JBS 0 Murdoch SCOTT mla UNIVERSITY automation+robotics MEAT LIVESTOCK AUSTRALIA final report Project code: A.TEC.0124 Prepared by: Jonathan Cook,Merv Shirazi Scott Automation and Robotics Graham Gardner Murdoch University Date published: 15 November 2016 PUBLISHED BY Meat and Livestock Australia Limited Locked Bag 1961 NORTH SYDNEY NSW 2059 X-Ray OCM Bone,Fat and Muscle Trials Final Report Meat Livestock Australia acknowledges the matching funds provided by the Australian Government to support the research and development detailed in this publication. This publication is published by Meat Livestock Australia Limited ABN 39 081 678 364(MLA).Care is taken to ensure the accuracy of the information contained in this publication.However MLA cannot accept responsibility for the accuracy or completeness of the information or opinions contained in the publication.You should make your own enquiries before making decisions conceming your interests.Reproduction in whole or in part of this publication is prohibited without prior written consent of MLA
Project code: A.TEC.0124 Prepared by: Jonathan Cook, Merv Shirazi Scott Automation and Robotics Graham Gardner Murdoch University Date published: 15 November 2016 PUBLISHED BY Meat and Livestock Australia Limited Locked Bag 1961 NORTH SYDNEY NSW 2059 X-Ray OCM Bone, Fat and Muscle Trials Final Report Meat & Livestock Australia acknowledges the matching funds provided by the Australian Government to support the research and development detailed in this publication. This publication is published by Meat & Livestock Australia Limited ABN 39 081 678 364 (MLA). Care is taken to ensure the accuracy of the information contained in this publication. However MLA cannot accept responsibility for the accuracy or completeness of the information or opinions contained in the publication. You should make your own enquiries before making decisions concerning your interests. Reproduction in whole or in part of this publication is prohibited without prior written consent of MLA. final report
Abstract An Automated Beef Rib Cutting system has been developed by Scott Automation Robotics (SCOTT)and is currently in production.This system utilises dual-energy x-ray (DEXA) hardware to drive automated cutting of beef carcases.There is currently a need in the industry for methods to objectively measure carcase characteristics for the purposes of grading.DEXA technology is a key enabler for this. The purpose of this project was to investigate the ability of this system to accurately perform objective carcase measurement (OCM)on beef sides for fat,lean and bone composition.A trial was first performed using a calibration object made from known compositions of fat and lean.These trials suggested that the system was capable of obtaining OCM data.A set of trials was then performed whereby six beef sides were scanned by the DEXA system and then by a CT scanner.From this,the DEXA images were analysed and models were built to predict the amount of lean,fat and bone present in each DEXA image.The CT data provided predictions for the amount of lean,fat and bone in each carcase side. These trials yielded promising results and a second set of trials was designed to build upon these findings.Another set of phantom trials were performed and a further eight sides were then scanned by the DEXA system,CT scanned and modelled as before.The modifications resulted in improved models with R2 values of 0.78 and 0.93 achieved for fat and bone, respectively.Alternatively there was no ability to predict CT lean%directly,although this can be calculated from the other two measures. This data demonstrates good potential for measuring carcase composition using DEXA values.The next phase of work should involve confirming these results within an expanded data set,while also testing the stability of this measurement across a variety of processing factors. Page 2 of 44
Page 2 of 44 Abstract An Automated Beef Rib Cutting system has been developed by Scott Automation & Robotics (SCOTT) and is currently in production. This system utilises dual-energy x-ray (DEXA) hardware to drive automated cutting of beef carcases. There is currently a need in the industry for methods to objectively measure carcase characteristics for the purposes of grading. DEXA technology is a key enabler for this. The purpose of this project was to investigate the ability of this system to accurately perform objective carcase measurement (OCM) on beef sides for fat, lean and bone composition. A trial was first performed using a calibration object made from known compositions of fat and lean. These trials suggested that the system was capable of obtaining OCM data. A set of trials was then performed whereby six beef sides were scanned by the DEXA system and then by a CT scanner. From this, the DEXA images were analysed and models were built to predict the amount of lean, fat and bone present in each DEXA image. The CT data provided predictions for the amount of lean, fat and bone in each carcase side. These trials yielded promising results and a second set of trials was designed to build upon these findings. Another set of phantom trials were performed and a further eight sides were then scanned by the DEXA system, CT scanned and modelled as before. The modifications resulted in improved models with R2 values of 0.78 and 0.93 achieved for fat and bone, respectively. Alternatively there was no ability to predict CT lean% directly, although this can be calculated from the other two measures. This data demonstrates good potential for measuring carcase composition using DEXA values. The next phase of work should involve confirming these results within an expanded data set, while also testing the stability of this measurement across a variety of processing factors
Executive Summary An Automated Beef Rib Cutting system has been developed by Scott Automation Robotics (SCOTT)which is currently in production in an Australian beef abattoir.The system utilises a dual-energy x-ray(DEXA)system in order to identify cut placement for that carcase.This system consists of separate source-detector pairs for each of the low-energy and high- energy x-ray images.These two images are then stitched together into one DEXA image. There is a need in the red meat industry to move towards methods of measuring carcase attributes in an objective manner.DEXA is one technological enabler for such measurement.Utilising DEXA technology for both automation and OCM concurrently,in one integrated system,presents a number of benefits,particularly surrounding the cost-benefit of such a system. This project thus aimed to evaluate the feasibility in utilising a system which has been designed and built for beef automation for OCM tasks as well.It will also explore the hardware requirements and commercial considerations for designing such systems in the future as well as the suitability of dual-hardware DEXA systems for the application. The first task was to get an initial assessment of whether the hardware was capable of producing consistent values for OCM measurements.This was achieved by scanning a tissue phantom-an object consisting of homogenous blocks of lean and fat,at varying compositions,which have been tested for chemical lean.The scans were completed successfully and analysis suggested the system was capable of producing consistent enough x-ray values to enable OCM calculation. Six beef sides were then selected and scanned with the system.These sides were then cut up and scanned with a CT scanner.The CT data was then used to predict the amount of fat, lean and bone in each of the sides.The DEXA images were analysed to see if the information could be modelled to predict CT composition.A number of challenges were experienced however which prevented an accurate model to be generated.One factor contributing to this was an effect along the height of the detector.The detectors in the system are 2500mm long and thus have significantly different x-ray flux along their lengths Compensating for this improved results significantly.Another effect found was that thin tissue information(approximately 10mm and less)was saturated in the low energy image. In the work completed in lamb,these tissue depths are known to contribute significantly to the OCM models.The loss of such information thus impacted the results negatively. Another set of trials was then conducted whereby scans were taken at production currents as well as a current low enough to avoid saturation of the detectors.Phantom scans were first performed which vindicated the positive effect of running at these lower currents-the information in the thinnest phantom were now visible and demonstrating consistent results. Another eight sides were then DEXA scanned,CT scanned and analysed. The results of the analysis on the additional eight sides yielded better results,particularly for predicting bone content.Fat and lean content however were unable to be modelled with a significant level of accuracy.R2 values of 0.4,0.45 and 0.82 achieved for lean,fat and bone,respectively.A number of possible limitations have been identified which may explain why this system is not able to achieve accurate OCM.It is suspected that the alignment Page 3 of 44
Page 3 of 44 Executive Summary An Automated Beef Rib Cutting system has been developed by Scott Automation & Robotics (SCOTT) which is currently in production in an Australian beef abattoir. The system utilises a dual-energy x-ray (DEXA) system in order to identify cut placement for that carcase. This system consists of separate source-detector pairs for each of the low-energy and highenergy x-ray images. These two images are then stitched together into one DEXA image. There is a need in the red meat industry to move towards methods of measuring carcase attributes in an objective manner. DEXA is one technological enabler for such measurement. Utilising DEXA technology for both automation and OCM concurrently, in one integrated system, presents a number of benefits, particularly surrounding the cost-benefit of such a system. This project thus aimed to evaluate the feasibility in utilising a system which has been designed and built for beef automation for OCM tasks as well. It will also explore the hardware requirements and commercial considerations for designing such systems in the future as well as the suitability of dual-hardware DEXA systems for the application. The first task was to get an initial assessment of whether the hardware was capable of producing consistent values for OCM measurements. This was achieved by scanning a tissue phantom – an object consisting of homogenous blocks of lean and fat, at varying compositions, which have been tested for chemical lean. The scans were completed successfully and analysis suggested the system was capable of producing consistent enough x-ray values to enable OCM calculation. Six beef sides were then selected and scanned with the system. These sides were then cut up and scanned with a CT scanner. The CT data was then used to predict the amount of fat, lean and bone in each of the sides. The DEXA images were analysed to see if the information could be modelled to predict CT composition. A number of challenges were experienced however which prevented an accurate model to be generated. One factor contributing to this was an effect along the height of the detector. The detectors in the system are 2500mm long and thus have significantly different x-ray flux along their lengths. Compensating for this improved results significantly. Another effect found was that thin tissue information (approximately 10mm and less) was saturated in the low energy image. In the work completed in lamb, these tissue depths are known to contribute significantly to the OCM models. The loss of such information thus impacted the results negatively. Another set of trials was then conducted whereby scans were taken at production currents as well as a current low enough to avoid saturation of the detectors. Phantom scans were first performed which vindicated the positive effect of running at these lower currents – the information in the thinnest phantom were now visible and demonstrating consistent results. Another eight sides were then DEXA scanned, CT scanned and analysed. The results of the analysis on the additional eight sides yielded better results, particularly for predicting bone content. Fat and lean content however were unable to be modelled with a significant level of accuracy. R 2 values of 0.4, 0.45 and 0.82 achieved for lean, fat and bone, respectively. A number of possible limitations have been identified which may explain why this system is not able to achieve accurate OCM. It is suspected that the alignment
between the low energy and high energy pixels,while sufficient for the purposes of cutting, aren't sufficient enough to allow accurate OCM analysis.The other key limitation is that the x-ray system doesn't scan the entire carcase-it was only designed to image the carcase ribs and,thus,doesn't capture the hindquarter.Finally,while scanning at a lower current enabled more accurate OCM measurements,it also negatively affects the system's ability to perform cutting. A second analysis was then performed whereby the image the was truncated at the 13th rib for each of the carcases.This ensured that all datasets contained the same carcase information(the forequarter only).In this case the prediction of CT bone composition was excellent,with R2 values as high as 0.78,and 0.93 when cold carcase weight was included in the model.There was also good precision for CT fat%prediction with R2 values as high as 0.71,and 0.78 when cold carcase weight was included in the model.Alternatively there was no ability to predict CT lean%directly,although this can be calculated from the other two measures. This data demonstrates good potential for measuring carcase composition using DEXA values.The next phase of work should involve confirming these results within an expanded data set,while also testing the stability of this measurement across a variety of processing factors. Page 4 of 44
Page 4 of 44 between the low energy and high energy pixels, while sufficient for the purposes of cutting, aren’t sufficient enough to allow accurate OCM analysis. The other key limitation is that the x-ray system doesn’t scan the entire carcase – it was only designed to image the carcase ribs and, thus, doesn’t capture the hindquarter. Finally, while scanning at a lower current enabled more accurate OCM measurements, it also negatively affects the system’s ability to perform cutting. A second analysis was then performed whereby the image the was truncated at the 13th rib for each of the carcases. This ensured that all datasets contained the same carcase information (the forequarter only). In this case the prediction of CT bone composition was excellent, with R2 values as high as 0.78, and 0.93 when cold carcase weight was included in the model. There was also good precision for CT fat% prediction with R2 values as high as 0.71, and 0.78 when cold carcase weight was included in the model. Alternatively there was no ability to predict CT lean% directly, although this can be calculated from the other two measures. This data demonstrates good potential for measuring carcase composition using DEXA values. The next phase of work should involve confirming these results within an expanded data set, while also testing the stability of this measurement across a variety of processing factors
Table of Contents 1 Background.......................................................................................................6 2 Project Objectives.… 7 3Methodology8 3.1 DEXA Scans of Tissue phantoms8 3.2 DEXA and CT scanning of six beef sides...............................9. 3.3 Second set of phantom and carcase scans at production X-Ray levels and reduced X-Ray levels........ .13 4 Results/Discussion… 17 4.1 DEXAScans of Tissue Phantoms1 4.2 DEXA and CT scanning of six beef sides....... .20 4.3 Second set of phantom and carcase scans at production X-Ray levels and reduced X-Ray levels................................... 28 4.3.1 Tissue Phantom Analysis..28 4.3.2 Carcase Data Analysis (8 sides).................... .31 5 Conclusions/Recommendations.................. .38 Page 5 of 44
Page 5 of 44 Table of Contents 1 Background.........................................................................................................................6 2 Project Objectives...............................................................................................................7 3 Methodology .......................................................................................................................8 3.1 DEXA Scans of Tissue Phantoms ..............................................................................8 3.2 DEXA and CT scanning of six beef sides...................................................................9 3.3 Second set of phantom and carcase scans at production X-Ray levels and reduced X-Ray levels. ........................................................................................................................13 4 Results/Discussion ...........................................................................................................17 4.1 DEXA Scans of Tissue Phantoms ............................................................................17 4.2 DEXA and CT scanning of six beef sides.................................................................20 4.3 Second set of phantom and carcase scans at production X-Ray levels and reduced X-Ray levels. ........................................................................................................................28 4.3.1 Tissue Phantom Analysis ..................................................................................28 4.3.2 Carcase Data Analysis (8 sides) .......................................................................31 5 Conclusions/Recommendations.......................................................................................38