Microstructure affects light scattering in apples

Postharvest Biology and Technology 159 (2020) 110996

Contents lists available at ScienceDirect

Postharvest Biology and Technology journal homepage: www.elsevier.com/locate/postharvbio

Microstructure affects light scattering in apples a,1

a,⁎,1

Zi Wang , Robbe Van Beers Bart Nicolaïa,c, Wouter Saeysa a b c

, Ben Aernouts

a,b

T a

a

, Rodrigo Watté , Pieter Verboven ,

KU Leuven Department of Biosystems, MeBioS, Kasteelpark Arenberg 30, 3001 Leuven, Belgium KU Leuven Department of Biosystems, Biosystems Technology Cluster, Campus Geel, Kleinhoefstraat 4, 2440 Geel, Belgium VCBT - Flanders Centre of Postharvest Technology, Willem de Croylaan 42, 3001 Leuven, Belgium

A R T I C LE I N FO

A B S T R A C T

Keywords: Dynamic controlled atmosphere Metamodel Biophotonics Contrast agent Image analysis X-ray computed tomography Spatially resolved spectroscopy

The success of long-term storage of apples under controlled atmosphere (CA) depends, amongst others, on the gas exchange properties of the fruit. As gas exchange is effectively dictated by the microstructure of the fruit, the ability to obtain microstructure data becomes critical to improve storage solutions. The current study complements scattering measurements by means of spatially resolved spectroscopy (SRS) with 3D microstructure data obtained with contrast enhanced X-ray computed micro-tomography (micro-CT). Complementary measurements with both techniques were performed on apples of different cultivars that have different optimal storage conditions and different fleshy microstructure (‘Kanzi’, ‘Braeburn’, ‘Jonagold’ and ‘Golden Delicious’). The mean reduced scattering coefficients of the subsurface tissue were calculated from SRS measurements in the 750 nm to 900 nm range at specific equatorial positions on the intact apple. Microstructural parameters such as cell size (equivalent spherical diameter), anisotropy, elongation, flatness, sphericity, object count, porosity as well as pore surface density were quantified and analyzed at the same spot up to 3 mm in depth from the fruit surface. A partial least squares regression model using the microstructural parameters as the different variables to predict the reduced scattering coefficient was built in order to identify the parameters contributing most to this relation. Both mean porosity and pore surface density showed the largest absolute regression coefficients. Furthermore, plotting the reduced scattering coefficient against mean porosity and pore surface density produced a linear relationship between the two parameters with an R2 of 0.89 for both sets of data. The linear relationship suggests that the porosity of individual fruit can be determined via optical SRS measurements. This could allow to sort fruit based on their porosity, thus promoting fine tuning of the storage strategies by reducing variation in the porosity and gas exchange rate of the fruit being stored together.

1. Introduction Apple fruit (Malus x domestica Borkh.) are placed into long-term storage after harvest to enable their off-season availability (Kader et al., 1989; Watkins and Nock, 2012). Advanced long-term storage strategies such as conventional and dynamic controlled atmosphere storage adapt the storage conditions to the gas exchange of the fruit being stored in order to minimize quality loss (Bessemans et al., 2016; Thewes et al., 2015). As the rate of gas transport/exchange is dependent on the concentration gradient resulting from respiration and physical properties such as the porosity of the fruit being stored, an accurate gauge of these physical characteristics may allow better understanding, management and control of storage (Ho et al., 2010; Schotsmans et al., 2004). However, since parameters such as porosity are an internal measure,

obtaining representative data is extremely difficult and necessitates the use of non-destructive techniques such as x-ray computed tomography (CT) (Herremans et al., 2014; Nugraha et al., 2019; Wang et al., 2018). CT, however, has considerable hardware and computational requirements, limiting its use in high throughput applications (De Schryver et al., 2016; Janssens et al., 2016; van Dael et al., 2018, 2017), while high resolution applications with so-called micro- or nano-CT typically require destructive sampling. Conversely, currently available automated quality assessment solutions are based on optical measurements (Studman, 2001), which have distinctive advantages in processing speed thanks to their relative simplicity. However, currently available techniques such as Vis/NIR spectroscopy have an output spectrum influenced by both the absorption and scattering of light, meaning that both chemical composition and physical characteristics of the fruit

⁎

Corresponding authors. E-mail addresses: [email protected] (R. Van Beers), [email protected] (W. Saeys). 1 These two authors contributed equally to this work. https://doi.org/10.1016/j.postharvbio.2019.110996 Received 1 April 2019; Received in revised form 19 August 2019; Accepted 23 August 2019 0925-5214/ © 2019 Elsevier B.V. All rights reserved.

Postharvest Biology and Technology 159 (2020) 110996

Z. Wang, et al.

2. Material and methods

affect the output data (Nicolaï et al., 2007). Due to the promising features of Vis/NIR spectroscopy, it has been used for the quality assessment of a wide range of agro-food products, including apple (Bobelyn et al., 2010; McGlone et al., 2002; Sun et al., 2017). More advanced instruments allow to extract the bulk optical properties (BOP) separately: the bulk absorption coefficient μa, the bulk scattering coefficient μs and the angular scattering pattern expressed by the anisotropy factor g (Tuchin, 2007). The bulk scattering coefficient μs and the anisotropy factor g are often combined into a single parameters: the reduced scattering coefficient μs’ (Tuchin, 2007). The extraction of the BOP can be achieved destructively by using double integrating sphere (DIS) measurements (Aernouts et al., 2013; Van Beers et al., 2017) or nondestructively by using time- or spatially resolved measurement techniques, among others (Cubeddu et al., 2001; Nguyen Do Trong et al., 2014; Nicolaï et al., 2008; Sun et al., 2016; Torricelli et al., 2015; Wang et al., 2017a). These advanced methods allow to attribute sample parameters, both chemical and physical, to the optical measurements using the BOP. Unsurprisingly, good correlations with chemical composition such as water and sugar content can be established with bulk absorption (Adebayo et al., 2016; Nguyen Do Trong et al., 2014), as it directly relates to the absorption bands created by present molecules (Lu, 2016). Although attempts have been made to link scattering properties to mechanical and structural properties of apple tissue during storage (Aernouts et al., 2011; Cen et al., 2013), no one-to-one relation between microstructural features and the optical properties has been reported to date. Without matching data, it is difficult to pinpoint the main causes of scattering in fruit tissue such as that of the apple, and even more difficult to properly interpret physical scattering data. It has, however, been hypothesized that physical parameters such as porosity and cell density would affect the output scattering spectrum, since the number and degree of refractive boundaries are directly determined by these parameters (Lu, 2016). One-to-one measurements with both optical and micro-CT instruments are required to investigate the correlation between optical measurements and physical parameters. Being the primary barrier protecting the fruit from the external environment, the surface and subsurface regions of a fruit (including the cuticle, epidermal, hypodermal and hypanthium layers) are both heterogeneous and dense (Schotsmans et al., 2004; Veraverbeke et al., 2003; Verboven et al., 2008). Apart from synchrotron scans of a single ‘Braeburn’ fruit (Verboven et al., 2013), micro-CT data of this layer is scarce. Datasets from multiple cultivars and fruit could not be readily obtained, as conventional benchtop systems cannot yield much analyzable data with such high density and low contrast. So, prior to being able to provide matching micro-CT data, additional developments must be made in the scanning methodologies for conventional micro-CT scanners to enhance the output data. Fortuitously, recent developments in micro-CT contrast enhancement have enabled the ability to acquire subsurface data, albeit modifications are necessary to adapt the published protocols for subsurface scans (Wang et al., 2017b). The aim of this research was to test the hypothesis that scattering coefficients measured with SRS on intact apples are correlated to microstructure properties of the subsurface tissues that affect gas exchange during storage. First, contrast enhancement and image analysis protocols for micro-CT were adapted to acquire 3D microstructural data such as porosity, and the cell and pore size distribution of subsurface samples of apples. Second, four cultivars of apples, ‘Kanzi’, ‘Braeburn’, ‘Jonagold’ and ‘Golden Delicious’ with known differences in porosity (listed in increasing order, with increasing gas diffusivity and decreasing susceptibility to hypoxia and storage disorders (Ho et al., 2010)) were examined on a one-to-one basis to determine which physical parameters have the strongest relation with changes in optical scattering.

2.1. Fruit source and selection Apples from the 2017 harvest season in Europe were purchased from a local supermarket. ‘Kanzi’, ‘Braeburn’, ‘Jonagold’ and ‘Golden Delicious’ (referred to as “Golden” further on) apples were acquired during the second week of March 2018 and stored at 4 °C in normal air until used for experimentation (< 7 d from date of purchase). Apples were of Belgian origin with the exception of the ‘Kanzi’ cultivar, which was from Italy. All fruit used were approximately 70 to 75 mm in diameter with a weight range from 145 to 191 g. All measurements were taken from equatorial regions of the apples, which were marked and subjected to both optical SRS and micro-CT imaging to obtain 1:1 measurements. For every cultivar, six apples were measured with four measurements per apple. However, not all measurements were utilized in the subsequent analysis, since unseen subsurface damage as well as movement artifacts and distortions during micro-CT imaging were inevitable. 2.2. Hyperspectral laser scatter imaging (HLSI) A HLSI setup, a form of non-contact spatially resolved spectroscopy (SRS), designed by Van Beers et al. (2015) was used to measure the apple samples at the marked positions. In short, combining a supercontinuum laser source with a monochromator allowed to scan wavelengths in the 500–1000 nm range in steps of 5 nm. Using a CCD camera (TXG-14NIR, Baumer, Frauenfeld, Switzerland), the diffuse reflectance at multiple distances from the point of illumination was captured. A schematic representation of the measurement configuration is shown in Fig. 1, while the reader is referred to Van Beers et al. (2015) for a more detailed description. Both a dark and a white reference were taken during every measurement campaign. The dark image was taken for all wavelengths by closing the monochromator shutter, while the light was guided into an integrating sphere in order to use the diffuse light exiting the sphere as a white reference signal. This allowed to convert the raw signals into a relative reflectance signal. After performing the dark and white corrections, the sample images were radially averaged resulting in a diffuse reflectance profile with relative reflectance as a function of the distance from the illumination center, further referred to as the SRS profiles (Van

Fig. 1. Schematic representation of the hyperspectral laser scatter imaging (HLSI) setup. Monochromatic light originating from the combination of a supercontinuum laser and a monochromator is focused on the sample under an angle of 20° to the normal. The resulting diffuse reflectance glow spot is imaged by a CCD camera. 2

Postharvest Biology and Technology 159 (2020) 110996

Z. Wang, et al.

2.5 μm. X-ray tube voltage and current were set at 60 kV and 175 μA respectively, and 2400 projection images were captured at 500 ms each, totaling in 20 min per scan. Octopus Reconstruction 8.9.4 (XRE, Gent, Belgium) was used for reconstruction via a filtered back projection algorithm. Ring artifact and noise filters were applied to improve the overall output quality of the reconstructed images (Boas and Fleischmann, 2012). All images were downscaled to 8 bits to reduce computational requirements during image processing.

Beers et al., 2015). This procedure was repeated for all of the locations on the fruit of each cultivar. 2.3. Bulk optical properties from HLSI measurements To determine the bulk optical properties (BOP) from the non-destructive HLSI measurements, a metamodeling approach was used (Aernouts et al., 2015; Watté et al., 2015). Materials with known optical properties - so called optical phantoms - were measured using the HLSI setup. This allowed to link the diffuse reflectance profiles of the phantoms with the BOP obtained from measuring the phantoms in a double integrating spheres (DIS) setup (Aernouts et al., 2013). The same set of liquid optical phantoms, based on Intralipid® 20% (Fresenius Kabi, Sweden) and the water-soluble azo dye Naphthol Blue Black (NBB) (195243, Sigma Aldrich, Missouri, USA), as described in Watté et al. (2015), was used for building the metamodel in this work. With these two components, a total of 49 calibration phantoms were created. For the calibration phantom set, seven scattering levels with a μs’ ranging from 3.1 cm−1 to 106.2 cm−1 at 600 nm, were made by changing the Intralipid® (IL) concentration. In addition, seven absorbing levels with a μa ranging from 0 cm−1 to 58.1 cm−1 at 620 nm were created, by controlling the added amount of a pre-made 5000 μM NBB stock solution. A metamodel was built to link the respective BOP values to the SRS profiles obtained for each of the liquid phantoms and considered wavelengths. Gaussian Process based Kriging surrogate interpolation in the ooDACE toolbox for Matlab was employed to build the metamodel (Couckuyt et al., 2012). A forward light propagation model was constructed, using the absorption coefficient μa, the reduced scattering coefficient μs’ and a predefined distance from the illumination point to obtain the reflectance value. Although this general model is independent of the wavelength, extrapolation outside the BOP range of the training set is typically detrimental (Watté et al., 2015). As the forward model describes the SRS profiles in function of the BOP, an inversion is necessary to go from an SRS measurement to the estimation of BOP values. To do this, an optimization procedure was used to find a set of BOP for which the simulated SRS profiles matched the measured SRS profiles for the measured sample. This is typically performed wavelength-by-wavelength where a cost function is minimized for each individual wavelength (Watté et al., 2015). The smaller this cost function, the closer the simulated SRS profile matches the measured one. The Nelder-Mead algorithm was used as optimization procedure for the inverse problem and several starting points evenly distributed over the search space were defined to obtain the best solution. Even though the cost function might not be strictly convex, this was shown to be a robust approach (Watté et al., 2013). The performance of this methodology was validated thoroughly on a similar set of liquid validation phantoms and the results are reported in Watté et al. (2015).

2.5. Micro-CT image processing The image processing workflow is a modified version of the one reported in (Wang et al., 2017b). While all the optical measurement sites were scanned via micro-CT, not all of the contrast enhanced microCT measurements generated tomographs of sufficient quality to be properly segmented. In short, to generate consistent and reproducible quantitative datasets useful in statistical analysis, some minimum quality requirements exist. The criteria for the current study were set as: good resolution and contrast, clear intercellular separation, no movement artifacts or distortions, and no significant sample tissue damage. Binning of the micro-CT samples was therefore necessary to ensure that the output tomographs could be successfully segmented and quantified in a reproducible manner. Of those binned images, only 7, 10, 4 and 12 image sets from ‘Kanzi’, ‘Braeburn’, ‘Jonagold’ and Golden cultivars, respectively, were considered to be top quality datasets, as the remaining data sets did not meet one or more of the selection criteria. Four image sets per cultivar were selected at random (maximum of one set per fruit, to prevent skewing of data towards any particular fruit) for further processing, resulting in a total of 16 image sets. Additionally, volumes of interest (VOIs) 3000 × 3000 × 3125 μm (l × w × d) were extracted from the reconstructed datasets to further reduce computational requirements. Image segmentation was done via a histogram based multi-thresholding module in Avizo 9.4 (FEI, Bordeaux, France). However, due to the stark anatomical differences between the cuticle, dermal and hypanthium layers, the VOIs were divided to ease image processing. At the approximate boundary of the hypodermis and the hypanthium, where an abrupt change in cell sizes can be observed, the VOI was split into two sub-VOIs (upper and lower). The cuticle was first extracted from the upper sub-VOI based on its distinctively lower greyscale value when compared to the dermal cell layers. Next, multi-thresholding was applied to mark black, grey and white pixels, which correspond to pores, hypodermal cells and cell boundaries, respectively. With the cuticle and the hypodermal cells marked, the remaining pixels in the middle of the two were marked as the epidermis. The hypanthium was segmented in a similar way, as described previously (Wang et al., 2017b). Despeckling and opening operations were then performed on both sub-VOIs to reduce the noise of the binary images obtained. Watershed transform was applied to both sub-VOIs separately to ensure individual cells are fully separated. Following intercellular separation, the upper and lower sub-VOIs were recombined into the initial complete VOI. Manual corrections were applied when necessary to reduce stitching artifacts. A border kill function was applied to remove incomplete cells at the boundaries of the VOI. Furthermore, a sphericity filter was applied (cutoff at 0.75) which removed nonsensical objects masquerading as segmented cells (typically 5–7 % of VOI). In all cases, binarized cells were labelled and their equivalent spherical diameter, volume and geometric center was calculated in addition to sphericity. A sliding window of 3000 × 3000 × 125 μm (adjusted for the curvature of the sample) was ran from the surface of the fruit inward until 3 mm in depth to obtain per depth profiles for both porosity and cell sizes. Depth reached by the sliding window is thus defined as the depth of the data point being analyzed. A porosity profile was created via pixel counting within the search window, while inclusion criteria for cells required their geometric centers to be within the search windows. Additional

2.4. Contrast enhancement and micro-CT measurements For all examined fruit, tissue samples of approximately 5 × 5 × 10 mm (l × w × d) were excised from the center of the marked measurement sites from the skin inwards, using an ultrathin razor blade with a thickness of approximately 125 μm. The released cytosol was drained with tissue paper and samples were subsequently placed in a freshly prepared 10% (w/v) cesium iodide contrast agent solution (Acros Organics, Geel, Belgium). All excised samples were incubated with the contrast solution at room temperature for 2 h with periodic agitation (tube inversions and flicking) at 15 min intervals. At the twohour mark, all samples were drained of excess contrast solution, wrapped in protective parafilm to prevent dehydration, and immediately subject to micro-CT scans. A Phoenix Nanotom micro-CT system (General Electric, Heidelberg, Germany) was used for all scans in the study. 12-bit greyscale images were captured on a 2304 × 2304 detector with voxel resolution of 3

Postharvest Biology and Technology 159 (2020) 110996

Z. Wang, et al.

Table 1 Subsurface parameters measured during quantitative image analysis. Measure

Brief description

Equivalent diameter Anisotropy Elongation Flatness Sphericity

Diameter of the sphere of same volume. It is a representation of cell size. 1 minus the ratio of the smallest to the largest eigenvalue of the covariance matrix. Measures the deviation of an object from a spherical shape. Ratio of the medium to the largest eigenvalue of the covariance matrix. Elongated objects will have small values close to 0. Ratio of the smallest to the medium eigenvalue of the covariance matrix. Flat objects have small values close to 0. Ratio of the surface area of a sphere (with the same volume as the given object) to the surface area of the object. The sphericity of a sphere is 1 and any object that is not a sphere will have sphericity less than 1. Number of cells and cell clusters within any given volume expressed in integers. Higher object count in a volume results in a higher object density. Count of voxels assigned to pores within a volume of interest (VOI), expressed as a percentage of total voxels within the VOI. Tissue with higher porosity contains more air. Surface area of pores divided by examined tissue volume. A proxy measure of the number of phase-change/refraction events encountered by light in the scattering process.

Object Count Porosity Pore surface density

hypodermis and hypanthium as well as pores were each assigned a colored label being purple, red, blue, green and yellow, respectively. Analysis of the cuticle layers indicates minimal differences in their thickness between the ‘Kanzi’, ‘Braeburn’ and ‘Jonagold’ cultivars at 18.49 ± 0.97 μm, 17.13 ± 0.58 μm, and 18.04 ± 0.49 μm, respectively. However, Golden apples have a significantly thicker cuticle layer at 22.12 ± 0.34 μm, as observable in the reconstructed image (Fig. 2d). Analysis of the single cell epidermal cell layer indicates negligible differences in layer thickness between the ‘Kanzi’, ‘Braeburn’, ‘Jonagold’ and Golden cultivars at 22.14 ± 1.09 μm, 20.32 ± 1.21 μm, 23.52 ± 0.62 μm and 21.53 ± 0.80 μm, respectively. It should also be noted that the hypodermal cells of ‘Kanzi’ apples are distinctive in shape (Fig. 2a), as they are flatter than the hypodermal cells found in the other three cultivars (Fig. 2b-d), with a mean flatness of 0.49 ± 0.01 versus 0.58 ± 0.02, 0.59 ± 0.03, 0.64 ± 0.01, respectively. Also, the analysis was based on a limited dataset which is not necessarily representative for the different cultivars.

geometric properties were also obtained without depth information or sphericity filtering which included anisotropy, elongation, and flatness. Lastly, object density (object count/mm3) was also obtained via the sliding search window method, but no filtering of any kind was included as the geometric quality or completeness of segmented objects do not negate their existence in the VOI. Table 1 provides an overview of the quantitative parameters with a brief description. 2.6. Statistical analysis All quantitative comparisons between the different examined apple subsurface samples were subjected to analysis of variance (ANOVA) with a sample size of four. Statistical analysis was performed utilizing Prism 7 (Graphpad Software, La Jolla, USA). Statistical significance was noted if the p-value obtained was less than 0.05. As the goal was to use microstructure-related parameters to explain the observed differences in light scattering, the importance of the different parameters was investigated by building a PLS model using the PLS Toolbox (version 8.6.1, Eigenvector Research Inc., Manson, WA, USA) in Matlab. To do so, eight microstructure-related parameters (Table 1) were used together in order to predict the reduced scattering coefficient (μs’) as determined using the HLSI technique. An ‘autoscale’ preprocessing was applied to the reduced scattering coefficient as well as to the microstructure-related parameters in order to achieve an equal scaling, making a comparison possible. The PLS model was built including a cross-validation step using venetian blinds with 8 splits. This allowed to evaluate the RMSEC and RMESCV values in order choose the number of latent variables to be included. Finally, the regression coefficients for the built PLS model were evaluated in order to identify the most important microstructural features, relating to light scattering. Lastly, the linear regression between the most important microstructural parameters and scattering was determined.

3.2. Subsurface porosity profile, cell size distribution and object density Subtle but distinctive differences in porosity are visible in 3D renders of the obtained datasets (Fig. 3a-d). Volume renderings of the pores show not only a significant increase in mean porosity going from the ‘Kanzi’ to Golden cultivars (in the previously established order), but also a thinner subsurface region dominated by small pores in the more porous cultivars such as ‘Jonagold’ and Golden. The mean porosity of the VOIs were first determined in order to confirm the visual observation as well as to determine the porosity variations in the entire VOIs between the four cultivars. Accordingly, the mean porosity of the cultivars of the ‘Kanzi’, ‘Braeburn’, ‘Jonagold’ and Golden cultivars was calculated to be 11.92 ± 0.83%, 13.95 ± 0.76%, 20.15 ± 1.50% and 26.12 ± 1.23%, respectively. In order to determine where the variation in porosity existed, a depth dependent porosity profile was generated and plotted. Graphs of porosity as a function of depth show clear differences between the cultivars (Fig. 4a). While tissue porosity of the initial 250 μm layer below the surface of the varying cultivars are not statistically different, a rapid increase in porosity can be observed in the ‘Jonagold’ and Golden cultivars. At depths of 375 μm and 625 μm or more below the skin, the difference in porosity between these cultivars became more statistically significant (Fig. 4a). It should be mentioned that only hypodermal and hypanthium cells were included in the analysis of cell sizes. Differences in cell sizes in the hypodermis and hypanthium are minimal in the examined VOI (Fig. 3eh). Statistics only denoted a significant difference between ‘Kanzi’ and the remaining cultivars in parts of the hypanthium where ‘Kanzi’ cells were notably larger (Fig. 4b). Average cell sizes did increase over the depth of the analyzed VOI from approximately 40 μm in the hypodermis to more than 150 μm in the hypanthium. In terms of object count, a statistically significant difference was observed between ‘Kanzi’ apples and the other cultivars in the first

3. Results 3.1. Subsurface microstructure In the reconstructed images of contrast enhanced micro-CT scans, a number of anatomical features are immediately visible (Fig. 2). Airspaces, whether they be outside (top of images) or inside of the fruit tissue (surrounded by cells) are marked by black pixels. Grey pixels mark both cuticle as well as cells (excluding cell walls), but their distinctive structural characteristics and location negates any difficulties in their recognition. White pixels mark cell walls (Wang et al., 2017b). The structural components are clearly distinguishable from each other, with the exception of the single layer of epidermal cells located directly under the cuticle. The combination of thicker cell walls and small epidermal cell sizes made that not all epidermal cells are clearly visible. Due to this inconsistency, only the thickness of the epidermal cell layer could be extracted (Fig. 2). Fig. 2e illustrates the segmentation and labelling of the subsurface samples. The cuticle, epidermis, cells of the 4

Postharvest Biology and Technology 159 (2020) 110996

Z. Wang, et al.

Fig. 2. Reconstructed images of contrast enhanced micro-CT scans of ‘Kanzi’, ‘Braeburn’, ‘Jonagold’ and Golden cultivars (a–d respectively). Air, cells and cell boundaries are marked by black, grey and white pixels respectively. A segmented and labelled version of subfigure d (shown as subfigure e), illustrates the different microstructural fractions that can be analyzed. The purple and red layers correspond to the cuticle and the epidermal layers. Blue and green structures mark the hypodermal and hypanthium cells within their respective layers. Yellow and black refer to pores and cell walls (thickened for demonstrative purposes) respectively. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

125 μm layer of the hypodermis. The object density of ‘Kanzi’ was less than 5000/mm3, versus that of ‘Braeburn’, ‘Jonagold’ and Golden apples for which it was more than 7000/mm3 (Fig. 4c). This is consistent with the observation that ‘Kanzi’ apples contained anatomically different hypodermal cells than the other cultivars (Fig. 2a). However, beyond the initial search window, object density did not statistically differ between the different cultivars, even though ‘Kanzi’ apples exhibited the trend of having lower object density beyond 1.5 mm under the surface of the fruit, due to the larger cells in the hypanthium.

(2017) for apple hypanthium and dermal (defined as the epidermal layer with some hypodermal cells) tissues separately. Moreover, in this wavelength region, reliable μs’ estimates were obtained. At other wavelengths, where high absorption values due to skin pigments or water are present (Van Beers et al., 2017), cross-talk between μa and μs’ was still present. An overview of the mean obtained reduced scattering coefficient μs’ values calculated over the entire wavelength range and per cultivar is given in Table 2. From Table 2 it is clear that the difference in μs’ is not significant between ‘Kanzi’ and ‘Braeburn’. Both Golden and ‘Jonagold’ have μs’ values that are significantly higher in comparison to ‘Braeburn’ and ‘Kanzi’.

3.3. Scattering properties The reduced scattering coefficient μs’ of the different apple samples was estimated with the built metamodel. The accuracy of this metamodel was quantified using a separate test set of optical phantoms, resulting in an R2 of 0.96 and an RMSEP of 3.2 cm−1 for estimating the reduced scattering coefficient. The mean μs’ values with standard deviation in the 750 nm–900 nm wavelength range are shown per cultivar in Fig. 5. As seen in Fig. 5, in the used wavelength region, an overall flat μs’ spectrum was obtained. This was also observed in Van Beers et al.

3.4. Microstructure – light scattering relations A PLS model was built using the eight autoscaled microstructural parameters as variables in order to predict the reduced scattering coefficient of individual locations on the fruit. For the microstructural parameters, instead of using the per-depth data like in Fig. 4, a volume average was taken at each measurement location, as the SRS measurements capture photons that have travelled through the different

Fig. 3. 3D volume renderings of internal pores (a–d) and cells post sphericity filtering (e–f) of ‘Kanzi’, ‘Braeburn’, ‘Jonagold’ and Golden cultivars respectively. In the cell renders, hypodermal cells and hypanthium cells are colored blue and green respectively. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) 5

Postharvest Biology and Technology 159 (2020) 110996

Z. Wang, et al.

Table 2 Mean reduced scattering coefficient and standard deviation over a wavelength range from 750 nm to 900 nm for the four apple cultivars (24 measurements per cultivar). Values without a common letter superscript are significantly different at p < 0.05. μs’ (cm−1) ‘Golden Delicious’ ‘Jonagold’ ‘Braeburn’ ‘Kanzi’

20.76 18.44 15.90 15.22

± ± ± ±

1.20 1.80 1.57 1.49

a b c c

Fig. 6. Regression coefficients of a PLS model using the eight autoscaled microstructure-related parameters (variables) to predict the reduced scattering coefficient μs’. Volume averages were utilized for the PLS analysis.

of 0.97 and an R2CV of 0.92. This already shows a clear connection between microstructure and scattering. To give a better idea on which parameters have the largest contribution to this prediction model, Fig. 6 shows the PLS regression coefficients of the eight microstructural parameters (variables) extracted from the micro-CT images. A higher absolute value of the regression coefficient means a higher contribution of that parameter in the prediction of the scattering coefficient μs’. From the analysis in Fig. 6, it is clear that the fruit porosity and pore surface density contribute the most to the prediction of μs’, with the largest regression coefficients. This could be expected as a higher porosity results in an increase in the surface area of cell to air (and vice versa) boundaries, and by extension the likelihood to reach a refractive index transition between cells and air (with indices of around 1.33 and 1 respectively). The change in refractive index across boundaries causes light to change its direction, thus to scatter. Other microstructural parameters have a lower contribution to the relation with μs’, shown as regression coefficients closer to zero. It should be noted that object count (of cells and cell clusters) did not contribute much to the prediction of the observed scattering. This suggests that the scattering contribution of cell walls, membranes and organelles (according to the Mie solution (Aernouts et al., 2014)) to the variation in observed scattering is less compared to the scattering by the encountered phase changes. To further investigate the relationship with porosity, Fig. 7 shows the obtained direct linear relation between mean porosity / pore surface density and the reduced scattering coefficient μs’. Results from the exact same measurement location for both the micro-CT and SRS setup were used to produce this figure. As shown in Fig. 7, a good linear relationship (R2 = 0.89 and RMSE = 0.97 cm−1) was found between mean tissue porosity and the predicted reduced scattering coefficient μs’. In addition, a clear trend is

Fig. 4. Porosity, cell size (equivalent spherical diameter) and object count (cells, cell fragments and clusters) as a function of depth from surface. Asterisks are included when the dataset started to become significantly different from the curve below (a) or is significantly different from the other three curves in the graph (b, c). Single asterisk denotes P < 0.05, double asterisk denotes P < 0.01. All graphs are with n = 4, plotted with standard error bars.

Fig. 5. Mean and standard deviation of the reduced scattering coefficient μs’ per cultivar in the 750 nm–900 nm wavelength range.

tissue layers. For μs’, like before, a mean value in the wavelength range between 750 nm and 900 nm was used for each measurement location. The PLS model was built using 5 latent variables and resulted in an R2C 6

Postharvest Biology and Technology 159 (2020) 110996

Z. Wang, et al.

Fig. 7. Relationship of the mean porosity and pore surface density as estimated by the contrast enhanced micro-CT with the reduced scattering coefficient μs’ estimated from SRS measurements.

4.2. Porosity has the strongest relation to variation in light scattering properties

visible within the different cultivars. This indicates that also the withincultivar variation in μs’ is related to differences in mean porosity. Pore surface density also gave a good linear relationship with μs’, showing an R2 of 0.89 and an RMSE of 0.94 cm−1. The other microstructural parameters only showed R2 values up to 0.28, confirming the results presented in Fig. 6.

The measured apple cultivars showed a high variability in their reduced scattering coefficient. The focus of this research was to investigate if the microstructure could explain this variability in scattering. Although significant differences in scattering between the cultivars were found, previous work on the estimation of BOP of both ‘Braeburn’ and ‘Kanzi’ during pre-harvest maturation showed different absolute values. The μs’ of ‘Braeburn’ in the same wavelength region was higher for the dermal region at 11.29 ± 1.52 cm−1 compared to 9.22 ± 1.36 cm−1 for ‘Kanzi’, while lower values were found for the ‘Braeburn’ hypanthium tissue (7.21 ± 0.82 cm−1) in comparison to that of ‘Kanzi’ (10.07 ± 1.46 cm−1) (Van Beers et al., 2017). Moreover, in the same study, the overall magnitude of μs’ for both ‘Braeburn’ and ‘Kanzi’ was lower in comparison to the values found here. A possible explanation is the measurement of intact apples in this study, resulting in BOP for the combination of dermal and hypanthium tissue. In addition, the fruit hypanthium tissue in Van Beers et al. (2017) was measured after adding water to the sample, filling some of the air pores inside the tissue, possibly resulting in an underestimation of μs’. Changes in the refractive index at boundaries causes scattering in a heterogeneous medium (Lu, 2016; Tuchin, 2007). However, considering that the subsurface region of apple fruit is highly complex, it was uncertain which morphological properties affect the scattering output of resolved optical methods the most. In theory, physical parameters such as cell size, density, and porosity all have an effect on scattering. Both the number of transitions and magnitude of change in refractive index are expected to alter the observed scattering (Lu, 2016). An increase in cell size, or rather, a reduction in the cell density has the fundamental effect of reducing the number of cell boundaries within the volume in question. Porosity has the opposite effect as it increases the number of cell-air and air-cell interfaces encountered by the incoming photons (by effectively increasing the pore surface density). However, no statements could be made to date, because only porosity data was available given the limitations of conventional microCT, and the difficulty was compounded by the fact that the refractive indices differ significantly between air and water (at 1.00 and 1.33 in the 589 nm range, respectively) (Hale and Querry, 1973). Given that the new contrast enhanced micro-CT provides far more information than just porosity, the parameter that causes the greatest scattering variability could be pinpointed. To start, an examination of cell sizes yields the observation that while cells do differ significantly in size at certain depths from the surface, the lack of variation between three of the four cultivars suggest that it plays a limited role in the scattering differences. Observations made from the cell density side of the dataset

4. Discussion 4.1. Light scattering data can be interpreted better via contrast enhanced micro-CT The flexibility, speed and non-destructive character of optical methodologies make these ideal for use in the agro-food industry as derivative instruments, whether handheld or in-line, and can be utilized to provide measurements for sorting and storage applications (Marques et al., 2016; McCormick and Biegert, 2018; Yan and Siesler, 2018). However, even advanced and resolved spectrographic methods are limited in their usefulness without insight in the sample properties at the basis of the observed signals (Van Beers et al., 2015). The only way to understand the cause of scattering variability is to study the underlying microstructure of the optical measurement site in 3D. Conventional micro-CT can provide data on porosity (Mendoza et al., 2007), and partial cell data if the porosity is high enough (Herremans et al., 2015). This severely limited the usefulness of conventional micro-CT in the current study, as the high density of the subsurface layer made it nearly impossible to obtain parameters other than porosity. For a more comprehensive analysis, additional subsurface parameters such as cell sizes, object density, cuticle thickness, etc. were necessary. The new protocol for contrast enhanced micro-CT was designed to provide such microstructural parameters from the optical measurement sites, thus enabling a 1:1 analysis to determine which variable(s) affect the observed difference in the NIR spectrum of the SRS measurement the most. Although other researchers tried to link the bulk optical properties to microstructural data, this was often performed based on a limited number of 2D measurements (Cen et al., 2013). Moreover, the measurements were performed over long periods of time causing large changes in microstructure. Although a relationship between cell size parameters and optical properties was established in storage conditions, no microstructural cause of fundamental scattering differences between the studied cultivars was shown (Cen et al., 2013). In the current study, a more fundamental approach was followed in which the variation in scattering between fruit and cultivars was investigated using detailed 3D microstructural information from the same tissue. 7

Postharvest Biology and Technology 159 (2020) 110996

Z. Wang, et al.

4.4. Limitations and further studies

suggest the same, as number of objects encountered in relation to the reduced scattering coefficient only yielded an R2 of 0.22, thereby also limiting its role in the observed scattering. However, the Mie solution to Maxwell’s equations suggests that the main scattering particles in the Vis/NIR spectrum would be the organelles in the cells, where higher cell density (and by extension smaller cells and higher organelle density) should contribute significantly to the observed scattering. The fact that this was not observed may suggest that the scattering effect due to the phase-changes between cells and pores dominates. The mean porosities calculated from the enhanced micro-CT data showed a significant difference between the four cultivars in a similar manner to the optical measures. While the mean porosity of ‘Kanzi’ samples is in agreement with previously reported values (12.1 ± 3.0%), the values for ‘Braeburn’, ‘Jonagold’ and Golden are significantly lower than the reported values (18.4 ± 4.4%, 24.5 ± 2.6%, and 29.8 ± 0.8%) (Herremans et al., 2015; Ting et al., 2013). However, it should be noted that these previous reported porosity values only included porosity from the hypanthium layer. Furthermore, at approximately 2 mm under the surface, the observed porosity profiles of all four cultivars begin to stabilize and are corresponding with literature values stated above, with the exception of ‘Kanzi’ which increased to 16.75 ± 1.64% (potentially due to its Italian origin, as a result of notably different climate and growing conditions). In any case, attempts to correlate reduced scattering coefficients with mean porosity as well as pore surface density both yielded an R2 of 0.89, which suggests that the observed variation in scattering is more correlated to the variation in porosity/pore surface density than that in cell parameters. Additionally, a PLS analysis of parameters in addition to the basic cell geometry and porosity information also indicated that porosity contributes most to predicting the variation in scattering properties. Based on these findings, we conclude that differences in porosity are the main cause for the observed differences in optical scattering among the investigated apples.

As this study is the first of its kind linking optical scattering data to 3D apple microstructure, drawing additional conclusions should be avoided. While a clear correlation between porosity and optical scattering has been observed, the current dataset is limited in size due to both yield and processing time restrictions. Furthermore, frosting was observed during the 2017 harvest season (especially on ‘Jonagold’ apples), and the corresponding yield of good quality micro-CT data was exceptionally low on ‘Jonagold’ (at 4 samples out of 24 measured). It could be hypothesized that the frost induced changes to the apple samples adversely affected the yield of contrast enhanced micro-CT scans. To avoid that one cultivar would be more represented than another, only 4 samples per cultivar were used in the statistical analysis portion of the study. While a clear relation was observed for these 16 samples, these results should be confirmed in a larger experiment. Two routes can be taken to accomplish this refinement process. If the researcher wishes to obtain data on cell and tissue parameters as well as porosity to confirm the link (or lack thereof) between scattering and specific tissue parameter such as those listed in Table 1, one to one optical scattering and contrast enhanced micro-CT measurements should be taken again for at least another 2 non-frosting or abnormal harvest seasons. Alternatively, if the researcher wishes to simply confirm that optical scattering is related to porosity and pore density, standard micro-CT scans of subsurface samples across at least 2 more seasons would suffice. However, it should be kept in mind that both methods have their limits. The first approach one produces more comprehensive datasets, but given data processing times are at 3 days per scan (with current software and computer hardware). The second approach takes less computation time, but only provides limited information. Regardless, as the theoretical use for this discovery is rapid individual sorting of apples via optical instruments, the sensitivity of the method for assessing porosity of variation within a cultivar should be verified on a larger set of samples (via the second method mentioned above). Moreover, certain apple cultivars with low gas diffusivity such as ‘Braeburn’ apples are particularly prone to storage disorders and quality loss even when stored in controlled atmospheric conditions (Ho et al., 2010). Given that the occurrence of the disorders varies from year to year and batch to batch (Herremans et al., 2013; van Dael et al., 2018), physiological variation in porosity is likely a key factor. Having the ability to measure porosity optically may thus be particularly useful to improve long-term storage of disorder-prone cultivars in a cost effective manner. However, it should be verified that the relation between porosity and the scattering coefficient is stable over different batches and seasons. Adjusting storage conditions to account for actual porosity of the batch being stored is easier to implement than elaborate systems such as dynamically controlled atmospheric (DCA) storage. Further, sorting fruit based on porosity may enable to group fruit in storage rooms according to similar porosity and thus increasing uniformity that may also facilitate DCA.

4.3. Optical measurement of porosity and its applications In long-term storage of apples, gas transfer into and out of the fruit is carefully controlled so that respiration is minimized without pushing the fruit into fermentation (Bessemans et al., 2016; Jayas and Jeyamkondan, 2002). It has been shown that porosity influences the gas transfer dynamics, because the internal pore network provides the interfaces in which gas exchange occurs (Ho et al., 2014, 2011). The porosity thus determines the optimal storage parameters (Ho et al., 2010; Thewes et al., 2015). Biological variation is inevitable and real porosity of the fruit can significantly drift out of the pre-determined and cultivar specific range. Thus, measurement of the fruit porosity may be useful for optimizing the storage conditions. To date, the only instruments that have been shown to measure and quantify internal porosity accurately are micro-CT (Herremans et al., 2015; Mendoza et al., 2007; Ting et al., 2013) and magnetic resonance imaging (Musse et al., 2010). Unfortunately, this kind of system is expensive and low in throughput, limiting its use in industrial applications. Falling back on optical measurements would be a more viable option. The results obtained in this study suggest that the reduced scattering coefficient scales linearly with porosity both between and within the cultivars. Optical measurements thus can be used to determine the porosity of individual fruit. Moreover, optical instruments are flexible in implementation and can be either machine mounted or handheld (McCormick and Biegert, 2018; Yan and Siesler, 2018). To this extent, Van Beers et al. (2017) tested a multispectral handheld sensor for SRS measurements on apple fruit, showing successful estimation of the BOP. In combination with its measurement and processing speed, this makes the technique suitable for large scale product sorting to bin and store fruit with similar porosity together. The reduced porosity variation within the batch could allow to tighten the control of storage conditions and by extension significantly reduce the losses due to fermentation.

5. Conclusions In this paper, light scattering was correlated to parameters relating to the subsurface microstructure of four different apple cultivars: ‘Kanzi’, ‘Braeburn’, ‘Jonagold’ and ‘Golden Delicious’. The reduced scattering coefficient μs’ was estimated from SRS measurements using a metamodeling approach, while eight parameters related to apple microstructure were extracted from contrast-enhanced micro-CT measurements at the exact same location. The enhanced micro-CT measurements provided numerical data on cell geometry and size, as well as porosity. From the extracted quantitative data, porosity was determined to be the structural parameter with the greatest variability between the four examined cultivars. Subsequently, using the regression coefficients from a PLS model using 8

Postharvest Biology and Technology 159 (2020) 110996

Z. Wang, et al.

the microstructural parameters as variables for the prediction of μs’, it was clear that both porosity and pore surface density were indeed the most important physical parameters for predicting the reduced scattering coefficient. Moreover, clear positive linear relationships between porosity / pore surface density and the reduced scattering coefficient (R2 = 0.89) were established. To our knowledge, this is the first study directly linking scattering information to microstructural data of apple subsurface layers. These results show the benefit of a complementary dataset, using information at a corresponding spatial location on the fruit. The observed strong correlation between porosity and scattering measurements can be utilized on its own to improve long-term storage strategies. However, additional assessments on the relationship between porosity, gas-exchange in long-term storage and the final quality of the stored fruit are necessary to ensure the usefulness and accuracy of optical evaluation techniques. As the current study is only a fraction of the necessary work required to fully exploit optical instruments in preserving fruit quality, further research is required to confirm these results and exploit their full potential.

Hale, G.M., Querry, M.R., 1973. Optical constants of water in the 200-nm to 200-μm wavelength region. Appl. Opt. 12, 555–563. https://doi.org/10.1364/AO.12.000555. Herremans, E., Verboven, P., Bongaers, E., Estrade, P., Verlinden, B.E., Wevers, M., Hertog, M.L.A.T.M., Nicolai, B.M., 2013. Characterisation of ‘Braeburn’ browning disorder by means of X-ray micro-CT. Postharvest Biol. Technol. 75, 114–124. https://doi.org/10.1016/j.postharvbio.2012.08.008. Herremans, E., Verboven, P., Defraeye, T., Rogge, S., Ho, Q.T., Hertog, M.L.A.T.M., Verlinden, B.E., Bongaers, E., Wevers, M., Nicolai, B.M., 2014. X-ray CT for quantitative food microstructure engineering: the apple case. Nucl. Instruments Methods Phys. Res. Sect. B Beam Interact. with Mater. Atoms 324, 88–94. https://doi.org/10. 1016/j.nimb.2013.07.035. Herremans, E., Verboven, P., Verlinden, B.E., Cantre, D., Abera, M., Wevers, M., Nicolaï, B.M., 2015. Automatic analysis of the 3-D microstructure of fruit parenchyma tissue using X-ray micro-CT explains differences in aeration. BMC Plant Biol. 15, 264. https://doi.org/10.1186/s12870-015-0650-y. Ho, Q.T., Verboven, P., Fanta, S.W., Abera, M.K., Retta, M.A., Herremans, E., Defraeye, T., Nicolaï, B.M., 2014. A multiphase pore scale network model of gas exchange in apple fruit. Food Bioprocess Technol. 7, 482–495. https://doi.org/10.1007/s11947-0121043-y. Ho, Q.T., Verboven, P., Verlinden, B.E., Herremans, E., Wevers, M., Carmeliet, J., Nicolaï, B.M., 2011. A three-dimensional multiscale model for gas exchange in fruit. Plant Physiol. 155, 1158–1168. https://doi.org/10.1104/pp.110.169391. Ho, Q.T., Verboven, P., Verlinden, B.E., Schenk, A., Delele, M.A., Rolletschek, H., Vercammen, J., Nicolaï, B.M., 2010. Genotype effects on internal gas gradients in apple fruit. J. Exp. Bot. 61, 2745–2755. https://doi.org/10.1093/jxb/erq108. Janssens, E., Alves Pereira, L.F., De Beenhouwer, J., Tsang, I.R., Van Dael, M., Verboven, P., Nicolaï, B., Sijbers, J., 2016. Fast inline inspection by Neural Network Based Filtered Backprojection: application to apple inspection. Case Stud. Nondestruct. Test. Eval. https://doi.org/10.1016/j.csndt.2016.03.003. Jayas, D.S., Jeyamkondan, S., 2002. Modified atmosphere storage of grains meats fruits and vegetables. Biosyst. Eng. 82, 235–251. https://doi.org/10.1006/bioe.2002.0080. Kader, A., Zagory, D., Kerbel, E.L., 1989. Modified atmosphere packaging of fruits and vegetables. Crit. Rev. Food Sci. Nutr. https://doi.org/10.1080/10408398909527506. Lu, R., 2016. Overview of Light Interaction With Food and Biological Materials, in: Light Scattering Technology for Food Property, Quality and Safety Assessment, Contemporary Food Engineering. CRC Press, pp. 19–41. https://doi.org/10.1201/ b20220-3. Marques, E.J.N., De Freitas, S.T., Pimentel, M.F., Pasquini, C., 2016. Rapid and non-destructive determination of quality parameters in the “Tommy Atkins” mango using a novel handheld near infrared spectrometer. Food Chem. 197, 1207–1214. https:// doi.org/10.1016/j.foodchem.2015.11.080. McCormick, R., Biegert, K., 2018. Monitoring the growth and maturation of apple fruit on the tree with handheld Vis/NIR devices. Nir News. https://doi.org/10.1177/ 0960336018814147. 096033601881414. McGlone, V.A., Jordan, R.B., Martinsen, P.J., 2002. Vis/NIR estimation at harvest of preand post-storage quality indices for ‘Royal Gala’ apple. Postharvest Biol. Technol. 25, 135–144. https://doi.org/10.1016/S0925-5214(01)00180-6. Mendoza, F., Verboven, P., Mebatsion, H.K., Kerckhofs, G., Wevers, M., Nicolaï, B., 2007. Three-dimensional pore space quantification of apple tissue using X-ray computed microtomography. Planta 226, 559–570. https://doi.org/10.1007/s00425-0070504-4. Musse, M., De Guio, F., Quellec, S., Cambert, M., Challois, S., Davenel, A., 2010. Quantification of microporosity in fruit by MRI at various magnetic fields: comparison with X-ray microtomography. Magn. Reson. Imaging 28, 1525–1534. https:// doi.org/10.1016/j.mri.2010.06.028. Nguyen Do Trong, N., Erkinbaev, C., Tsuta, M., De Baerdemaeker, J., Nicolaï, B., Saeys, W., 2014. Spatially resolved diffuse reflectance in the visible and near-infrared wavelength range for non-destructive quality assessment of “Braeburn” apples. Postharvest Biol. Technol. 91, 39–48. https://doi.org/10.1016/j.postharvbio.2013. 12.004. Nicolaï, B.M., Beullens, K., Bobelyn, E., Peirs, A., Saeys, W., Theron, K.I., Lammertyn, J., 2007. Nondestructive measurement of fruit and vegetable quality by means of NIR spectroscopy: a review. Postharvest Biol. Technol. 46, 99–118. https://doi.org/10. 1016/j.postharvbio.2007.06.024. Nicolaï, B.M., Verlinden, B.E., Desmet, M., Saevels, S., Saeys, W., Theron, K., Cubeddu, R., Pifferi, A., Torricelli, A., 2008. Time-resolved and continuous wave NIR reflectance spectroscopy to predict soluble solids content and firmness of pear. Postharvest Biol. Technol. 47, 68–74. https://doi.org/10.1016/j.postharvbio.2007.06.001. Nugraha, B., Verboven, P., Janssen, S., Wang, Z., Nicolaï, B.M., 2019. Non-destructive porosity mapping of fruit and vegetables using X-ray CT. Postharvest Biol. Technol. 150, 80–88 https://doi.org/S0925521418311323. Schotsmans, W., Verlinden, B.E., Lammertyn, J., Nicolaï, B.M., 2004. The relationship between gas transport properties and the histology of apple. J. Sci. Food Agric. 84, 1131–1140. https://doi.org/10.1002/jsfa.1768. Studman, C.J., 2001. Computers and electronics in postharvest technology - A review. Comput. Electron. Agric. 30, 109–124. https://doi.org/10.1016/S0168-1699(00) 00160-5. Sun, J., Künnemeyer, R., McGlone, A., 2017. Optical methods for firmness assessment of fresh produce: a review, in: postharvest handling. InTech. https://doi.org/10.5772/ intechopen.69256. Sun, J., Künnemeyer, R., McGlone, A., Rowe, P., 2016. Multispectral scattering imaging and NIR interactance for apple firmness predictions. Postharvest Biol. Technol. 119, 58–68. https://doi.org/10.1016/j.postharvbio.2016.04.019. Thewes, F.R., Both, V., Brackmann, A., Weber, A., De Oliveira Anese, R., 2015. Dynamic controlled atmosphere and ultralow oxygen storage on “Gala” mutants quality maintenance. Food Chem. 188, 62–70. https://doi.org/10.1016/j.foodchem.2015.

Acknowledgements Zi Wang was funded by the KU Leuven Research Council in the context of C1 project C16/16/002. Robbe Van Beers was supported by both the Institute for the Promotion of Innovation through Science and Technology (IWT-Flanders, SB grant 131777) and KU Leuven (PDM/ 18/119). Ben Aernouts was funded as a postdoctoral fellow of the Research Foundation Flanders (FWO, Brussels, Belgium, grant12K3916N). References Adebayo, S.E., Hashim, N., Abdan, K., Hanafi, M., 2016. Application and potential of backscattering imaging techniques in agricultural and food processing – a review. J. Food Eng. 169, 155–164. https://doi.org/10.1016/j.jfoodeng.2015.08.006. Aernouts, B., Erkinbaev, C., Watté, R., Van Beers, R., Do Trong, N.N., Nicolai, B., Saeys, W., 2015. Estimation of bulk optical properties of turbid media from hyperspectral scatter imaging measurements: metamodeling approach. Opt. Express 23, 26049. https://doi.org/10.1364/OE.23.026049. Aernouts, B., Trong, N.N.D., Watté, R., Bruggeman, W., Tsuta, M., Verboven, P., Nicolai, B., Saeys, W., 2011. Food quality control by combining light propagation models with multiple vis/NIR reflectance measurements. Nir News 22, 14–16. https://doi.org/10. 1255/nirn.1237. Aernouts, B., Watté, R., Van Beers, R., Delport, F., Merchiers, M., De Block, J., Lammertyn, J., Saeys, W., 2014. Flexible tool for simulating the bulk optical properties of polydisperse spherical particles in an absorbing host: experimental validation. Opt. Express 22, 20223. https://doi.org/10.1364/OE.22.020223. Aernouts, B., Zamora-Rojas, E., Van Beers, R., Watté, R., Wang, L., Tsuta, M., Lammertyn, J., Saeys, W., 2013. Supercontinuum laser based optical characterization of Intralipid® phantoms in the 500-2250 nm range. Opt. Express 21, 32450–32467. https://doi.org/10.1364/OE.21.032450. Bessemans, N., Verboven, P., Verlinden, B.E., Nicolaï, B.M., 2016. A novel type of dynamic controlled atmosphere storage based on the respiratory quotient (RQ-DCA). Postharvest Biol. Technol. 115, 91–102. https://doi.org/10.1016/j.postharvbio. 2015.12.019. Boas, F.E., Fleischmann, D., 2012. CT artifacts: causes and reduction techniques. Imaging Med. 4, 229–240. https://doi.org/10.2217/iim.12.13. Bobelyn, E., Serban, A.-S., Nicu, M., Lammertyn, J., Nicolai, B.M., Saeys, W., 2010. Postharvest quality of apple predicted by NIR-spectroscopy: study of the effect of biological variability on spectra and model performance. Postharvest Biol. Technol. 55, 133–143. https://doi.org/10.1016/j.postharvbio.2009.09.006. Cen, H., Lu, R., Mendoza, F., Beaudry, R.M., 2013. Relationship of the optical absorption and scattering properties with mechanical and structural properties of apple tissue. Postharvest Biol. Technol. 85, 30–38. https://doi.org/10.1016/j.postharvbio.2013. 04.014. Couckuyt, I., Forrester, A., Gorissen, D., De Turck, F., Dhaene, T., 2012. Blind kriging: implementation and performance analysis. Adv. Eng. Softw. 49, 1–13. https://doi. org/10.1016/j.advengsoft.2012.03.002. Cubeddu, R., D’Andrea, C., Pifferi, A., Taroni, P., Torricelli, A., Valentini, G., Dover, C., Johnson, D., Ruiz-Altisent, M., Valero, C., 2001. Nondestructive quantification of chemical and physical properties of fruits by time-resolved reflectance spectroscopy in the wavelength range 650–1000 nm. Appl. Opt. 40, 538. https://doi.org/10.1364/ AO.40.000538. De Schryver, T., Dhaene, J., Dierick, M., Boone, M.N., Janssens, E., Sijbers, J., van Dael, M., Verboven, P., Nicolai, B., Van Hoorebeke, L., 2016. In-line NDT with X-Ray CT combining sample rotation and translation. NDT E Int. 84, 89–98. https://doi.org/10. 1016/j.ndteint.2016.09.001.

9

Postharvest Biology and Technology 159 (2020) 110996

Z. Wang, et al.

P., Nicolai, B.M., 2008. Three-dimensional gas exchange pathways in pome fruit characterized by synchrotron X-Ray computed tomography. Plant Physiol. 147, 518–527. https://doi.org/10.1104/pp.108.118935. Verboven, P., Nemeth, A., Abera, M.K., Bongaers, E., Daelemans, D., Estrade, P., Herremans, E., Hertog, M., Saeys, W., Vanstreels, E., Verlinden, B., Leitner, M., Nicolaï, B., 2013. Optical coherence tomography visualizes microstructure of apple peel. Postharvest Biol. Technol. 78, 123–132. Wang, A., Lu, R., Xie, L., 2017a. Improved algorithm for estimating the optical properties of food products using spatially-resolved diffuse re fl ectance. J. Food Eng. https:// doi.org/10.1016/j.jfoodeng.2017.05.005. Wang, Z., Herremans, E., Janssen, S., Cantre, D., Verboven, P., Nicolaï, B., 2018. Visualizing 3D food microstructure using tomographic methods: advantages and disadvantages. Annu. Rev. Food Sci. Technol. 9, 323–343. https://doi.org/10.1146/ annurev-food-030117-012639. Wang, Z., Verboven, P., Nicolai, B., 2017b. Contrast-enhanced 3D micro-CT of plant tissues using different impregnation techniques. Plant Methods 13, 105. https://doi. org/10.1186/s13007-017-0256-5. Watkins, C.B., Nock, J.F., 2012. Controlled-atmosphere storage of ‘Honeycrisp’ apples. HortScience 47, 886–892. https://doi.org/10.21273/HORTSCI.47.7.886. Watté, R., Aernouts, B., Van Beers, R., Saeys, W., 2015. Robust metamodel-based inverse estimation of bulk optical properties of turbid media from spatially resolved diffuse reflectance measurements. Opt. Express 5, 27880–27898. https://doi.org/10.1364/ OME.5.0027880. Watté, R., Nguyen Do Trong, N., Aernouts, B., Erkinbaev, C., De Baerdemaeker, J., Nicolaï, B., Saeys, W., 2013. Metamodeling approach for efficient estimation of optical properties of turbid media from spatially resolved diffuse reflectance measurements. Opt. Express 21, 32630. https://doi.org/10.1364/OE.21.032630. Yan, H., Siesler, H.W., 2018. Hand-held near-infrared spectrometers: state-of-the-art instrumentation and practical applications. Nir News. https://doi.org/10.1177/ 0960336018796391. 096033601879639.

04.128. Ting, V.J.L., Silcock, P., Bremer, P.J., Biasioli, F., 2013. X-ray micro-computer tomographic method to visualize the microstructure of different apple cultivars. J. Food Sci. 78. https://doi.org/10.1111/1750-3841.12290. Torricelli, A., Contini, D., Dalla Mora, A., Tamborini, D., Villa, F., Tosi, A., Spinelli, L., 2015. Recent advances in time-resolved NIR spectroscopy for nondestructive assessment of fruit quality. Chem. Eng. Trans. 44, 43–48. https://doi.org/10.3303/ CET1544008. Tuchin, V., 2007. Methods and algorithms for the measurement of the optical parameters of tissues. Tissue Optics 143–256. https://doi.org/10.1117/3.684093.ch2. Van Beers, R., Aernouts, B., León Gutiérrez, L., Erkinbaev, C., Rutten, K., Schenk, A., Nicolaï, B., Saeys, W., 2015. Optimal illumination-detection distance and detector size for predicting braeburn apple maturity from Vis/NIR laser reflectance measurements. Food Bioprocess Technol. 8, 2123–2136. https://doi.org/10.1007/ s11947-015-1562-4. Van Beers, R., Aernouts, B., Watté, R., Schenk, A., Nicolaï, B., Saeys, W., 2017. Effect of maturation on the bulk optical properties of apple skin and cortex in the 500–1850 nm wavelength range. J. Food Eng. 214, 79–89. https://doi.org/10.1016/j.jfoodeng. 2017.06.013. van Dael, M., Verboven, P., Dhaene, J., Van Hoorebeke, L., Sijbers, J., Nicolai, B., 2017. Multisensor X-ray inspection of internal defects in horticultural products. Postharvest Biol. Technol. 128, 33–43. https://doi.org/10.1016/j.postharvbio.2017.02.002. van Dael, M., Verboven, P., Zanella, A., Sijbers, J., Nicolai, B., 2018. Combination of shape and X-ray inspection for apple internal quality control: in silico analysis of the methodology based on X-ray computed tomography. Postharvest Biol. Technol. https://doi.org/10.1016/j.postharvbio.2018.05.020. Veraverbeke, E.A., Verboven, P., Van Oostveldt, P., Nicolaı ̈, B.M., 2003. Prediction of moisture loss across the cuticle of apple (Malus sylvestris subsp. Mitis (Wallr.)) during storage: part 2. Model simulations and practical applications. Postharvest Biol. Technol. 30, 89–97. https://doi.org/10.1016/S0925-5214(03)00082-6. Verboven, P., Kerckhofs, G., Mebatsion, H.K., Ho, Q.T., Temst, K., Wevers, M., Cloetens,

10