Automated Discrimination of Tissue Boundaries Using Ultrasound Images

Abstract

Balance between human body compositions, e.g. bones, muscles, and fat, is one of the major and basic indicators related to personal health. Body composition analysis using ultrasound imaging has been developed rapidly. However, interpretation of ultrasound images is conducted manually, and accuracy and confidence in interpretation requires experience. This paper proposes an automated approach to determine boundaries between tissues, with which thickness of subcutaneous fat and muscles can be obtained. A portable B-mode ultrasound echographic device was used in this study. The proposed automated discrimination method utilizes statistical characteristics of anatomy, intensity and shape information about fat and muscle tissues and bone to determine boundaries. Furthermore, the proposed method was developed considering real-time and high-speed implementation and reduction of computational cost. Validity of the proposed method was evaluated in nine subjects (one woman, eight men; ages 22-59 yr) at three anatomical sites. Experimental results show that the proposed method can discriminate the thickness with high accuracy. Finally, a sensitivity analysis of discrimination error to changes in parameters of the proposed method was carried out.