nondestructive analysis of salt, water, and protein in dried salted cod using computed tomography

E:Fo

odEn

ginee

ring&

Phys

icalP

rope

rties

JFS E: Food Engineering and Physical Properties

Nondestructive Analysis of Salt, Water,and Protein in Dried Salted Cod usingComputed TomographyT.T. HASETH, M. HØY, B. EGELANDSDAL, AND O. SØRHEIM

ABSTRACT: Computed X-ray tomography (CT) was used to determine NaCl, water, and protein levels in dried saltedcod. Cod fillets were salted and dried, and CT was conducted several times during the process. Also, homogenizedcod samples with a chemical composition covering the typical composition of cod during salting and drying wereproduced, and CT scanned. Chemical composition of fillet and model samples was predicted from CT images ac-quired at 80, 110, and 130 kV. The best average prediction errors (RMSECV) obtained for homogenized samples were0.6% NaCl, 1.3% water, and 1.5% protein; all explained variances were R2 = 0.99 or above. The best RMSECVs andexplained variances for cod fillet samples were 0.9% NaCl (R2 = 0.96), 0.8% water (R2 = 0.99), and 1.4% protein(R2 = 0.79). Combining CT values from 2 or 3 voltages gave the best predictions except when predicting salt in codfillet, where 1 voltage was sufficient.

Keywords: chemical composition, computed X-ray tomography, dried salted cod, nondestructive analysis, salt

Introduction

Salting has been used for preservation of cod for thousands ofyears (Horner 1997), and even though the preservation aspect

is less important today, products like salt-ripened cod and driedsalted cod (clip fish, bacalao) are still appreciated for their sen-sory properties. However, consumer trends, such as preferences forready-to-eat meals and low salt results in product development, forexample lightly salted fillets (Larsen and others 2008), and in im-provement of production technologies like salting (Barat and oth-ers 2002, 2003; Martınez-Alvarez and others 2005) and desalting(Barat and others 2004).

As new technologies and products are developed, a fast and non-destructive method for analyzing traits, like chemical compositionand salt distribution, would be useful for optimizing unit opera-tions, such as salting and desalting of cod. One such fast and non-destructive method could be computed X-ray tomography (CT),which is an imaging technique able to visualize structures of lowcontrast inside an object (Goldman 2007). The method measuresthe attenuation of X-rays sent through an object. The X-ray atten-uation is dependent in part on the density of the scanned object(Seibert and Boone 2005) and thus various tissues can be differen-tiated and, possibly, chemical components quantified.

CT is well-known within animal science for determination ofthe separation between fat and lean tissue as first described bySkjervold and others (1981). Within aquaculture, CT is used for pre-diction of body composition of various, mainly fatty, fish species(Gjerde 1987; Rye 1991; Romvari and others 2002; Kolstad andothers 2004). Use of CT within lean fish species is more limited;

MS 20080936 Submitted 11/21/2008, Accepted 1/13/2009. Authors Haseth,Høy, and Sørheim are with Nofima Mat AS, Osloveien 1, N-1430 As, Nor-way. Authors Haseth and Egelandsdal are with Dept. of Chemistry, Biotech-nology and Food Science, Norwegian Univ. of Life Sciences, P.O. Box 5003,N-1432 As, Norway. Author Haseth is also with Animalia—Meat and Poul-try Research Centre, P.O.Box 396 Økern, N-0513 Oslo, Norway. Direct in-quiries to author Haseth (E-mail: [email protected]).

however, CT has proved useful in determination of dry matter inAtlantic cod (Kolstad and others 2008).

By use of CT, the density differences between the major chemicalconstituents of dry-cured ham have been shown to be in the order:fat < water < protein < NaCl (Haseth and others 2007), which wasalso utilized to quantify the amount of NaCl. X-ray attenuation isenergy-dependent (Seibert and Boone 2005), and the energy spec-trum sent through a scanned object is determined by the voltagelevel set on the scanner. Scanning at 2 or 3 voltage levels rather than1 when predicting NaCl may result in high prediction accuracies,as shown both for dry-cured ham (Haseth and others 2008) andsalted, smoked salmon (Segtnan and others, 2009). To the authors’knowledge, CT has not been used to study salt and other chemi-cal compounds in salted cod, or to study heavily salted fish, neitherqualitatively nor quantitatively.

The objective of this study was to calibrate the CT scanner forquantitative and distributional analysis of salt in dried salted cod,and to investigate the possibility of CT to quantify water and pro-tein content in addition to salt. To optimize the determinations, thesamples were scanned at different voltages and the images prepro-cessed to remove nonmuscle pixels.

Materials and Methods

Both designed, ground cod model samples, and fillets fromfresh, salted, and salted/dried cod were used to make predic-

tive models of NaCl and water from CT values.

Cod model samplesA design of 45 homogenized cod model samples and 14 repli-

cates was created, with a composition spanning 0.1% to 67% NaCl,14% to 92% water, and 7% to 46% protein. The wide range was toinclude all conceivable chemical compositions during productionof dried salted cod, also with potential salt crystallized within themuscle. Sample size was 275 g. Five samples were removed dueto preparation difficulties. Images of samples with more than 30%

C© 2009 Institute of Food Technologists R© Vol. 74, Nr. 3, 2009—JOURNAL OF FOOD SCIENCE E147doi: 10.1111/j.1750-3841.2009.01102.xFurther reproduction without permission is prohibited

E:FoodEngineering&PhysicalProperties

Salt analysis in dried salted cod by CT . . .

NaCl were eventually kept out from the statistical analysis; sam-ples with such high salt content were not well measured by the CTscanner, probably because NaCl crystallinity resulted in a too highdegree of X-ray reflection. This left 35 samples distributed on 23design points; a summary of their reference values is given inTable 1.

To prepare the cod model samples, cod muscles (Gadus morhua)were ground twice in a grinder (Mohle FDR905, Th. Mohle Maschi-nenfabrik, Remscheid, Germany) and analyzed for water (81.8%),protein (19.1%), fat (0.6%), and NaCl (0.1%). Ground cod muscle,freeze-dried cod muscle (Christ Gamma 1-16LSC, Martin ChristGefriertrockungsanlagen GmbH, Osterode am Harz, Germany), salt(Norsal R© finely refined, GC Rieber Salt AS, Oslo, Norway), andwater were weighed in to obtain the designed chemical samplecompositions. The samples were blended thoroughly, filled in 13-cm wide polyamide bags, vacuum-packaged, and pressed. Thesamples were CT-scanned and the images analyzed as describedsubsequently.

Cod filletsSix gutted and cleaned farmed cods (Gadus morhua) stored on

ice, gutted weight 2.7 to 4.3 kg (Akvaforsk, Averøy, Norway) were fil-leted 2-d postmortem, providing 12 fillets. They were salted using 2commercial methods, “fast” and “slow.” With the fast method, thefillets were brine-injected to 15% weight increase (25% NaCl; WS80,Suhner AG, Bremgarten, Switzerland), brine-salted (25% NaCl) for2 d, and finally kench-cured (dry salting with drainage) (Horner1997) for 10 d. With the slow method, the fillets were pickle-salted(dry salting with brine formation) for 21 d before kench-cured for5 d. The fillets from each fish were assigned to one salting methodeach. Computed tomography was conducted regularly during theproduction process. Selected scanned cross-section volumes werecut out and analyzed chemically for water, fat, protein, and NaCl

Table 1 --- Summary of reference data for NaCl and fat incod model samples and fillet samples.

Min Max Mean SD N

NaCl in model samples (%) 0.1 30.7 12.9 10.5 35NaCl in fillet samples (%) 0.1 22.3 13.5 7.3 42

Water in model samples (%) 36.5 92.4 63.7 17.2 35Water in fillet samples (%) 51.7 81.4 63.0 10.7 42

Protein in model samples (%) 7.0 45.9 23.6 12.5 35Protein in fillet samples (%) 14.7 26.0 20.1 3.2 33

Values expressed as percentage by weight.

Figure 1 --- Image of a cod filletwith the 5 scan positions marked.The plastic needle markingposition 1 is placed under themarking of position 1 (to the left),barely visible.

after each scanning. In addition, 3 dried salted cod fillets werebought from a commercial producer, CT scanned, and analyzedchemically for NaCl, protein, water, and fat in the scanned volumes.In total, 42 samples from the fillets were analyzed (protein only in33 samples because of insufficient amount of sample). A summaryof the reference values for NaCl and water are given in Table 1.

Reference analysesReference analyses were conducted on the raw material for

model samples, and on the muscular part of the scanned cross-sections of fillets. Water content was determined as weight-loss af-ter drying at 105 ◦C for 16 to 18 h (NMKL 1991). Protein contentwas determined as Kjeldahl protein (NMKL 2003). NaCl content ofmodel sample fillet and dried salted cod was determined as solu-ble chloride by end-point titration with silver nitrate (model sam-ples [IDF 1997], dried fillets [NMKL 2004]). NaCl content of saltedcod was determined as soluble chloride by ion chromatography(NSLAD 1969). The fat content was set to 0%. The chemical compo-sition of each sample was recalculated to amount to 100%.

CT scanningAll samples were sequentially CT scanned (Siemens Somatom

Emotion, Siemens AG, Erlangen, Germany). Scanning protocols aregiven in Table 2. The samples were CT scanned at the 3 availablevoltages. The 2 scan positions on the homogenized model sampleswere 7.5 mm on each side of the center of the sample. The modelsamples were scanned once; the 35 samples that were made useof resulted in 210 images. The 5 scan positions on the cod filletsare shown in Figure 1. A plastic needle marked the position of the1st scan; the 4 subsequent scans were taken at 5-cm intervals. Thesalted fillets were scanned 6 times during processing, providing 90images of each fillet and 1080 images in total. The commercial driedsalted cods were scanned once, providing 45 images.

Table 2 --- CT protocols used for scanning cod.

Current (mA) 106

Voltage (kV) 80, 110, 130Scan time (s) 1Convolution kernel B50SSlice with (mm) 10Resolution (pixels/mm) 3.4a/2.6b

Nr of scan positions 2a/5b

aModel samples.bFillet samples.

E148 JOURNAL OF FOOD SCIENCE—Vol. 74, Nr. 3, 2009

E:Fo

odEn

ginee

ring&

Phys

icalP

rope

rties


CT measures attenuation in a 3-dimensional slice of tissue. Themeasurements are preprocessed and reconstructed by instrumen-tal software and displayed as a 2-dimensional CT image. The CTimage has 512 × 512 pixels; each pixel has a gray value given inHounsfield Units (HU) (Hounsfield 1980). The CT value expressesthe X-ray attenuation in the corresponding volume element (voxel)in the scanned object.

Image analysesThe acquired CT images were of DICOM format, 16 bit, where

the 12 bit of image data has 212 = 4096 values, corresponding to CTvalues ranging from –1024 (air) to 3071 HU. The CT images were im-ported into MATLAB (MATLAB version 7.4 R2007a, The MathworksInc., Natick, Mass., U.S.A.) for image analyses.

For cod model sample images, a mask was created to remove thebackground and sample edge, as proposed by Haseth and others(2008). Frequency histograms were generated for each masked im-age, with range from –25 HU to 1774 HU and histogram intervalsof 5 HU. The HU range covered all fish tissues at all productionstages. Mean CT value was calculated from the relevant range ofeach frequency histogram. The mean CT values were correlated tothe chemical composition of the same volume.

For cod fillet images, a mask removing pixels containing air (dueto for example gaping) and bone, and also the background andsample edge including skin, was created. Frequency histograms ofHU = (0, 550) were generated with histogram intervals of 5 HU, andmean CT value was calculated.

Statistical analysesThe obtained CT values were used to make predictive models for

NaCl, water, and protein in cod model samples and in cod fillets,using the statistical software The Unscrambler R© version 9.6 (CAMOSoftware AS, Oslo, Norway). Predictors were mean CT values at 1, 2,or 3 voltages. The calibration of mean CT values against referencevalues was performed using partial least squares (PLS) regression(Martens and Næs 1989). In multiple linear regression (MLR), lowcorrelation between variables is an assumption; however, CT valuesat different voltages are highly correlated. PLS regression handleshighly correlated variables by summarizing them in possibly fewerprincipal components (PCs). In addition, there is the possibility forgraphical interpretation using PLS regression.

The calibration models were validated using cross-validationwith parallel samples kept in the same validation segment. Theadded amount of NaCl, yi, and the predicted NaCl content, yi, of thevalidation samples, were used to calculate prediction errors of thecross-validated calibration models, expressed as root mean squ-are error of cross-validation (RMSECV). The RMSECV is defined as:

RMSECV =√

1n

n∑i=1

(yi − yi )2

where i denotes the number of calibration samples from 1 to n. RM-SECV measures the prediction error in the same units as the origi-nal response variable.

Results and Discussion

Density changes in salted codAs CT values are a measure of density (Hounsfield 1980), the dis-

tribution of CT values in a cod sample is related to its chemicalcomposition. During production of dried salted cod, salt diffusesinto and water out of the cod (Thorarinsdottir and others 2004),and such a change in chemical composition can be reflected in a

change in the distribution of CT values (Haseth and others 2007).Statistics for CT values in model and fillet samples are given inTable 3. The effect of pickle-salting on the distribution of CT valuesin cod (at 110 kV) is demonstrated in Figure 2. The fresh, unsaltedfillet had a mean CT value of 64 HU, which is somewhat lower thanreported by Kolstad and others (2008). The difference could be dueto both the image analysis and use of different CT scanners. After1-d pickle-salting, mean CT value had increased to 151 HU, after6 d to 315 HU, and after 21 d to 368 HU. The CT values would typi-cally be 10% to 20% higher at 80 kV and 5% to 10% lower at 130 kV,as previously stated for pork (Haseth and others, 2008).

The fresh fillet had a narrow range of CT values (Figure 2). Duringpickle-salting, the salt diffused into the fillet from the surface of thesample. This resulted in a salt gradient from the surface towards thecenter of the fillet during the first days of salting, which can be seenas a long “tail” for the fillet salted for 1 d in Figure 2. The gradientwas reduced as the salting proceeded.

Quantification and distribution of NaClBjørkevoll and others (2004) raised concerns for difficulties in

prediction of salt distribution and final salt content in ready-to-use rehydrated clip fish. Both distributional and quantitative issuescan be overcome with a proper CT-salt calibration. In this study,calibration models for salt were calculated for both homogenizedmodel samples and fillet samples. The prediction error of NaCl infillet samples was independent of which and how many voltageswere used for prediction. All (1 to 3) voltages were significant; how-ever, only one (out of 1 to 3) principal component was used in theregression model. This 1st PC separated on salt content. The 2ndPC separated out the commercial dried salted samples, and the3rd PC separated the brine-injected samples at the time just afterinjection. Average prediction errors were 1.4% to 1.5% NaCl, whichamounts to 6.3% to 6.7% of the salt range, and the explainedvariances were R2 = 0.96. It should be noted that the chemical

Table 3 --- Summary of CT values (HU) in model and filletsamples used for calibration.

Min Max Mean SD N

80 kV in model samples 40.1 721.7 336.9 223.6 3580 kV in fillet samples 63.5 480.5 346.3 157.5 42



0

200

400

600

800

1000

1200

1400

-50 50 150 250 350 450

CT value (HU)

Fre

qu

en

cy (

nr

of

pix

els

)

Unsalted fillet

Salted

1 day

Salted

6 days

Salted

26 days

Figure 2 --- Frequency distributions of CT values at 110 kVin a cross-section of a cod fillet during salting. The filletis the same as displayed in Figure 4.

Vol. 74, Nr. 3, 2009—JOURNAL OF FOOD SCIENCE E149



distribution in fillet samples was not even, and that there was apredominance of highly salted samples. In another study, aroot mean square error of prediction (RMSEP) of 0.34% NaCl (10%of the salt range) and R2 = 0.90 was achieved for salted/smokedsalmon with low salt content, with fat content predicted from NIRincluded in the model (Segtnan and others 2009). According toHaseth and others (2008), a lean system is comparable to a larger-fat-range system where fat content is known.

The salt content increased most during the first 6 days of salting,only slightly the next 6 days, and then it seemed to slowly equili-brate when reaching approximately 15% NaCl, depending on thesalting method. Prediction of salt in samples ranging from freshuntil just reaching salt ripened was more accurate than predic-tion in all samples, providing R2 = 0.98 and RMSECV = 0.9% NaCl(5.6% of the salt range). As before, all (1 to 3) voltages were sig-nificant; however, only one (out of 1 to 3) principal componentwas used in the regression model. The reason for the higher accu-racy should be that after reaching salt ripened, the HU value maychange in salted samples, but not necessarily due to increased saltcontent.

With homogenized model samples, average prediction errors(RMSECVs) of 2.3% to 2.8% NaCl and R2 = 0.92 to 0.95 were ob-tained using 1 voltage. The results improved significantly using 2 orall 3 voltages, reducing the RMSECVs to 0.6% to 0.7% NaCl, amount-ing to 2% of the salt range, and an explained variance of R2 = 0.996.In comparison, RMSEPs of 0.3% to 0.5% NaCl was found in groundpork using 2 or 3 voltages when fat content was included in themodels (Haseth and others, 2008), corresponding to 2% to 3% oftheir salt range.

In Figure 3, predicted values for NaCl are plotted against refer-ence values for the best models provided all analyzed samples weremodeled. The prediction accuracy was higher in the homogenizedmodel system (A) than in the fillet samples (B). According to the cal-ibration equation for salt in fillet samples (Figure 3B), 1% salt equals18.2 HU. In comparison, an increase of 16.5 HU corresponded to a1% increase in NaCl in fresh pork (Haseth and others 2007). Thecalibration equation for CT values in salt in cod acquired at 110 kV(HU110) (Figure 3B) was

%NaCl = 0.055 ∗ HU110 − 2.6

Compared to images of homogenized samples, the fillet imageswere challenging to analyze. Gaping and loose structure gave pixelscomposed of both air and muscle. Removal of the sample edge inthe images caused problems for newly salted samples, where muchof the salt still was situated close to the sample edge. Also, removalof the bones from the images was not straightforward. A fixed rangefor bone CT values could not be set without overlap with the CT val-ues of muscle, because most bones in unsalted fillet were less densethan fully salted cod muscle, and also the fish bones became denserduring salting. These phenomena can be seen when comparing up-per and lower image in Figure 4 and although they were reduced byimage analysis and segmentation, some bone pixels remained andsome muscle pixels were removed from the image before calcula-tion of mean CT value. Another approach for handling problems asdiscussed previously, could be to simply leave edge, bones and soon in the images and calibrate the scanner with such remaining inthe images. This would perhaps be a better approach for potentialindustrial applications.

Figure 4 shows CT images where the predicted salt content is in-dicated with a calibration bar. Upper image displays fresh, unsaltedfillet, as indicated by the blue color. After 1 d in brine (middle im-

age), the salt has diffused into the outer areas of the fillet. The mus-cle is salt saturated on the surface and there is a gradient towardsthe sample center without salt. After 19 d in brine (lower image),the fillet is salt ripened.

The salt equations from the homogenized model samples, whichcan be regarded as “global,” were also used to predict the salt con-tent of the intact fillets. The model samples covered a large chem-ical range, so they should in principle be able to predict salt inany fish system chemically covered by this range. The best predic-tions were obtained when a 1-voltage model was used, performingequally well as the fillet models. Inclusion of more voltages gavepoorer predictions; perhaps more voltages provided models spe-cific for the homogenized samples, while the 1-voltage model wasmore universal. It could also be that the 2-voltage model was moresensitive to noise in the measured CT value, which was more pro-nounced in the fillet images as discussed previously. A range of pos-sible applications of CT on cod/salt systems can be made, however,mainly within product and process development and as a researchtool. CT could be used for optimizing both the amount and dis-tribution of salt during processes like cod desalting/rehydration(Bjørkevoll and others 2004; Erikson and others 2004; Andres andothers 2005) or salting (Lauritzsen and others 2004; Barat andothers 2006; Gallart-Jornet and others 2007). For optimization oflow-salt products (Larsen and others 2008), the prediction ac-curacy should preferably be even higher than obtained in thisstudy.

0 10 20 300

5

10

15

20

25

30

Reference NaCl content

Pre

dic

ted

Na

Cl

co

nte

nt

0 5 10 15 200

5

10

15

20

Reference NaCl content

Pre

dic

ted

Na

Cl

co

nte

nt

Exp.var: 0.996RMSECV: 0.6N = 35


A

B

Figure 3 --- Determination of salt (A) in homogenized modelsamples, modeled with 80 and 130 kV, and (B) in cod fillet,modeled with 110 kV.


E:Fo

odEn

ginee

ring&

Phys

icalP

rope

rties


Quantification of waterThe CT values within a cod varied from head (lower) to tail

(higher), which was also found by Kolstad and others (2008), whorelated this variation to variation in dry matter percent. In the fresh,lean cod muscle, consisting of mainly water and protein, thesecompounds will probably be highly correlated and thus it is a pos-sibility that variation in protein is what is actually measured, due tothe higher density of protein.

For prediction of water in cod fillets in our study, an explainedvariance of R2 = 0.99 and RMSECV = 1.1% water was found whencombining 80 and 110 kV, which is a relevant error for dried cod.In this case, adding the 3rd voltage did not add extra informationto the system. The RMSECV amounted to 4% of the water range,which is similar to the findings of Kolstad and others (2008); how-ever, only 1 voltage gave poor water predictions in our study. Sim-ilar results were found for homogenized samples, except 110 and130 kV was the optimal combination in that case. All (1 to 3) volt-ages were significant in any regression model. When 2 or 3 voltageswere modeled, 2 principal components were used, except only 1 forthe 110 + 130 kV combination in fillet samples. The reason for dif-ference between fillet and model samples in optimal voltages wasnot clear. It could be due to the differences the width of the fil-let compared with model sample distributions. The homogenizedmodel samples were not homogenous on pixel level. In particulardid the freeze-dried cod used in the model samples cause inho-mogeneity, resulting in broader distributions for such samples, butalso other model samples had broad distributions. This may causethe less noisy high-voltage images to be better predictors for water.

Prediction of water in cod fillet only using samples ranging fromfresh until just reaching salt ripened slightly improved the predic-tions of water. The best combination 80 + 110 kV provided RM-SECV = 0.8% NaCl and R2 = 0.98, with 2 PCs in the model. Any othervoltage combination gave models where only 1 PC was used.

CT has previously been shown feasible for measurement of drymatter (DM%) in fresh, unsalted Atlantic cod (Kolstad and others2008). They found an overall explained variance of R2 = 0.55 anda prediction error of SEP = 0.66 for 3 batches of cod (DM% range15.2 to 24.5); R2 = 0.87 and SEP = 0.48 was found for the best pre-dicted batch. Their SEP amounted to approximately 5% of their wa-ter range. Water or dry matter is also determined by CT in fatty

Figure 4 --- CT images at 110 kV,where the predicted salt content (%)is indicated with a calibration bar.The images are taken at the sameposition in the same fillet anddisplay unsalted fillet (upper image),after 1 d in brine (middle image), andafter 19 d in brine (lower image).

fishes (Gjerde 1987; Rye 1991) with explained variances between0.62 and 0.85.

Provided all analyzed samples were modeled, predicted watervalues for the best models are plotted against reference values inFigure 5. In contrast to NaCl, the prediction accuracy for water wassimilar for both homogenized model samples (A) and fillet samples(B). In both systems, 2 voltages gave more information on watercontent than 1 voltage. The correlation between water and salt isvery high for the fillet samples (Table 4), and thus we cannot besure whether it is salt, water, or both components that we actuallymeasure. The use of a designed, homogenized system weakenedthe correlations usually present in intact fillets, and the predictionaccuracy of NaCl and water in the homogenized samples thereforesuggested that both water and NaCl can be measured by CT.

Quantification of proteinProtein in fish has previously been predicted from CT values with

varying success. Protein was predicted in 4 freshwater fish species(Romvari and others 2002) with R2 = 0.87. They used fishes of vary-ing fat content, and from their plot between measured and esti-mated protein, it can be seen that the lean fish (0.8% fat) was notvery well predicted, in contrast to the fattier fish species. Predictionof protein in salmon (Rye 1991) was not successful (R2 ≤ 0.06%),perhaps due to a narrow protein range, while the explained vari-ance for protein was somewhat higher (R2 = 0.46) in rainbow trout(Gjerde 1987).

In this study, 2 voltages were necessary to predict protein fairlywell. The prediction accuracies for protein were similar to or lowerthan those found for salt and water. This is in agreement with theresearch referred previously. Predicted protein values for the bestmodels are plotted against reference values in Figure 6. For ho-mogenized samples (A), an explained variance of R2 = 0.99 andRMSECV = 1.5% (4% of range) was found using 110 and 130 kV.For fillet samples (B), an explained variance of R2 = 0.79 andRMSECV = 1.4% (12% of range) was found using 80 and 110 kV. Pre-diction of salt in samples ranging from fresh until just reaching saltripened did not improve the prediction of protein. As for water, all(1 to 3) voltages were significant in any regression model. When 2or 3 voltages were modeled, 2 principal components were used, ex-cept only 1 for the 110 + 130 kV combination in fillet samples. The




40 60 80

40

50

60

70

80

90

Reference water content

Pre

dic

ted

wa

ter

co

nte

nt

50 60 70 80

50

55

60

65

70

75

80

Reference water content

Pre

dic

ted

wa

ter

co

nte

nt



A

B

Figure 5 --- Determination of water (A) in homogenizedmodel samples, modeled with 110 and 130 kV, and(B) in cod fillet samples, modeled with 80 and 110 kV.

Table 4 --- Correlations between chemical components incod model samples and fillet samples.

Model samples Fillet samples

Water and salt −0.70 −0.97Water and protein −0.81 −0.97Protein and salt 0.15 0.69

correlation between protein and water was high in both cases, par-ticularly for fillet samples (Table 4). Prediction of protein in filletusing models from the homogenized model was unsuccessful, re-sulting in high deviations and high RMSEPs.

Conclusions

CT can be used for nondestructive quantitative and distribu-tional analysis of salt in cod during salting and drying. From

the same CT measurements, quantitative measurements of water,and less accurately protein, can be made. All over, the best RM-SECVs were 0.9% salt, 0.8% water, and 1.4% protein in fillet sam-ples, and 0.6% salt, 1.3% water, and 1.5% protein in homogenizedmodel samples. One voltage was sufficient for salt analysis in codfillet, while 2 voltages were necessary for prediction of water andprotein.

AcknowledgmentsThe Research Council of Norway supported this study throughgrant nr 153381/140. We thank Knut Dalen and Turid Mørkøre fromthe Norwegian Univ. of Life Sciences, and Karin Solgaard, Tom Chr.

10 20 30 40

10

15

20

25

30

35

40

45

Reference protein content

Pre

dic

ted

pro

tein

co

nte

nt

15 20 2514

16

18

20

22

24

26

Reference protein content

Pre

dic

ted

pro

tein

co

nte

nt



A

B

Figure 6 --- Determination of protein (A) in homogenizedmodel samples, modeled with 110 and 130 kV, and(B) in cod fillet, modeled with 80 and 110 kV.

Johannessen, and Anne-Kari Arnesen from Nofima Food, for dis-cussions and their technical assistance.

ReferencesAndres A, Rodriguez-Barona S, Barat JM. 2005. Analysis of some cod-desalting process

variables. J Food Engineer 70(1):67–72.Barat JM, Rodrıguez-Barona S, Andres A, Fito P. 2002. Influence of increas-

ing brine concentration in the cod-salting process. J Food Sci 67(5):1922–5.

Barat JM, Rodrıguez-Barona S, Andres A, Fito P. 2003. Cod salting manufacturinganalysis. Food Res Int 36(5):447–53.

Barat JM, Rodrıguez-Barona S, Andres A, Visquert M. 2004. Mass transfer analysis dur-ing the cod desalting process. Food Res Int 37(3):203–8.

Barat JM, Gallart-Jornet L, Andres A, Akse L, Carlehog M, Skjerdal OT. 2006. Influ-ence of cod freshness on the salting, drying and desalting stages. J Food Engineer73(1):9–19.

Bjørkevoll I, Olsen J-V, Olsen RL. 2004. Rehydration of salt-cured cod using injectionand tumbling technologies. Food Res Int 37(10):925–31.

Erikson U, Veliyulin E, Singstad TE, Aursand M. 2004. Salting and desalting of freshand frozen-thawed cod (Gadus morhua) fillets: a comparative study using Na-23NMR, Na-23 MRI, low-field H-1 NMR, and physicochemical analytical methods. JFood Sci 69(3):E107–14.

Gallart-Jornet L, Barat JM, Rustad T, Erikson U, Escriche I, Fito P. 2007. A comparativestudy of brine salting of Atlantic cod (Gadus morhua) and Atlantic salmon (Salmosalar). J Food Engineer 79(1):261–70.

Gjerde B. 1987. Predicting carcass composition of rainbow trout by computerized to-mography. J Anim Breed Genet (Zeitschrift Fur Tierzuchtung Und Zuchtungsbiolo-gie) 104(1–2):121–36.

Goldman LW. 2007. Principles of CT: radiation dose and image quality. J Nucl MedTechnol 35(4):213–25.

Horner WFA. 1997. Preservation of fish by curing (drying, salting and smoking). In:Hall GM, editor. Fish processing technology. 2nd ed. London: Chapman & Hall.p 32–73.

Hounsfield GN. 1980. Computed medical imaging—Nobel lecture, December 8, 1979.J Comput Assist Tomogr 4(5):665–74.

Haseth TT, Egelandsdal B, Bjerke F, Sørheim O. 2007. Computed tomography forquantitative determination of sodium chloride in ground pork and dry-curedhams. J Food Sci 72(8):E420–7.


E:Fo

odEn

ginee

ring&

Phys

icalP

rope

rties


Haseth TT, Høy M, Kongsro J, Kohler A, Sørheim O, Egelandsdal B. 2008. Determina-tion of sodium chloride in pork meat by computed tomography at different volt-ages. J Food Sci 73(7):E333–9.

IDF. 1997. Butter. Determination of the salt (chloride) content (potentiometric titra-tion method). Brussels, Belgium: Int Dairy Federation.

Kolstad K, Vegusdal A, Baeverfjord G, Einen O. 2004. Quantification of fat depositsand fat distribution in Atlantic halibut (Hippoglossus hippoglossus L.) using com-puterised X-ray tomography (CT). Aquaculture 229(1–4):255–64.

Kolstad K, Mørkøre T, Thomassen MS. 2008. Quantification of dry matter% and liq-uid leakage in Atlantic cod (Gadus morhua) using computerised X-ray tomography(CT). Aquaculture 275(1–4):209–16.

Larsen R, Olsen SH, Kristoffersen S, Elvevoll EO. 2008. Low salt brining of pre-rigorfilleted farmed cod (Gadus morhua L.) and the effects on different quality parame-ters. LWT Food Sci Technol 41(7):1167–72.

Lauritzsen K, Akse L, Johansen A, Joensen SU, Sorensen NK, Olsen RL. 2004. Physicaland quality attributes of salted cod (Gadus morhua L.) as affected by the state ofrigor and freezing prior to salting. Food Res Int 37(7):677–88.

Martens H, Næs T. 1989. Multivariate calibration. Chichester, U.K.: John Wiley & sons.Martınez-Alvarez O, Borderıas AJ, Gomez-Guillen MC. 2005. Sodium replacement

in the cod (Gadus morhua) muscle salting process. Food Chem 93(1):125–33.

NMKL. 1991. Nr 23. Moisture and ash. Gravimetric determination in meat and meatproducts. Nordic Committee on Food Analysis.

NMKL. 2003. Nr 6. Nitrogen. Determination in foods and feeds according to Kjeldahl.4th ed.: Nordic Committee on Food Analysis.

NMKL. 2004. Nr 178. Chloride (salt). Determination in foods by potentiometric titra-tion. Nordic Committee on Food Analysis.

NSLAD. 1969. AD/01 modified: salt-innhold i næringsmidler (salt content in food).Norges Slakteri Laboratorium. Provided by AnalyCen, Moss, Norway.

Qamhiyeh S. 2007. A Monte Carlo study of the accuracy of CT-numbers for rangecalculations in Carbon ion therapy [PhD thesis]. Heidelberg, Germany: Ruperto-Carola Univ. of Heidelberg. 141 p.

Romvari R, Hancz C, Petrasi Z, Molnar T, Horn P. 2002. Noninvasive measurement offillet composition of four freshwater fish species by computer tomography. AquacInt 10(3):231–40.

Rye M. 1991. Prediction of carcass composition in Atlantic salmon by computerizedtomography. Aquaculture 99(1–2):35–48.

Segtnan V, Høy M, Sørheim O, Kohler A, Lundby F, Wold J, Ofstad R. 2009. Noncon-tact salt and fat distributional analysis in salted and smoked salmon fillets usingX-ray computed tomography and NIR interactance imaging. J Agric Food Chem.Forthcoming.

Seibert JA, Boone JM. 2005. X-ray imaging physics for nuclear medicine technologists.Part 2: X-ray interactions and image formation. J Nucl Med Technol 33(1):3–18.

Skjervold H, Grønseth K, Vangen O, Evensen A. 1981. In vivo estimation of bodycomposition by computerized tomography. Zeitschrift fur Tierzuchtung undZuchtungsbiologie 98(1):77–9.

Thorarinsdottir KA, Arason, S, Bogason, SG, Kristbergsson, K. 2004. The effects of var-ious salt concentrations during brine curing of cod (Gadus morhua). Int J Food SciTechnol 39(1):79–89.


nondestructive analysis of salt, water, and protein in dried salted cod using computed tomography

Documents