Automatic edge detection of foetal head and abdominal circumferences using neural network arbitration

Abstract

Image recognition has increasingly become vital in the rapid development in industrial and medical applications. The need for better image processing techniques has lead to the emergence of many approaches and techniques, which aim to achieve faster, cheaper and more user-friendly operations. Multiscale analysis for edge detection is a powerful approach to image recognition. However, this approach suffers huge computational and time cost which, consequently, jeopardise its efficiency. This paper presents a new technique in image processing that aims to reduce high computational cost, to provide rapid 3-dimensional objects recognition and has a wide domain of applications. The novel Automatic Edge Detection Scheme (AEDS) combines neural network technology with multiscale analysis, to achieve automatic recognition of high and low contrast images that contain two or three-dimensional objects. The AEDS is based on applying scale space analysis using the Laplacian of a Gaussian edge detection operator, together with a neural network model. This technique will be utilised to detect foetal head and abdominal circumferences. Monitoring the development of a baby during gestation using ultrasound scan, provides vital information on the growth of the foetus and can predict any abnormalities such as Down's syndrome. Providing a fast, user-friendly and low-cost system for the automatic detection, prevents the possibility of human errors in obtaining foetal measurements. In addition, the new automatic image recognition technique can be efficiently applied in other fields.