AIR2-CT94-1416 Developing Analytical Methods from Fluorescence Spectra and Video Images with Relevance to Food and Non-Food Agro-Industrial Engineering Using New Computer Based Evaluation Techniques (AFFLUENCE) - Final Report

Contacts
Summary Information



To find similar Items, click on a keyword below:
AIR Cluster V - Speciality Chemicals : Fibre : Fine Chemicals : Straw : Vegetable Oil/Fat

Results
The following results were obtained.
Establishment and development of hardware

• Testing standard laboratory spectrofluorimeters (KVL, NTUA, INRA, SFK, UMS and Vioryl)
• Testing a multichannel video imaging low magnification microscope ("DIPIX" by KVL and UMS) and a similar high magnification microscope (INRA and NTUA).
• Establishing and testing a multichannel video imaging CCD camera, including an image board (INRA, sfim ODS, CETIOM) (Bertrand et al. 1996)
• Testing a new video imaging spectrograph which allows spacial spectral landscape through software binning (KVL, SFK, UMS) (Brøndum et al. 1999a)

Software development

• Introduction of multiway analysis (KVL) such as PARAFAC (Bro 1998, Andersson and Bro 1998). Thanks to the specificity of the fluorescence phenomenon it was possible to deconvolute the characteristic excitation and emission spectra of the underlying chemical components from a complex fluorescence landscape entirely by mathematics. The results were checked by chemical separation (chromatograpy).
• It was found that chemometric software could be used to standardize and calibrate spectrofluorimeters (KVL) (Nørgaard 1995a,b).
• Newly adapted discriminant analysis algorithms supported the evaluation of data from the multichannel CCD camera. For many types of samples only a few pictures at different excitation wavelengths were needed for differentiation (INRA) (Baldwin et al. 1999).
• A special software for homogeneity was developed in which the unique capabilities of the imaging spectrograph to make spacial fluorescence landscapes was exploited (KVL, SFK) (Brøndum et al 1999a).

Applications
The basic prenormative research described above gave a platform for a range of applications which gave indications that the fluorescence option could be developed as a valid complement to current near infrared screening methods. Fluorescence spectroscopy was used in monitoring colour precursors in a sugar process (KVL, Munck et al. 1998), oxidation in fish and meat (SFK, Brøndum et al. 1999b,c), flour (UMS, Vioryl), soap (Vioryl) and in frying oil (KVL, Engelsen 1997). Multichannel fluorescence imaging (INRA, Bertand et al. 1996, Novales et al. 1996, Baldwin et al. 1999) was used to differentiate between seeds, including weeds (CETIOM), wheat flour of varying quality, yeast and dough. Fluorescence spectroscopy and imaging could distinquish between the different botanical components separated in wheat mills (UMS) and follow the extraction of lignin in the paper process (NTUA, Billa et al. 1999) and to certify the purity (humus) of drinking water and follow the maturation of cheese (KVL student projects). It was concluded that fluorescence is much more sensitive than most other spectroscopic techniques and that it should be considered not generally, but specifically applicable screening method. The original endeavour to develop fluorescence analysis combined with chemometrics on the pre-competitive level analogous with the near infrared technique has thus been highly successful in establishing a firm basis regarding software and hardware specifications as well as a rationale of how to search for further and improved applications.

Discussion
The work provided in this EU-project clearly demonstrates the potential of fluorescence spectroscopy and imaging combined by chemometrics and discriminant analysis in practice in industry and in the research laboratory. The potential which should be exploited includes great benefits such as very high specificity and sensitivity, but also limitations.

Some of the latter are inherent in the method e.g. few chemicals fluoresce. Others limitations might be overcome e.g. the presently available instrumentation which is not optimized for practical use outside the research laboratory (e.g. too small measurement area and sensitivity to harsh environmental conditions "on-line").

Spectrofluorimeters are difficult to standardize because of reference to zero light and because the spectrum of the xenon lamp used for illumination changes with time. KVL has shown that it is possible by regular calibration to overcome these limitations with chemometric software (Nørgaard 1995a,b,c; Nørgaard 1996).

Technology sometimes relies solely on indirect, only empirically validated control methods without checking on the underlying mechanisms. This is, however, unacceptable for science. In this project we have proven with chromatography (Munck et al. 1998) that, because of the unique specificity of fluorescence excitation/emission landscapes, it is possible to employ an exploratory strategy to identify up to seven excitation and emission spectra of the underlying chemical entities solely by mathematical separation using the PARAFAC algorithm (Andersson and Bro 1998). This holds a great potential for the future of science as well as technology. Now, in specific cases it is possible to explore matrices by directly evaluating fluorescence landscapes with PARAFAC and to obtain a first indication of the underlying fluorescent compounds, which of course must be checked. Bro (1998) has demonstrated the universal applicability of the PARAFAC algorithm in dividing a multitude of incoming radio signals to an antenna into their original components.

In principle, the project has also proven (INRA) that with astonishly few exposures it is possible to exploit multichannel imaging (Betrand et al. 1996) by discriminant analysis in an artificial learning process in both macro (Novales et al. 1996) and micro (Baldwin et al. 1997) evaluation. The project has also sponsored the development of an imaging card (sfim ODS) to speed up data aquisition which is fundamental for the future competitiveness of employing multichannel imaging "on-line" in industry. It has also been shown that multichannel imaging and discriminant analysis is an efficient tool for the scientist e.g. in evaluating microscopical preparations. However, in order to obtain positive results an in-depth knowledge of microscopy, including sample preparation, is nessesary. This also applies to the "macro" tests, indicating that there are often enough naturally labelled differences in fluorescence to discriminate between the natural objects tested.

Basic research in the laboratory in designing a workable prototype for multichannel fluorescence imaging has been successful in principle (INRA). All this is very promising. However, most of the work still remains, if the final target is to launch an industrial application. There is an obvious need for an apparatus for the automatic analysis of seed purity where the different contaminants could be defined as well as for machines which can discriminate for more valuable products. But we can conclude that fluorescence should be launched not as a general method but for carefully selected specific applications. These could, for example, easily be screened for by spectrofluorimetry and evaluated with the PARAFAC software (KVL), providing an indication of which fluorophores are in work.

We have also found that multichannel imaging can be complemented with new equipment in the form of a CCD imaging spectrograph (Brøndum et al 1999a). Within a fraction of a second it can take both an image and a characteristic spacial spectral landscape which amplifies possibilities for very specific studies e.g. for industrial "on-line" homogeneity monitoring for several fluorophores simultaneously.

A wide range of applications with regard to food and non-food technology have been explored by the participants in model trials in the laboratory and with samples from industry. These have brought forth positive options to be further explored but also an awareness of the limitations of the techniques.

In particular, the use of fluorescence spectroscopy and imaging for monitoring the botanical and chemical composition of wheat flour streams in flour mills has been studied in numerous trials. These experiments have had a great value in comparing widely different equipment. The answer is unanimous. In principle, it is possible to use fluorescence spectroscopy and imaging to monitor e.g. ash content in materials varying from 0.35 to 1.6% by all the applied instruments. But to be attractive for industry the sensitivity and precision have to be improved in order to monitor ash within a narrow scale such as 0.35 - 0.80%. Near infrared reflection (NIR), as explored by partner UMS is not ideal either with regard to sensitivity and precision, but it seems more versatile, because it may also simultanously monitor particle size by modelling scatter which is not possible with fluorescence. This gives valuable information to the miller on how to set the rollers.

An interesting observation by VIORYL is that storage of wheat flour can be monitored by fluorescence spectroscopy. VIORYL also has positive experience in monitoring oxidation processes of soap in order to help optimize raw materials to avoid odour development in the final products, as described earlier. KVL has confirmed this observation in the above mentioned study with vegetable frying oil for cooking. However, the fluorescence information regarding oxidation in this study does not come from peroxides as VIORYL suggests, but rather from aldehydes which may form Shiff's bases which are known to fluoresce and which are also detected by the anisidine number.

Another confirmation of the sensitivity of fluorescence in monitoring oxidation products is given by the warmed-over flavour (WOF) example (Brøndum et al. 1999c).

The rather complete industrial example of fluorescence monitoring from raw material to finished product in beet sugar processes given by KVL (Munck et al. 1998a) demonstrates the strength of the exploratory spectrofluorometric approach. Sugar and sugar production fluids are, however, very homogeneous products which facilitate the use of the currently used limited laboratory spectrofluorimeters which can only see through a very narrow window of approx. 1 x 9 mm.

Sugar is the purist of all food components marketed, having a typical purity of 99.999%. Colour formation is a serious problem in the sugar industry which has to be counteracted by SO₂ and CaO to prevent oxidation and formation of reducing sugars by inappropriate pH. Fluorescence is able to monitor fluorescent amino acids and phenols which may potentially react with small amounts of reducing sugar available and produce coloured melanoidines and melanines which also fluoresce. The sugar industry could be interested in introducing fluorescence for monitoring, if it could be done in the process. One is hesitant, because the problem of calibration will require time and money as well as specially trained chemometricans and because there is a lack of suitable "on-line" instruments on the market.

In traditional chemical analysis one starts by defining the hundreds of chemical substances involved in a process, as was done for the beet sugar industry by Madsen et al. 1978 in order to understand colour formation. If the target hypothesis is to find easily identifiable indicator substances, we suggest that our exploratory inductive method by introducing a multivariate screening method in the global area of the sugar factory would be more economical than a deductive strategy based only on a priori chemical knowledge, chromatography and classical statistics as studied in the local area - the research laboratory (Munck et al. 1998). This experience should be able to be generalized for other industrial applications.

The fundamental work on fish (KVL and SFK) demonstrates in a preliminary study that fish deterioration could be monitored by spectrofluorimetry and storage time could be predicted. This is a positive indication of the feasibility of future development. It needs optimization of the equipment in making it more handy and precise. It would also demand a very time-consuming calibration work to make a library with fluorescence signatures e.g. for cod with known size, catch place, time of year, temparature profile, time and temperature of storage, etc.. The samples must be analyzed biochemically and micro-biologically in order to understand how the screening method might work.

The spectroscopic work on meat (SFK) (Brøndum et al. 1999b,c) is difficult because of the non-homogeneity of the product and the instruments which are inadequate for the purpose. The correlation coefficients for quality parameters are lower with fluorescence than NIR and H-NMR, but we should remember that the spectrofluorimeter measures only 1 x 9 mm, NMR a circle with a diameter of 20 mm and the H-NMR apparatus a plug with a diameter of 18 mm and a length of 20 mm. The trials are interesting enough to be repeated, when suitable spectrofluorometric equipment for measuring larger areas become available. The imaging spectrograph principle (Brøndum et al. 1999a) could be used, but the instrument at KVL is too clumsy and sensitive to be moved to a slaughter house. The target for pork should be directed toward the prediction of pH drop 24 hours post mortem by measurements made immediately (5-10 minutes) after slaughter. This will also take care of drip loss and the problem of warmed-over flavour (WOF) when storing cooked meat.

The fluorescence studies which have been done (NTUA) with lignocelluloses validated with classical chemical analyses and advanced P-NMR methods clearly show that spectrofluorimetry evaluated by chemometrics is ideal for following extraction processes for lignin in the laboratory in the black liquor as well as on raw material, pulp and paper (Billa et al. 1999). There should be several applications ripe for industrial practice. The PARAFAC algorithm should also be applied to better define the fluorophores at work in situ. KVL has positive experience from student work using fluorescence to characterize municipal water (Christensen and Pedersen 1995) and cheese (Munck et al. 1999). Fluorescence spectroscopy could differentiate between surface (lake) water and deep well water. Fluorescence information could predict the content of humus (r = 0.96) as measured by the permanganate oxidation test. Humus mainly consists of degraded lignin skeletons. Monitoring water quality with spectrofluorimetry and evaluating with the PARAFAC software should be the future of water control with regard to regular consumer water and polluted industrial/sewage water.

Another remarkable feature of whole spectra fluorescence screening is its specificity by providing well-defined indicator substances. In the DIPIX fluorescence microscope study (UMS) with wheat flours mentioned above the DIPIX microscope plus the incorporated simple software could not model a narrow range of ash variation in the region 0.38 - 0.58% in a sample of ll7 samples from nine different mills (Mikkelsen 1996). However. in a PCA diagram there was a clear difference, just as in the sugar factory case described above, between the content of aleurone and pericarp between the different mills. In a study of 98 cheeses of the same type analyzed by spectrofluorimetry we have seen a clustering effect in a PCA plot due to production type and site (Munck et al. 1999) which tentatively can been associated with the addition of specific enzymes. In these experiments fluorescence was also shown to work as an indicator of maturation. The fluorescence diminishes when the maturation process proceeds, implying protein hydrolysis. The fluorophore, tryptophane, is likely to be involved here. KVL and INRA also have positive experience in using spectrofluorimetry to predict molecular structure and enzyme activity which could be of great interest to the pharmaceutical industry.

Exploitation
Development is hampered by the lack of suitable spectrofluorometers and imaging devices which can monitor large areas. An EU Opus project IN309051 with the aim of the improving industrial spectrofluorimetry was thus started in 1999 (KVL). There is also a lack of chemometric training and understanding of the exploratory, reversed engineering approach. This implies first measuring and then evaluating in order to build up a library of information to be used by the computor as a basis for artificial intelligence (see information on chemometrics at http://newton.mli.kvl.dk). It took 30 years to establish the currently very active near infrared screening technology. There has never been a greater gap between the potential of the new screening technology and its use today, considering the numerous other ranges in the spectrum which remain unexploited. The limiting factor is that cooperation and investments are needed - in this case from very different sources - to make a functioning chain. It is the combination of hardware, software, application knowledge and basic research which will forward the new technology of direct instant analysis e.g. exploring the world with fluorescence spectroscopy and imaging as our expanded senses. The European Union has an important role to play in catalyzing this development.

Conclusion
This EU project has shown that fluorescence spectroscopy and imaging combined with new software and applications pioneered by the research of the project has a great potential in developing new screening tools for food and non-food production chains. This would be a valuable supplement to the near infrared spectroscopy now available on the market.

The limiting factor for use of fluorescence analysis in industry is that suitable robust instruments with optimized sampling devices are generally unavailable. The interest of industry in developing such instrumentation seems small, because mechanics and optics are just a minor part of the development costs. The new chemometric and discriminant software requires in each case the establishment of a spectral library with the appropriate calibrations to relevant physical and chemical analyses for establishing a source for artificial intelligence.

This can only be done in a collaborative effort by several partners, including instrument manufacturers and laboratories from industries where the application is to take place as well as with researchers from universities and branch institutes who are knowledgeable in software and in applications in general. The economic coordinative incentive is lacking because such a difficult process involving many specialised partners is difficult to manage.

There has never been a larger gap between the use of technology in practice on one hand and the great potential for using modern technology to explore and control real processes outside the research laboratory on the other. The coordination problem seems too great even for large companies and national states who wants to see fast cash.

We would therefore recommend that EU take new initiatives in new programmes to support developments in artificial intelligence and computer technology in combination with new instrumentation for control of the positive and negative sides of modern technology as it rapidly evolves in industry, society and nature.

References
Andersson, C.A. (1999). Direct orthogonalisation. Chemometrics and Intelligent Laboratory Systems, 47, 51-63.
Andersson, C.A. and Bro,R. (1998). Improving the speed of multiway algorithms. Part I:Tucker 3. Chemometrics and Intelligent Laboratory Systems, 42 (1-2),193-103.
Baldwin, P. Bertrand,D., Novales,B., Bouchet, B. and Collobert, G.G.D.J. (1999) Chemometric labelling of cereal tissues in multi-channel fluorescence microscopy images. Analytical Chemistry, 69 (21), 4339-4348.
Billa,E., Argyropoulus,D.S. and Koukios, E.G. (1999) Recent advances in residual kraft lignins characterisation combining P-NMR and fluorescence spectroscopy by chemometrics. Progress in analytical methodologies applied to ligno-cellulosic material, Tappi Press. In press,
Bertrand.D., Novales,B., Baldwin, P.M. Bouchet, B. and Gallant, D.J. (1996). Multichannel imaging of cereal products. Proceedings of the 46th Australian Cereal Chemistry Conference,Sydney, 4-6 September, 1996. C.W. Wrigly (ed.) Grain Research and Development Corporation publ, Canberra (Aus.) 87-91.
Bro,R. (1998). Multiway analysis in the food industry: Models algorithms and applications. Thesis. Univ. of Amsterdam and Royal Vet. and Agric. Univ., Frederiksberg, Denmark. pp 228.
Brøndum,J., Munck,L. Larsen, E. and Zappey, H. (1999a). Imaging spectroscopy applications in food quality. Food Technology, The Royal Vet. and Agric. Univ. Frederiksberg. Denmark. Manuscript pp.28.
Brøndum,J., Engelsen,S.B., Munck,L., Henkel, P., Karlsson,A. and Tornberg, E. (1999b). Prediction of water holding capacity and composition in porcine meat with spectroscopy. IBID. Manuscript pp 17. To be submitted to "J. of Food Control".
Brøndum, J., Byrne, D.V., Bak,L.S., Bertelsen, G. and Engelsen S.B. (1999c). Warmed-over flavour in porcine meat - a combined spectroscopic, sensory and chemometric study. IBID. Manuscript pp37. To be submitted to "Meat Science".
Engelsen, S.B. (1997). Explorative spectrometric evaluations of frying oil deterioration. JAOCS. 74 (12).
Fulcher, R.G. (1989). Fluorescence microscopy applications in food analysis. In: Fluorescence analysis in foods.
L.Munck (ed.). Longmans Scientific, 59 -109.
Harschman, R. (1970). Foundations of the PARAFAC procedure: Models and conditions for an exploratory multimodal factor analysis. UCLA working papers in phonetics. 1 - 84.
Jensen, S.Aa., Renberg, S. and Munck, L. Fluorescence analysis in fish and meat technology. In Fluorescence analysis in food. L. Munck(ed.) Longmans Scientific, 181-192.
Madsen, R.F., Kofod Nielsen, W. and Winstrøm-Olsen, T.E. (1978). Formation of colour compounds in production of sugar from sugar beet. Sugar Technology Reviews, Elsevier, Amsterdam, 6, 49-115.
Mikkelsen, E. (1996). Prediction of baking performance of commercial wheat flour using explorative data analysis. Master thesis. Food Technology. Dept. of Dairy and Food Science, Royal Vet. and Agric. Univ., Frederiksberg, Denmark, pp.66.
Munck, L. (1989a). Fluorescence analysis in foods. L.Munck (ed) Longmans Scientific, pp.289.
Munck, L. (1989b). Practical experiences in the development of fluorescence analyses in an applied food research laboratory. In: Fluorescence analyses in foods. L. Munck (ed.) Longmans Scientific, 1-32.
Munck, L. Nørgaard, L., Engelsen, S.B., Bro, R. and Andersson, C.A. (1998). Chemometrics in food science - demonstration of the feasibility of a highly explorative, inductive evaluation strategy of fundamental scientific significance. Chemometrics and Intelligent Laboratory Systems, 44, 31 - 60.
Munck, L., Nørgaard,L., Bro,R., Andersson, C.A., Wittrup, C. and Hansen, P.W. (1999). New chemometric multivariate methods for an overview of food processes. In: Proceedings of 25th International Dairy Congress, September 21-24, Aarhus, Denmark, 372 - 382.
Novales, B., Bertrand, D., Devaus, M.F. and Roberts, P.S.A. (1996). Multispectral imaging of food products. Journal of the Science of Food and Agriculture, 71, 376-382.
Nørgaard, L. (1995a). Direct standardisation in multi-wavelength fluorescence spectroscopy. Chemometrics and Intelligent Laboratory Systems, 29, 283-293.
Nørgaard, L. (1995b). A multivariate apporach to fluorescence spectroscopy. Talanta. 42, 1305-1324.
Pedersen, B. and Martens, H. (1989). Multivariate calibration of fluorescence data. In: Fluorescence Analysis in Foods. L.Munck (ed.). Longmans Scientific, 215 - 267.
Williams, P.C. and Norris (eds.)(1987). Near Infrared Technology in Agricultural and Food Industries. Amer. Assoc. Cereal Chem., St.Paul,MN, USA.