SBIR-STTR Award

Over the past 15 years, the emphasis on selecting for leaner pigs with larger loineye areas and high percent lean cuts has resulted in meat with lower water-holding capacity, lower intramuscular fat (IMF), and lower scores for juiciness, flavor and overall pork quality. Swine breeders need a genetic selection tool to help reverse this trend. The purpose of this project is to build on several years of research experience with ultrasound and prediction of IMF for live beef cattle by applying the technology to live swine. Phase I involves serially scanning 400 live animals to develop mathematical models to characterize the IMF in pork samples from each pig. A database of image texture parameters that are predictive of IMF will be developed. Pigs will be harvested, visual marbling scores will be obtained at the packing plant, tissue samples will be taken at the scanning site, and a slice of the longissimus muscle from the 10th to 11th rib interface will be transported to a meat laboratory for chemical analysis. Linear and nonlinear IMF prediction models will be developed using the image texture parameters as the dependent variables and the tissue IMF (chemical fat) as the independent variable. Phase I will prove the ability to enhance IMF prediction accuracy using ultrasound in live pigs by a minimum of 10 percent over the current published research. In addition, Phase I will demonstrate the ability to port an IMF prediction model from one ultrasound equipment type to another with an incremental standard error of prediction of less than 0.25% IMF. OBJECTIVES: The project (Phase I) will demonstrate the feasibility of developing and commercializing an approach for predicting pork quality in live animals using new digital beam forming ultrasound technology now available in low-cost portable real-time ultrasound equipment. Several technical questions need be addressed to establish the feasibility of the proposed approach. 1. Can current research models be enhanced for the prediction of IMF in swine? What parameters should be measured to accurately characterize pork tissue quality in live animals? What physical limitations exist for end users to collect the necessary data on live animals? How can these be minimized? What statistical design is needed to test a prediction model of pork quality using data collected by ultrasound technology? How many animals are needed to develop technology with sufficient statistical significance, accuracy, repeatability, and bias in estimates of heritability? 2. How can the costs of expensive (and extensive) laboratory testing during model development be minimized for varying types of image capturing equipment? These questions will be answered by a public-private effort including Biotronics, Inc. (the small business applicant), the Department of Animal Science at Iowa State University, and potential end users of the technology to be developed. Technical objectives for Phase I are to: 1. Prove that current research models can be enhanced for the prediction of IMF in swine. This objective includes developing image processing algorithms to quantify swine ultrasound texture/marbling; and developing and validating (proof-of-concept) the prediction model for intramuscular fat. 2. Demonstrate the feasibility of porting IMF prediction models from one equipment type to a second equipment type by using tissue mimicking phantoms. APPROACH: TASK 1: Review and refine scanning protocol for new ultrasound equipment types by serially scanning the longissimus muscle in pigs as they mature, and using alternative gain settings and optimal frequency (using a multi-frequency transducer) so that a matrix of data can be established for use in optimizing age to scan the pig at and the equipment gain settings and optimal frequency using multi-frequency tranducers to use. Up to 340 pigs, preferably of Duroc or Duroc cross breeding, will be required to achieve the level of statistical significance for this objective. The pigs will be scanned three times at 3 week intervals, with the last scan being within 5 days of harvest. Two contemporary groups of size 10-15 pigs each will be harvested within 5 days of the second scan to add numbers to the lower end of the % IMF range. Two ultrasound equipment types will be used to scan each pig. TASK 2: Develop a database of image texture parameters using techniques including but not limited to Fourier Transformation, texture gradient, and co-occurrence matrices for all images. Conduct a literature review to determine if there are other image processing algorithms and characterizing parameters that need to be incorporated in this task. A given set of precisely defined region of interest (ROI) will be used to obtain multiple sets of parameters within each image. TASK 3: Harvest the pigs, obtain visual marbling scores at the packing plant, and secure tissue samples at the scanning site. After harvest, a slice of the longissimus muscle from the 10th to 11th rib interface will be analyzed at the ISU Meat Laboratory for carcass IMF percentage by the method outlined in Bligh and Dyer (1959). TASK 4: Perform a statistical analysis to develop linear or nonlinear IMF prediction models using the image texture parameters as the dependent variables and the tissue IMF as the independent variable. The data will be randomized into independent development (60 % of the data) and validation (40 % of the data) data sets. The statistical analysis will be performed within equipment gain settings and scanning period. IMF % prediction models will be compared on model standard error of prediction, standard error of repeatability, bias and rank correlations. TASK 5: Perform a statistical analysis that will take the prediction model developed from one equipment type to the second equipment type using the characterization of each equipment type from images captured from the tissue-mimicking phantom. TASK 6: Validate the equipment-adjusted IMF prediction models on a sample of 60 market ready pigs and compare the equipment adjusted IMF model to the equipment-specific IMF prediction model. Validation scanning will be done within 5 days of harvest. The pork carcasses will be evaluated for visual marbling, and a rib sample will be returned to the ISU Meat Laboratory to chemically determine IMF % for each pig carcass

In the past 15 years, the emphasis on selection for leaner pigs with larger loineye areas and high percent lean cuts has resulted in meat with lower water-holding capacity, lower percent intramuscular fat (% IMF) or marbling, and lower scores for juiciness and flavor. Consumers and an ever growing segment of the pork industry know that more optimal levels of marbling must be put back into the high-value pork cuts. Biotronics, Inc. is developing technology to solve this problem by using real-time ultrasound to measure the % IMF in live pigs and hot carcasses. The premise of the project is that swine breeders can be rewarded financially for production of pigs resulting in higher levels of % IMF if the pork packing plant realizes premiums for their product. The project will refine research models to accurately characterize the % IMF levels in the pork longissimus dorsi muscle in live and hot carcasses, demonstrate an operational system using the technology in the packing plant, and document the anticipated economic impact of the technology for the packer and the swine breeder. Results of the project will allow swine breeders to make genetic selections for optimal levels of % IMF in their seed stock and will give the packing industry an accurate measurement of % IMF to sort and target higher value carcasses to niche markets. All segments of the pork industry will benefit, with the ultimate benefit to be realized by the consumer. OBJECTIVES: In the past 15 years, the emphasis on selection for leaner pigs with larger loineye areas and high percent lean cuts has resulted in meat with lower water-holding capacity, lower percent intramuscular fat (% IMF), and lower scores for juiciness and flavor. Research has confirmed that some minimum level of % IMF, also called marbling, must be maintained for optimal eating quality, and Pacific Rim markets seek pork with even higher levels of % IMF. The premise for the project is that the general consumer and targeted niche markets in the U.S. and internationally will pay a premium for pork cuts that meet specification standards for % IMF. The use of real-time ultrasound and sophisticated image processing algorithms have allowed the beef cattle industry to evaluate livestock animals directly to measure % IMF in the longissimus dorsi (ld) muscle. This live animal evaluation using ultrasound has been widely adopted on an international scale by the beef industry. Although not widely adopted by the swine industry, ultrasound processing methods for % IMF prediction have been successfully demonstrated with live swine. The goal of the project is to develop a system for the accurate measurement of % IMF in swine using real-time ultrasound technology. The technology would allow the seed stock industry to select sires and replacement gilts that are genetically superior for meat quality. The same technology structure would allow the packing industry to noninvasively scan and sort pork loins for quality without breaking the loins. Considering that % IMF is a highly heritable trait, the combination of measurements in all segments of the industry and a pricing schedule for quality in the end product, the new technology will bring about a dramatic change in pork quality. The real-time ultrasound processing system to be developed includes the computer hardware, software, and associated interfaces to enable scanning and prediction of % IMF in either live swine or hot carcasses. The Phase I research proved that current research models can be enhanced for the prediction of % IMF in swine. Objectives of the full R&D program are to make additional refinements to the prediction model for accurately characterizing % IMF levels in the pork ld muscle tissue in live animals and hot carcasses. The project will develop computer processing algorithms required for the system to evaluate hot carcasses. The project will assemble and demonstrate an operational system in the packing plant and will develop documentation to assist in the commercialization of the product to swine breeders and the packing segment of the industry. APPROACH: A major goal in Phase II will be to increase the repeatability of predicting the percent of intramuscular fat (% IMF) and further reduce the root mean square error (RMSE) of prediction in both live animals and carcasses. The live animals and carcasses will be scanned using an Aloka SSD 500V real-time ultrasound scanner with both the 12cm and 18cm linear array transducers. The company will subcontract with Iowa State University to supply a portion of the live animals and will partner with one or more packing plants for the hot carcass segment. Algorithms for calculating variations in ten texture parameters from ultrasound images will include histogram, gradient, Fourier, and co-occurrence techniques. The optimal prediction model will be implemented in the real-time ultrasound processing system to be developed. Increasing the repeatability of the prediction model within animal and within carcass will involve studying the reduction in the standard error of the measurement (% IMF prediction) across images within an animal and within the image. General linear model analyses will be conducted to develop alternative regression models for predicting % IMF in live swine and carcasses using texture parameters identified. The images from the live animals and carcasses will be randomly sampled to develop the alternative models using actual % lipid as determined through chemical extraction as the independent variable. The images not chosen for model development will be used for model validation. The statistical tools for analysis will include linear, nonlinear, and transformation procedures to determine the best fit of the data. Distribution of residuals will be studied to determine if transformations are appropriate. The primary statistic for determining the best model will be the RMSE. Other statistics that will be used will be R-Square, Mallows Cp statistic, and Pearson product-moment correlation. The processing algorithms will be driven by the need to be near real-time to keep up with chain speeds in the packing plant. A pilot study will be conducted to test the % IMF prediction accuracy on hot carcasses. On each of two separate scanning dates, a large sample of carcasses will be manually scanned with the technology. Results from the pilot study will be used to develop in-plant training and system maintenance and servicing procedures. Procedures to test individual scanning technician aptitude and dexterity for this type of work will be developed. Quality control procedures will be developed that will provide the needed feedback on individual scanning technician performance. The project will develop the first functional prototype system will include an Aloka SSD 500V ultrasound scanner, either a 12cm or 17cm linear array transducer, software, interfaces, and image processing computer. Plant personnel will undergo an intensive training program in the system operations and scanning protocol. Operational tests in the plant will be conducted. An economic impact study will be undertaken to define issues associated with commercialization of the system.