| Autor: |
Yokota, Shinichiroh, Doi, Shunsuke, Fukuhara, Masakazu, Mitani, Tomohiro, Nagashima, Satomi, Gonoi, Wataru, Imai, Takeshi, Ohe, Kazuhiko |
| Zdroj: |
Health & Technology; Jan2023, Vol. 13 Issue 1, p65-73, 9p |
| Abstrakt: |
Purpose: Accurate disease diagnosis from radiology reports is important in medical treatment. Preventing physicians from overlooking the findings of relevant radiology reports is thus an important global issue. This study is part of a project that aims to develop and verify a program that detects and notifies physicians of diagnoses that do not appear in the patient's medical records. Methods: In this study, we developed software functions that (1) extract diagnoses from radiology reports, (2) search medical records for the extracted diagnoses and output them and their corresponding ICD10 codes, which do not appear in the patients' medical records, and (3) automatically execute all processes daily and notify auditors by email. We verified seven cases including diagnoses suspected of being overlooked that are automatically extracted by our system from 1,194 radiology reports. Results: By checking the output obtained from the system constructed by incorporating a high-performance sentence classifier for disease name polarity classification with AUC = 0.972, it was possible to extract disease names that physicians might have overlooked. We could extract a past actual overlooked case by this algorithm. From verifying seven cases, we assessed that the physician might not have considered one diagnosis in one case. Conclusion: These results demonstrate that the developed system is useful when auditors check patients' medical records to confirm the status of overlooked reports and diagnoses. In future work, we will develop additional functions and tools for incorporation into the proposed system to facilitate its application in clinical practice. [ABSTRACT FROM AUTHOR] |
| Databáze: |
Complementary Index |
| Externí odkaz: |
|
Full text from SpringerLink