Related Experiment Video
Updated: Jul 30, 2025
14:08
Automated Midline Shift and Intracranial Pressure Estimation based on Brain CT Images
Published on: April 13, 2013
42.7K
Text Analysis of Radiology Reports with Signs of Intracranial Hemorrhage on Brain CT Scans Using the Decision Tree
А N Khoruzhaya1, D V Kozlov2, К M Arzamasov3
1Junior Researcher, Department of Innovative Technologies; Research and Practical Clinical Center for Diagnostics and Telemedicine Technologies of the Moscow Health Care Department, Bldg 1, 24 Petrovka St., Moscow, 127051, Russia.
Sovremennye Tekhnologii V Meditsine
|May 14, 2023
Summary
This study developed a decision tree algorithm to automatically detect intracranial hemorrhage (ICH) signs in brain CT reports, achieving high accuracy. The algorithm aids in classifying CT scans, but requires manual review for quality control.
Area of Science:
- Medical Informatics
- Radiology
- Machine Learning
Background:
- Automated analysis of medical reports is crucial for efficient healthcare.
- Intracranial hemorrhage (ICH) detection in brain CT scans requires accurate and timely classification.
Purpose of the Study:
- To develop and evaluate a decision tree algorithm for binary classification of brain CT text reports.
- To automatically identify the presence or absence of intracranial hemorrhage (ICH) signs.
Main Methods:
- Utilized a dataset of 34,188 brain CT studies from URIS UMIAS.
- Employed Natural Language Toolkit (NLTK) and scikit-learn for data preprocessing and classification.
- Developed a decision tree model trained on 2786 reports and tested on 1194 reports, using keyword and stop-phrase analysis.
Main Results:
- The algorithm achieved a sensitivity of 0.94, specificity of 0.88, and F-score of 0.83 in classifying CT reports for ICH.
- Demonstrated high accuracy in distinguishing between CT reports with and without signs of ICH.
Conclusions:
- The developed algorithm effectively performs binary classification of brain CT reports for ICH detection.
- The algorithm can assist in creating datasets for radiological analysis but necessitates manual quality control.
