Computer reconstruction of pine growth rings using MRI

Abstract

This work explores the use of magnetic resonance imaging (MRI) for nondestructive determination of wood characteristics and for 3D wood modeling. In this context, one of the applications under development is the automatic recognition and reconstruction of rings from transversal images obtained from MRI scanners. The algorithm analyzes a set of transversal MRI images, detecting and reconstructing growth ring edges. The information generated is then interpolated in order to obtain an accurate 3D picture of the log and its fundamental constituents (individual rings, knots, defects, etc). Results also show that the technique has potential for defect recognition, providing a powerful tool for future developments in wood analysis. The results are encouraging and further research is needed to develop automatic detection not only of rings, but also of different types of defects that are of paramount importance in the sawmill and plywood industries.

Introduction

This study is part of a research project aimed at optimizing the utilization of Chile's main productive forest resource: radiata pine (Pinus radiata D. Don). The overall objective is modeling the effect of site, silviculture and genetic material on tree architecture and wood quality, considering variables such as knots distribution, size and shape and annual ring width and shape (eccentricity and irregularities).

To accomplish the latter the modeling has been divided into several parts. One of them is aimed at describing the internal structure of wood in three dimensions, i.e., the shape of annual rings, size of rings and ring irregularities due to branching.

The decision to use magnetic resonance imaging (MRI) [1], [2] to acquire the internal wood information was based on several studies [3], [4], [5], [6], [7], [8], [9], [10] that showed its potential, along with the experience of this research team.

This article describes an algorithm aimed at finding and reconstructing growth rings nondestructively, describing the ring evolution along the log from a 3D perspective.

Information on site quality, forest and wood density, effects of prunings and other forest management techniques can be obtained from ring analysis, e.g., in dense forests the growth rings tend to be thinner, reflecting the competition for sunlight.

Most of the existing computer algorithms developed for ring analysis have been implemented based on destructive methods and designed for dendrochronological studies. In those, wooden disks are cut from the log, and a small portion of them is analyzed using superficial optical scanners or CCD cameras [11], [12], [13].

Rauschkolb [14] developed an algorithm for growth ring boundary detection from a digitized image of a cross-sectional disk. Here, ring edges are searched in perpendicular directions over the surface, looking for the largest positive derivatives.

The way the image is acquired in Rauschkolb's case, has proved to be very effective, but it is unnecessarily restricted in the directions over which the rings are sought.

Using a computed tomography (CT) scanner, Som et al. [15], described two approaches for enhancing the visualization of rings. In the first one, the rings are sought along fixed directions using the Laplacian of a Gaussian filter [16], [17], [18]. The second approach extracts ridges and valleys from rings using a morphological technique, namely the top-hat transform [16], [18]. This technique detects edges between rings, but it is limited only to visualization enhancements of rings.

The use of CT for nondestructive analysis of wood has been widely reported in the literature. This is not so for MRI, where little work has been published [3], [4], [5], [6], [7], [8], [19]. Apart from cost of the equipment used in each case, no research has been reported in order to compare the performance of CT versus MRI.

According to Chang [20], MRI images have a better contrast than CT images, especially when comparing defective areas of the log. MRI is capable of detecting three different characteristics of the object being scanned (proton density, T₁ and T₂), contrary to CT scanning, which is limited to density differences.

Figure 1 illustrates the effects produced by different types of contrasts. The same log has been scanned using different value for the parameters echo time (TE), sequence repetition time (TR) and inversion time (TI).

By adjusting the scans to the proper parameters, the pith of the log can be emphasized, while still observing good contrast for the rings (Fig. 1a).

Figures 1b and 1c in turn, show a marked difference in ring and knots definition. Finally, in Fig. 1d, decay in ring and knot contrast is observed, however, the phloem surrounding the log (external bright circle) can be easily distinguished. The phloem corresponds to cells that transport sap from the leaves to the different organs.

Additionally, MRI produces images in any direction, contrary to CT, which is limited only to transversal images of the body being scanned.

Finally, Chang [3] points out how moisture content affects both technologies. High water content increases the amount of hydrogen in the log but diminishes the log density, positively affecting the quality of MRI images. This does not apply to CT images. The opposite applies when moisture content decreases. Keeping in mind that the final objective of nondestructive wood analysis is to obtain information about internal structures prior to the sawing process [21], [22] (where logs still have high moisture content), the differences stated above favor the MRI technique.

In spite of previous efforts, there are no methods attempting to identify rings by their geometric parameters, both in two and three dimensions. This work fills this gap.

Futhermore, this work demostrates, with experimental results, the potential of MRI as a tool for nondestructive internal analysis of wood characteristics, not only for research but also for industrial purposes as an alternative to the CT scanners.

The developed algorithm analyzes a set of transversal images obtained through MRI of a radiata pine log. In each image the algorithm reconstructs growth rings in 2D. After analyzing all the images and using interpolations, the growth rings are reconstructed in three dimensions. The use of MRI and the algorithm developed are novel.

Although the main objective of this article is the development of an algorithm to reconstruct the growth ring of a tree in a three dimensions, attention was also placed to defect detection, since by trying to find rings, the method has to handle those cases where defects like knots or rotten wood appear.

Segmentation methods had been widely used in defect detection [9], [10], [21], [23], [24], [25], [26], [27], [28]. Most recent research [22], [29], [30] report the use of artificial neural networks in order to combine the segmentation and labeling process.

Literature can also be found on 3D modeling of external and internal defects in the case of CT scanners. In this context, a significant contribution has been made by Jaeger et al. [31].

The novel approach used in this article for defect detection, is based in the fact that ring shape is strongly altered by the presence of defects, so the same basic principle used to identify rings can also apply to defect detection.