Analysis of capsule endoscopy images related to gastric ulcer using bidimensional empirical mode decomposition

Capsule endoscopy (CE) is a novel technology that allows direct noninvasive visualization of the entire small intestine. CE permits a detailed examination in the ambulatory setting, allowing identification of clinically relevant lesions, and it is appealing to both patients and providers. In this context, advanced image processing analysis could facilitate the physician to come up with a diagnosis, providing him/her with the appropriate classification indicators. Towards this direction, in this work, the Bidimensional Empirical Mode Decomposition (BEMD) was applied to small intestine images generated by a CE system (i.e., the Pillcam SB capsule) to extract their Intrinsic Mode Functions (IMFs). The latter could be used as a new classification domain, as they reflect different modes included in the original signal and related to the underlined pathology. BEMD is more advantageous compared to other techniques (such as Fourier analysis, Wavelets, AM-FM decomposition) due to its adaptation to the nonstationary character of the signals (i.e., most of natural images exhibit such behaviour), and the extraction of global structures due to its better stability. In this paper, the BEMD analysis is focused on endoscopic images related to gastric ulcer, which is the most popular disease of the gastrointestinal tract. The corresponding IMFs reveal differences in structure and provide features from their finest to their coarsest scale, structuring a new analysis, recognition and classification domain.