African Journal of
Biotechnology

  • Abbreviation: Afr. J. Biotechnol.
  • Language: English
  • ISSN: 1684-5315
  • DOI: 10.5897/AJB
  • Start Year: 2002
  • Published Articles: 12496

Full Length Research Paper

Artificial neural network analysis of ultrasound image for the estimation of intramuscular fat content in lamb muscle

Piotr Åšlósarz1*, Marek Stanisz1, Piotr Boniecki2, Andrzej Przybylak2, Dariusz Lisiak3 and Agnieszka Ludwiczak1
  1PoznaÅ„ University of Life Sciences, Department of Small Mammal Breeding and Animal Origin Materials, ZÅ‚otniki, SÅ‚oneczna 1, 62-002 Suchy Las, Poland. 2PoznaÅ„ University of Life Sciences, Institute of Agricultural Engineering, Wojska Polskiego 50, 60-625 PoznaÅ„, Poland. 3Institute of Agricultural And Food Biotechnology, Division of Meat And Fat Technology, GÅ‚ogowska 239, 60-111 PoznaÅ„, Poland.
Email: [email protected]

  •  Accepted: 26 August 2011
  •  Published: 21 September 2011

Abstract

 

The purpose of the study was to evaluate the effectiveness of ultrasound image analysis of sections of the longissimus dorsi muscle in lambs using artificial neural networks to estimate intramuscular fat content. Ultrasound images of the musculus longissimus dorsi cross-sections on the right flank behind the last rib were collected from 169 live lambs of both sexes prior to slaughter at the age of approximately 4 months. The analyses were conducted using a Pie Medical 100LC ultrasound scanner with an 8.0 MHz linear probe. On recorded ultrasound images using the MultiScan® ver. 12.05 software (Computer Scanning Systems Ltd.), measurements were taken for the thickness of subcutaneous fat, as well as the depth and cross-section area of the muscle. Further, on each ultrasound image, 10 frames of 1 × 1 cm were marked, which were converted to the ASCII format using the scale from 0 – a black pixel, to 255 – a image pixel of maximum brightness. After slaughter the content of extraction fat was determined using laboratory analyses in samples of the longissimus dorsi muscle collected on the site of ultrasound examinations. The data analysis was performed using a simulator of artificial neural networks implemented in the Statistica ver. 7.1 software package. The multi-layer perceptron proved to be the neural model with the best validation parameters (correlation = 0.858, mean prediction error = 0.151). All the tested network models resulted in an overestimation of the estimated fat contents. Despite that fact, obtained results are promising and more accurate than the previously applied regression method.

 

Key words: Sheep, intramuscular fat, neural network, ultrasonography.