Global weighted LBP based entropy features for the assessment of pulmonary hypertension

Abstract

Pulmonary hypertension (PH) is characterized by elevated pulmonary arterial pressure. Echocardiography, or cardiac ultrasound, is a helpful imaging tool to screen for PH. However, expert interpretation is required for successful screening. Development of a more automated method for diagnosis of PH would be useful to minimize error, thereby improving patient health. This task is challenging and the literature pertaining to the problem is still nascent. In this paper, we propose a computer aided diagnosis (CAD) tool, using ultrasound images, to expedite the screening of PH. Textural components play a significant role in ultrasound imaging for the efficient identification of PH. The extraction of such features is accomplished by computing several entropy measurements over a globally weighted local binary pattern (LBP). Thereafter, the blend of ranked maximum and fuzzy entropy features are input to a support vector machine, resulting in a maximum accuracy of approximately 92%. A comparison with variants indicates improved performance of the proposed globally weighted LBP.