Browsing by Author "Kocatekin, Tuğberk"

Filter results by typing the first few letters
Now showing 1 - 1 of 1
  • No Thumbnail Available
    Item
    Text mining in Turkish radiology reports
    (Bahçeşehir Üniversitesi Fen Bilimleri Enstitüsü, 2013-09) Kocatekin, Tuğberk; Ünay, Devrim
    Text mining and text classification is a popular area of machine learning and information retrieval. Automated categorization and analysis of medical documents may improve work flow, and aid in better diagnosis and therapy planning. There is already some research done on analysis and categorization of radiology reports. However, to the best of our knowledge there is no prior work on anatomical region based classification of Turkish radiology reports. In order to fill this gap, this thesis focuses on dictionary-based classification of Turkish radiology reports into anatomical regions. The proposed solution is intented to automatize, speed up, and improve the accuracy of the task of classifying these documents, which is manually realized traditionally. The proposed solution, implemented in Bash environment, consists of header-footer removal, Turkish character elimination, stemming, word frequency analysis, normalization, and scoring steps. Training (n=69) and performance evaluation (n=161) of the system is realized using a total of 230 Turkish radiology reports from 8 different anatomical regions acquired from routine clinical practice. F-score of the system is measured as 98,6%, and it is observed that the proposed system correctly identifies the actual classes of 7 reports that were previously misclassified by the radiology staff. In order to improve the accuracy of the system one can increase the size of the training set, incorporate natural language processing solutions, or make use of ontologies that encode anatomical/pathological knowledge. In addition to that, the proposed system can be integrated with speech processing solutions to automatically create radiology reports from audio recordings of radiologists. Lastly, the system can be further improved by user feedback.