved by feeding more data [1]. Suffering from the biased simulator, previous MAD methods cannot fully demonstrate their advantages. To eliminate this bias and inspired by the propensity score matching technique with SCM, we propose a propensity-based patient simulator to effectively answer unrecorded inquiry by drawing knowledge from the other records; Bias (ii) inherently comes along with the passive manner of collecting MAD data. To this end, we propose a progressive assurance agent, which includes the dual processes accounting for symptom inquiry and disease diagnosis. The inquiry process is driven by the diagnosis process in a top-down manner to inquire about symptoms for enhancing diagnostic confidence. The diagnosis process can reason within that mental representation by intervening with imaginary questions.
Keywords:
- Correction
- Source
- Cite
- Save
- Machine Reading By IdeaReader
65
References
0
Citations
NaN
KQI