Related Experiment Videos
Automatic segmentation of lung fields in chest radiographs
B van Ginneken1, B M ter Haar Romeny
1Image Sciences Institute, Utrecht University, The Netherlands. bram@isi.uu.nl
Medical Physics
|December 1, 2000
Summary
Accurate lung field segmentation in chest X-rays is crucial for automated analysis. A novel hybrid approach combining rule-based and pixel classification methods achieves over 94% accuracy, nearing expert performance.
Area of Science:
- Medical Imaging
- Computer-Aided Diagnosis
- Radiology
Background:
- Accurate delineation of anatomical structures in chest radiographs is vital for automated image analysis.
- Applications include tuberculosis screening and computer-assisted diagnosis (CAD).
Purpose of the Study:
- To develop and compare algorithms for automatic lung field segmentation in chest radiographs.
- To evaluate the performance of different segmentation techniques against interobserver variability.
Main Methods:
- Comparison of segmentation techniques including matching, pixel classifiers, a novel rule-based edge/ridge detection scheme, and a hybrid approach.
- Performance evaluation using accuracy metrics on a test set of 115 chest radiographs.
Main Results:
- The hybrid scheme, combining rule-based and pixel classification, demonstrated the best performance.
- Accuracy exceeded 94% across the test set, with an average accuracy of 0.969 +/- 0.0080.
- This performance closely approximates interobserver variability (0.984 +/- 0.0048).
Conclusions:
- A hybrid segmentation approach offers robust and accurate lung field delineation in chest radiographs.
- The developed methods are computationally efficient, suitable for standard PC platforms.
- This technique shows promise for enhancing automated analysis in radiological applications.