진단 프로세스 기반 설명 가능한 전문가 지식 모델링에 관한 연구
A Study on Modeling Explainable Expert Knowledge Based on Diagnosis Process
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
본 논문은 환자의 증상으로부터 의사의 최종치료를 위한 진단결정에 이르는 이유를 묻는 진단프로세스가 각 단계에서 내리는 질의와 대답 (Question and Answer)을 통하여 이루어진다는 점에 주목한다. 이에 따라 각 단계의 질의 (question)을 만드는 모델을 제시한다. 본 모델은 전문의에 의한 임상 진단의 흐름을 반영하여 각 단계별 질문으로 구성하였다. 진단 과정의 인과관계를 추론하기 위해 수행되었으며 다양한 Task를 구성하기 위한 전문의의 분석과정을 포함하였다. 이는 각 환자의 증상에 대하여 의사가 내린 진단결정으로 데이터의 설명가능성을 보장할 수 있으며, 추론적 접근이 가능하다. 최종적으로 의사의 질의와 대답에 대한 보고서와 전문가의 수동 주석을 통한 보고서를 비교 평가한다. 평가는 baseline-model로 Visual commonsense reasoning model, VI-bert, Biobert를 사용한다. 그 결과 BLEU-1 to 2의 평균 성능은 각각 3.5904, 2.8007, 3.4627를 보였으며, rogue-1, 2, l의 평균 성능은 각각 0.1606, 0.3428, 0.0797를 보였다. 이는 보고서 생성에서 데이터에 의료 지식이 필요하다는 것을 보여주며, 인과관계를 학습하기 위한 모델에 적용할 수 있는 설명가능한 데이터라고 볼 수 있다.
Abstract
This paper notes that the diagnostic process of asking why the patient's symptoms lead to the doctor's diagnostic decision for final treatment is made through questions and answers made at each stage. Accordingly, we present a model that creates a question for each step. This model consists of questions at each stage by reflecting the flow of clinical diagnosis only by specialists. It was performed to infer the causal relationship of the diagnostic process and included the analysis process of a specialist to construct various tasks. This is a diagnostic decision made by a doctor for each patient's symptoms, which can guarantee the explanability of the data, and an inferential approach is possible. Finally, we compare and evaluate reports on doctor's questions and answers with reports through manual commentary by experts. The evaluation uses Visual commonsense reasoning model, VI-bert, and Biobert as baseline models. As a result, the average performance of BLEU-1 to 2 was 3.5904, 2.8007, and 3.4627, respectively, and the average performance of Rogue-1, 2, and l was 0.1606, 0.3428, and 0.0797, respectively. This shows that medical knowledge is required for the data in report generation, and can be viewed as Explainable data that can be applied to models for learning causality.
참고문헌 (16)
- J. E. Brush Jr, "How expert clinicians intuitively recognize a medical diagnosis", The American journal of medicine, 130(6), pp.629~634 (2017)
- H. Suri, "MeDiaQA: A question answering dataset on medical dialogues", arXiv preprint arXiv:2108.08074, ~ (2021)
- a de-identified publicly available database of chest radiographs with free-text reports, "A. E. W. Johnson", Scientific Data, ~ (2019)
- A. B. Abacha, "Overview of the vqa-med task at imageclef 2021: Visual question answering and generation in the medical domain", Proceedings of the CLEF 2021 Conference and Labs of the Evaluation Forum-working notes, pp.21~24 (2021)
- Y. Du, "A survey of vision-language pre-trained models", arXiv preprint arXiv:2202.10936, ~ (2022)
- A. Mogadala, "Trends in integration of vision and language research: A survey of tasks, datasets, and methods", Journal of Artificial Intelligence Research, 71, pp.1183~1317 (2021)
- T. Brown, "Language models are few-shot learners", Advances in neural information processing systems, 33, pp.1877~1901 (2020)
- W. Chen, "Visual-linguistic causal intervention for radiology report generation", arXiv preprint arXiv:2303.09117, ~ (2023)
- S. C. M. Gowda, "Pulling up by the causal bootstraps: causal data augmentation for pre-training debiasing", arXiv preprint arXiv:2108.12510, ~ (2021)
- M. Watson, "Agree to disagree: When deep learning models with identical architectures produce distinct explanations", Proceedings of the IEEE, 2022, ~
- M. Watson, "Learning how to MIMIC: Using model explanations to guide deep learning training", Proceedings of the IEEE, 2023, ~
- R. Zellers, "From recognition to cognition: Visual commonsense reasoning", Proceedings of the IEEE, 2019, ~
- W. Su, "Vl-bert: Pre-training of generic visual-linguistic representations", arXiv preprint arXiv:1908.08530, ~ (2019)
- C. Deng, "Extraction of gene-disease association from literature using BioBERT", The 2nd International Conference on Computing and Data Science (2021)
- K. Papineni, "Bleu : a method for automatic evaluation of machine translation", Proceedings of the 40th annual meeting of the Association for Computational Linguistics (2002)
- C. Y. Lin, "Rouge : A package for automatic evaluation of summaries" (2004), Text summarization branches out