Retrieval and classification of pneumoconiosis chest radiograph images using multiscale AM-FM methods

We propose the use of Amplitude-Modulation Frequency-Modulation (AM-FM) features for representing and retrieving X-Ray images with pneumoconiosis. The AM-FM features are estimated using multiscale filterbanks with wave-lengths related with the standard sizes for grading the level of opacities in X-Rays. The extracted AM-FM features represent opacity profusion in terms of instantaneous frequency (IF) and instantaneous amplitude (IA) features. Here, IF estimates in the medium and high scale frequencies can be used to capture early disease symptoms. AM-FM features from the low and medium scale frequencies are associated with advanced disease stages. We demonstrate the performance of the system in X-ray image retrieval and classification applications.