School of Information Science

Permanent URI for this college

http://etd.aau.edu.et/handle/123456789/26

Browse

Browsing School of Information Science by Author "Abebe, Bethelhem"

Now showing 1 - 1 of 1
  • No Thumbnail Available
    Item
    Applying Text Mining Techniques to Extract Knowledge from Cancer Patients’ Medcial Records-The case of Tikur Anbessa Specialized Hospital
    (Addis Ababa University, 2018-01) Abebe, Bethelhem; Yifiru, Martha (PhD); Yilma, Mengistu (PhD)
    Background: Currently, there are a lot of medical texts accumulated than interpreted. These texts have to be organized and analyzed effectively in order to be useful. Nowadays text mining has become very important in analyzing these medical texts and finding patterns effectively. The medical records of chronic ill patients (especially cancer) contain a lot of information both image and textual formats which are vital to immediate patient care which could also help in different researches and finding difficult cases. Objective: The aim of this project is applying text mining techniques to extract knowledge from oncology patients’ medical records. Method: In order to conduct this project, data was collected from oncology patients’ medical records of Tikur Anbessa specialized hospital. The CRISP methodology was applied for describing the pattern in these medical records. For extracting the patterns from 137 medical records, R software was used. After creating a corpus and having pre processed the medical records, a pattern was extracted using hierarchical and k means clustering algorithm. The patterns extracted from these two algorithms were compared and evaluated. To evaluate the pattern extracted both the subjective and objective evaluation approaches were used. The subjective evaluation was done with the help of ten physcians (both residents and oncologists). For the objective evaluation Rand index, accuracy, precison, recall and F measures were performed. Result: According to the assessment of the medical records indicated, searching the necessary medical records from the record room was difficult almost impossible, their follow up formats are disorganized and the physicians’ handwriting is illegible. These make knowledge discovery difficult, time taking and tiresome. As the objective evaluation methods showed the hierarchical algorithm performed better than the k means (Rand index=66.2%, Accuracy=48.8% and Precision and recall=65.6%) and 50% of the physicians chose also the hierarchical algorithm. During the subjective evaluation, out of the ten physicians, three of them (30%) did not have any idea as which one is better because the idea was new to them and difficult to understand the patterns. Two of them (20%) preferred the k means to the hierarchical because the hierarchical seemed complicated to them. The rest (50%) chose the hierarchical algorithm since it tried to show almost the necessary pattern found in the patients’ medical records.Conclusion: In general text mining eases access of the necessary knowledge rather than going through patients’ medical records and any one can get the necessary knowledge from the pattern extracted in the oncology patients’ medical records. From the project it is easy to see that the hierarchical algorithm performed better. Recommendation: Text mining has different applications and each application has different benefits in the medical health care. And different kinds of knowledge can be discovered and predicted not only from cancer medical records but also from other chronic illnesses that need further researches and experiments. For researchers, there is a great need of text mining applications in the medical domain specially using clustering algorithms in order to extract new knowledge. Also the efficiency of the k means and hierarchical clustering needs to be improved. For the health practitioners and Tikur anbessa specialized hospital, this application will give them a great benefit, so handling the patients’ medical record in a proper and organized way will give opportunity to give quality of care for the patient. Also for the physicians and different researchers, it will give ease acccess of the necessary data for knowledge extraction and patient management system. For software development business organizations, there is a great opportunity to work on the text mining area especially in the medical domain which needs more structuring and handling medical data.