Classifying Medical Imaging Modalities Using Visual and Textual Features
Author Information
Author(s): Han Xian-Hua, Chen Yen-Wei
Primary Institution: Ritsumeikan University
Hypothesis
Can combining visual and textual features improve the classification of medical imaging modalities?
Conclusion
The proposed method achieved an accuracy rate of about 97% in classifying medical imaging modalities.
Supporting Evidence
- The method achieved a classification rate of 97% on the ImageCLEF 2010 dataset.
- Local classifiers were designed to improve accuracy for easily misclassified modalities.
- Different types of visual features were combined to enhance classification performance.
Takeaway
This study shows how using both pictures and words can help computers better understand different types of medical images.
Methodology
The study used a combination of visual features (like histograms and SIFT descriptors) and textual features (binary histograms from image captions) for classification using SVM.
Limitations
The study did not release ground-truths for the evaluated and test datasets, which may affect the validation of results.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website