Abstract (300 words maximum)
Electronic health records are digital versions to manage and facilitate consultations and follow-up on treatments with the patients. It includes information like medical history, diagnosis, medications and test results of a patient. This digital system has been adopted by the majority of developed countries across the globe. Clinical services use ICD codes [International Classification Codes] to code the diseases and medical condition of an individual. ICD codes have a significant role in secondary purposes including funding, insurance claim processing and research. The main challenge with the use of ICD codes for a patient is to decode its meaning due to the complexity of technical medical terminology, lack of common language/context and no reference available on EHR system to direct translate it. To overcome this challenge, we would like to propose an AI-powered model to help the patients with decrypting the description of the ICD code mentioned on the medical record. The model would use unstructured clinical summaries and map it to the most relevant ICD codes. Model includes usage of Pytorch for deep model building, Flask for flexible web framework design, publicly available kaggle dataset for ICD-10 code library and clinical narrative generation and BERT (Bidirectional Encoder Representations from Transformers) for extracting the medical concepts and map them to the appropriate ICD codes. By leveraging BERT’s ability to understand and predict, we will be targeting to bridge the gap between patient comprehension and ICD Codes with an accuracy of ~ 60-80%. This will help patients to avoid any anxiety about misinterpretation, simple understandable explanations and continual usage of digital systems.
Academic department under which the project should be listed
CCSE - Computer Science
Primary Investigator (PI) Name
Md Abdullah Al Hafiz Khan
Mapping the Lexicon of Healthcare: Connecting ICD codes to Clinical narratives
Electronic health records are digital versions to manage and facilitate consultations and follow-up on treatments with the patients. It includes information like medical history, diagnosis, medications and test results of a patient. This digital system has been adopted by the majority of developed countries across the globe. Clinical services use ICD codes [International Classification Codes] to code the diseases and medical condition of an individual. ICD codes have a significant role in secondary purposes including funding, insurance claim processing and research. The main challenge with the use of ICD codes for a patient is to decode its meaning due to the complexity of technical medical terminology, lack of common language/context and no reference available on EHR system to direct translate it. To overcome this challenge, we would like to propose an AI-powered model to help the patients with decrypting the description of the ICD code mentioned on the medical record. The model would use unstructured clinical summaries and map it to the most relevant ICD codes. Model includes usage of Pytorch for deep model building, Flask for flexible web framework design, publicly available kaggle dataset for ICD-10 code library and clinical narrative generation and BERT (Bidirectional Encoder Representations from Transformers) for extracting the medical concepts and map them to the appropriate ICD codes. By leveraging BERT’s ability to understand and predict, we will be targeting to bridge the gap between patient comprehension and ICD Codes with an accuracy of ~ 60-80%. This will help patients to avoid any anxiety about misinterpretation, simple understandable explanations and continual usage of digital systems.