AI model distinguishes COVID-19 from pneumonia in X-rays
New ML model achieves 75.4% accuracy using texture-based analysis...
Independent researcher Yogisri Pujitha Chinthoti published a pilot study on arXiv demonstrating a machine learning approach to differentiate COVID-19 from other pneumonias using chest X-ray analysis. The work leverages classical texture descriptors—Histogram of Oriented Gradients (HOG) and Gray-Level Co-occurrence Matrix (GLCM)—paired with traditional classifiers like logistic regression, random forests, and support vector machines.
The study evaluated 668 public chest radiographs (posteroanterior/anteroposterior views from 408 patients) from the COVID-19 Image Data Collection. Under patient-level 5-fold stratified cross-validation to prevent data leakage, the best model achieved a mean accuracy of 75.4% and an AUC of 0.755, modestly outperforming the 71.6% majority-class baseline. While promising, the authors emphasize the limitations of small sample size and advocate for a future multi-modal deep learning architecture combining convolutional and transformer encoders across imaging modalities.
- Model uses HOG and GLCM texture descriptors with classical ML classifiers (logistic regression, random forest, SVM)
- Achieves 75.4% accuracy and AUC 0.755 on 668 public chest X-rays, outperforming 71.6% baseline
- Study highlights need for larger, multi-institutional datasets for robust, generalizable multi-modal deep learning
Why It Matters
Early AI tools for rapid pulmonary disease triage could reduce diagnostic delays and improve patient outcomes in resource-limited settings.