Automatic computation of mandibular indices in dental panoramic radiographs for early osteoporosis detection

https://doi.org/10.1016/j.artmed.2020.101816Get rights and content

Highlights

Abstract

Aim

A new automatic method for detecting specific points and lines (straight and curves) in dental panoramic radiographies (orthopantomographies) is proposed, where the human knowledge is mapped to the automatic system. The goal is to compute relevant mandibular indices (Mandibular Cortical Width, Panoramic Mandibular Index, Mandibular Ratio, Mandibular Cortical Index) in order to detect the thinning and deterioration of the mandibular bone. Data can be stored for posterior massive analysis.

Methods

Panoramic radiographies are intrinsically complex, including: artificial structures, unclear limits in bony structures, jawbones with irregular curvatures and intensity levels, irregular shapes and borders of the mental foramen, irregular teeth alignments or missing dental pieces. An intelligent sequence of linked imaging segmentation processes is proposed to cope with the above situations towards the design of the automatic segmentation, making the following contributions: (i) Fuzzy K-means classification for identifying artificial structures; (ii) adjust a tangent line to the lower border of the lower jawbone (lower cortex), based on texture analysis, grey scale dilation, binarization and labelling; (iii) identification of the mental foramen region and its centre, based on multi-thresholding, binarization, morphological operations and labelling; (iv) tracing a perpendicular line to the tangent passing through the centre of the mental foramen region and two parallel lines to the tangent, passing through borders on the mental foramen intersected by the perpendicular; (v) following the perpendicular line, a sweep is made moving up the tangent for detecting accumulation of binary points after applying adaptive filtering; (vi) detection of the lower mandible alveolar crest line based on the identification of inter-teeth gaps by saliency and interest points feature description.

Results

The performance of the proposed approach was quantitatively compared against the criteria of expert dentists, verifying also its validity with statistical studies based on the analysis of deterioration of bone structures with different levels of osteoporosis. All indices are computed inside two regions of interest, which tolerate flexibility in sizes and locations, making this process robust enough.

Conclusions

The proposed approach provides an automatic procedure able to process with efficiency and reliability panoramic X-Ray images for early osteoporosis detection.

Introduction

Osteoporosis is a disease of low bone mineral mass and microarchitectural deterioration of bone, which leads to increased fracture risk [1]. The oral implications of osteoporosis include loss of teeth, loss in alveolar bone height, erosion of inferior mandibular cortex, reduced mandibular inferior cortical width and are broadly reported in the literature [2]. Dental panoramic radiographies provide excellent information to identify evidences of osteoporosis, where different indices can be obtained for early detection. A number of mandibular cortical indices, including the mandibular cortical width (MCW), panoramic mandibular index (PMI), mandibular ratio (M/M) and mandibular cortical width (MCI) have been developed to assess and quantify the quality of mandibular bone mass and to observe signs of resorption on panoramic radiographs for identification of osteopenia [3]. We propose a new automatic procedure where the human knowledge is directly mapped to the computer-based system, with the aim of data can be stored for posterior analysis.

It is well known that automatic imaging analysis from dental radiographies represents a challenge. In this regard, at least two grand challenges were held in dental X-ray image analysis automatization sponsored by the IEEE International Symposium on Biomedical Imaging 2015 [4]. The first one was intended for detecting anatomical landmarks and analysis for cephalometric and the second for detecting caries in bitewing radiographies. In this regard the full automatic process must be conveniently designed in order to apply the required intelligence or reasoning for early osteoporosis detection in patients, still requiring an extra effort [5], where this paper is focused. Osteoporosis studies have been addressed based on different indices from dental panoramic radiographs and have been defined and validated in the scientific dentistry community [1,[6], [7], [8], [9], [10], [11], [12], [13]]. Fig. 1(a) displays an original radiography with a framework where indices are to be computed and Fig. 1(b) interest lines (a, b, c, d, e, f) and points (A, B, C, D, E, F) on such framework. The following is the list of relevant indices to be computed, on which a comparative analysis is performed:

  • a)

    Mandibular Cortical Width (MCW) or Mental Index (MI) or Cortical Width Index or Mandibular Cortical Thickness [3,[14], [15], [16]]: width of the lower border of the mandible below the two mental foramina. Distance between points A and B, d(B,A), Fig. 1(b).

  • b)

    Panoramic Mandibular Index (PMI) [17,18]: ratio of the width of the mandibular cortex, d(A,B), to the distance from the mental foramen to the inferior margin of the mandibular cortex. Two PMI values can be measured, Fig. 1(b), distinguishing between superior margin (PMIs, involving the distance between points E and A, d(E,A)) and inferior margin (PMIi, distance between points C and A, d(C,A)), depending on whether the distance is measured from above or below limits from the foramen. The final PMI index is computed as the average value of both, PMIs (=d(B,A)/ d(E,A)) and PMIi (=d(B,A)/ d(C,A)), i.e. PMI = (PMIs + PMIi)/2

  • c)

    Mandibular ratio (M/M) [8,19]: total mandibular height (distance between points F and A, d(F,A)) divided by the height from the center of the mental foramen to the inferior mandibular border, distance between points D and A, d(D,A), i.e. M/M = d(F,A)/ d(D,A).

  • d)

    Mandibular Cortical Index (MCI), Klemetti index (KI) or Cortical Erosion [16,20,21]: to determine the mandible bone porosity in the mental region, assuming that osteoporosis is directly related to low bone mass and deterioration of the bone tissue. Detection of local variations in gray intensity levels is the main chosen approach as an alternative method to the computation of bone mineral density, which is generally determined using dual energy X-ray absorptiometry [22]. IDIOS [23] provides a reference score of 64.77 % with significant evidences of osteoporosis.

Regardless of the effectiveness of the different indices to assess the degree or level of osteoporosis, [24,25], from the point of view of automatic image segmentation and computation, the problem is constrained to the identification of specific structures and their geometric measurements (morphometric analysis), ratios (MCW, PMI, M/M) and textures (KI) in the mandible. In [26] authors reported that PMI is only useful when the margin of the mental foramen is clear. On the contrary MCW has fewer drawbacks than PMI. This means that the automatic image segmentation must be designed to cope with this drawback. The following structures are to be automatically segmented and identified as follows, where lowercase and uppercase letters are referred to Fig. 1(b):

  • a)

    Straight parallel line, a, following the long axis of the mandible and tangential to the inferior border, i.e. lower jaw.

  • b)

    Mental foramen (Mf) region with its central point (D) defined by this structure.

  • c)

    Straight perpendicular line, e, to the line a passing through point D, intersecting line a at point A.

  • d)

    Straight parallel line, b, to the line a through the point B in the upper border of the mental region (mandibular cortex).

  • e)

    Two straight parallel lines c and d to the line a passing through lower (C) and upper (E) points defined by the intersection of line e with the lowest and most superior points respectively defining Mf.

  • f)

    Curved or straight line, f, defining the border that separates the gum from the exposed part of the tooth (gum-line).

  • g)

    Texture characterization for porosity determination at the region below the mental foramen.

In [27] authors developed a manual computer aided diagnosis system for determining MCW based on gradient analysis of edges. Active shape contour models were early used in [28,29], where the main problem is to envelop (identify) the region to be segmented where the contour is to be fitted. In [30] a point distribution model was built by manual annotation of endosteal and periosteal edges, used to guide the shape of the cortex and to define the mental foramen region, i.e. for mandibular cortex localization, this approach assumes the search of the desired structures by means of a kind of mask, without considering that many of them are complex in nature presenting different irregularities. Kavitha et al. [31,32] proposed an image segmentation approach consisting of five steps: 1) contrast stretching on a ROI around the mental foramen; 2) image enhancement, based on histogram equalization, thresholding, and high-pass filtering; 3) cortical margin measurement, where the Chessboard distance, with dynamic programming, is used for determining the medial axis of the cortical bone, and then the cortical margins; 4) distance measurement, where a second-order polynomial is fitted to the upper boundary, which allows to measure the cortical width at each point along the tangential line to the lower boundary; 5) these distances are used as features in a Support Vector Machines (SVM) classifier for determining bones affected or not by osteoporosis. The main drawback of this method is the isolation of the cortical bone region, because of the irregular, and most times whimsical shapes. Also Kavitha et al. in [33,34] determined the width of the inferior cortical structure based on dynamic programming, based on a continuous distance evaluation through the region of interest where the cortex is located, but this requires a clear definition of region’s boundaries. In [35] authors proposed an automatic approach for mandible segmentation in panoramic X-ray based on horizontal integral projections, with a modified Canny edge detector followed by morphological operations, and guided by a template manually designed by three expert dentists. Low-quality X-rays images, blurred or with the patient malposed, were excluded from the dataset. A mandibular mask, guiding the image segmentation process, followed by a modified Canny edge detector and active contour models (by adjusting a set of control points) was also the approach applied [36], contour edges require a clear differentiation on image intensity levels, so that the contour can follows and adjust the border of the mandible. Muramatsu et al., [15], measure the MCW by applying the Canny edge detector on previously defined ROIs for detecting the lower border of the mandible progressing upwards, under the assumption that residues appear when the cortex begins its erosion. Piecewise lines appear, which are detected by convergence index filter for gradient vector fields [37], while determining ridged textural features. The border to be detected appears on the densest parts (i.e. without ridges). The main handicap of this approach is that dense and homogenous parts are required without ridges and also that patients begin to have osteoporosis. Also the determination of ROIs is critical for fitting the edges.

Naik et al., [38], applied different image processing techniques for automatic segmentation of lower jaw and then mandibular bone in dental digital panoramic X-rays. Five strips are located on lower jaw, where the central strip guides the process for segmentation the lower jaw from upper jaw. The cutting position of lower jaw, from the central part, is located after applying Otsu, [39], thresholding followed by horizontal projection to determine a valley near bottom end. Edge points in the mandible are located through the Canny operator with identification of intensities on this part. Edge detection operators were applied in [40] to measure the cortical width identifying lower/upper in lower jawbone for MCW computation. In [41] selective edges in three directions (45°, 90°, 135°) were extracted by Laplacian of Gaussian second derivative detector [42], with different standard deviations, a path following brightness values is adjusted from a starting point. It is well known that edge extractors (first and second derivative) are strongly affected by noise or undesired effects and the identification of control points for adjusting the active contour or starting points represents a serious challenge. Moreover, capricious forms of lower jawbone border greatly difficult its detection.

In [43,44] a segmentation process is applied on a ROI to separate trabecular and porous areas based on top/bottom-hat filtering, followed by histogram expansion and adaptive thresholding for contrast enhancement. A binary image is obtained and analysed based on the distribution of black/whites areas to determine densities. Co-ocurrence matrices were also proposed for texture description followed by fuzzy-based classification approaches [45], or based on Bayes, Fuzzy K-means and SVM [46]. Texture based descriptors followed by classification methods, including also SVM, was a recurrent approach on this regard [47,48]. The setting of the involved parameters in the above classification approach is always a handicap. Related to MCI [49] and [50] summarize the appearance of the cortical bone, below the mandibular foramen, using the classification commonly accepted by the scientific community as follows [10,51,52]: C1 (normal), when the endosteal margin is even and well differentiated; C2 (moderately eroded) because of endosteal cortical residues due to lacunar resorption; C3 (severely eroded), with high level of porosity. Correlation between MCW and also osteoporosis and porosity levels were studied in [53,54] with a clear positive correlation. Muramatsu et al. [15] identify ridges at the upper border limit of the mandibular foramen regions as textural features for porosity description. In this regard, Fig. 2 displays three illustrative examples (a), (b) and (c) with different degrees of porosity that could be associated to types C1, C2 and C3 respectively corresponding to three segmented ROIs, as described below. With identical purpose, [55] applied high-pass filtering to enhance underlying structures, i.e. edges. High pass filtering is very noise sensitive depending always of some image threshold that must be previously stablished, which can be avoided by applying adaptive filtering [56]. Indeed, once the ROI is selected, Fig. 4(a), the local variance image (texture description) on an 8-neighbourhood for each pixel is computed, followed by global histogram equalization, obtaining the image (Iv), Fig. 2(a–c) upper parts, where foreground and background can be distinguished by considering local variations (key issue) to obtain the binary image (Ib), Fig. 2(a–c) lower parts. This is carried out as follows:

  • a)

    Apply a mean filter to Iv with size w (set to 11 in this work) to obtain Im (a median filter is also feasible).

b) Ib=1ifImIvst0otherwisewithstthe standard deviation ofIv

We can see from the above figures, how porosity can be identify on the elongated jaw bone regions, so that the density of black points increases according to the qualification in C1, C2 and C3.

IDIOS (Index of Dental-imaging Indices of Osteoporosis Screening) [23], is a method to validate indices in osteoporosis, provides the highest score for MCW (78.32 %) from the 104 indices analyzed and the fourth position for MCI (64.77 %). PMI and M/M achieved respectively 59.84 % and 50.00 %. IDIOS uses: a) statistical (sensitivity and specificity) values to determine positive and negative cases of osteoporosis accurately; b) kappa statistics and correlations to measure when an observer obtains same scores on same subjects; c) objectivity, based on measures and calculations; d) use of software for analysis; e) differentiation between bone fragility groups (normal, osteopenia, osteoporosis). Under these criteria the above scores can be clearly improved if an automatic methodology, as the one presented in this paper, is applied. Additionally [2], reported that there is not agreement in evaluating mandibular cortical indices, causing discrepancies and differences. In this regard a procedure based on imaging processing can contribute to minimize this effect. This serves as reinforcement to justify this proposal.

Naik et al. [38] pointed out that X-ray panoramic images are complex enough from the point of view of image segmentation, requiring the specific combination and integration of several image segmentation methods to achieve the proposed goal. This is the approach addressed on this paper. As described below in Section 2.1, technically these images are complex by different reasons.

Under the above assumptions we have developed an automatic image processing strategy for computing three indices (MCW, PMI, M/M) and also to determine porosities, aligned with KI. The main contributions defining the full process are synthetized in the following steps, as parts of it:

  • 1)

    Identify dental artificial structures (implant, fillings, crowns), to be considered during the subsequent processes. This is carried out by applying a fuzzy K-means classification approach, [57], where four classes are automatically segmented based on intensity brightness with a previous training process and assuming these artificial structures display the higher intensity levels. Higher intensity levels are replaced by mean intensity values surrounding these areas. The training, which learns specific parameters, is only required (executed) one time for each type of images, i.e. for each device.

  • 2)

    Adjust tangent straight lines to the lower border of the lower jawbone as follows: a) find an elongated region defining the bone by computing the variance (texture analysis) in the image, assuming the mandibular cortex contains a sufficient degree of consistence although it is affected by porosity; b) apply grey-scale dilation to enhance the lower border [42], followed by binarization and labelling to the elongated region with high variance values; c) find edges of the elongated region; d) fit up to three straight lines, based on the Hough transform [42], to the lower border to cope with up to three different curvatures if any (more curvatures should be also possible).

  • 3)

    Identify the mental foramen (Mf) region, its limits and centroid, making the following assumptions on it: a) contains low intensity levels (dark); b) tends to be circular or ellipsoidal with a high degree of solidity and without holes); c) the centroid is placed near the central part in the region. A multi-threshold strategy [39], with binarization is applied, followed by morphological operations, [42,] and region labelling, obtaining properties from the labelled regions, [50], to determine the best Mf candidate, together with its centroid (point D Fig. 1b). Determine the minimum distance from D to all tangent straight lines obtained above, selecting the corresponding line as the best one, line a Fig. 1(b).

  • 4)

    Determine the perpendicular line, e Fig. 1(b), to line a passing through D and obtain the intersection point A; considering the mean intensity level of Mf, based on the multi-thresholding process and labelling expressed above, the line e and the centroid D, determine the intersection points C and E. From D, following the direction of line e in the two opposite senses, three consecutives transitions of labels determine points C and E. Two parallel straight lines, c and d, to the line a, passing through points C and E, are obtained respectively.

  • 5)

    From point A, the line a is imaginary moved up following the line e keeping the parallelism with the original line a, obtaining the corresponding profiles along the displaced line a and on the image adaptively filtered, Fig. 2 (lower parts). When three consecutive profiles contain a given number of black points, the intermediate line is considered as the line b and the point B is obtained as the intersection between line e and b.

  • 6)

    Determine a sub-region (sROI) subsidiary to the initial ROI, where the f line (curved or straight) is to be detected. The following assumptions are made: a) inter-teeth gaps exist, with darker intensity levels than teeth and gums; b) the base of these gaps defines the line to be detected. Based on these assumptions the process is designed as follows: a) enhance the intensity of gaps by applying grey-scale erosion [42]; b) identify saliency regions corresponding to artificial artefacts in teeth with the higher intensity levels (blank or near blank parts) by applying the visual saliency concept in images [61]; c) identify interest points based on well-known SURF (Speed Up Robust Features) descriptor [62]; d) find alignments of candidate points with a posterior polynomial fitting by least-squares.

The automatic segmentation process expressed above represents the mapping of human knowledge/intelligence in the context of Artificial Intelligence, inside the computer science paradigm. The proposed approach transfers human intelligence to a machine, with the aim of early diagnosis of osteoporosis for detecting significant bone structures, making the main contribution with novelty with respect manual measurement methods. In this regard, two main levels of intelligence are considered in this work:

  • a)

    Image understanding: convenient sequencing of advanced image processing and computer vision techniques to extract the underlying embedded knowledge in the raw images, sometimes hidden for the human analyst.

  • b)

    Expert system: first step of this kind of intelligent systems where mandibular indices are data, coming from observations, for posterior analysis through inferential statistics in the context of inductive reasoning or machine learning for decision making after training.

Additionally, specific intelligent-based techniques are included in the sequenced process: K-means for excluding artificial structures and decision making for determining the best Mf candidate (points 1 and 3 above).

The outline of this paper is organized as follows. Section 2 describes the materials used and also the proposed approach. Section 3 provides the results including quantitative evaluation results. Conclusions and future works are given in Section 4.

Section snippets

Materials

This study was carried out on a set of 370 dental panoramic digital radiographs from as many patients acquired with a Villa Sistemi Medicali. Radiograph EVO DIGITAL, CEI-OPX/105 machine with 72 kV, 6 mA and 14.4 s of exposure. They were stored in JPEG format with image resolution of 1536 × 2573 pixels as intensity images with sufficient quality.

The images were processed by Matlab R2018b [63], using an Intel Core i7 2.0 GHz processor, 8 GB RAM and Windows 10 Pro operating system (64-bits). These

Image description and calibration

A total of 370 images have been analysed, all captured as described in Section 2.1. Each image contains a particular and peculiar morphology with different singularities and features.

As mentioned before (Section 2.2), two calibration processes are applied in order to: a) define ROIs limits; b) stablish correspondence between pixels in the image and length units (expressed in mm) in the patient's jaws in vertical and horizontal dimensions.

a) Calibration to determine ROI limits

With 60 of these

Conclusions

The study proposes a new automatic image processing approach in order to compute relevant mandibular indices from dental panoramic radiographies to detect the deterioration of significant bone structures. The method has been designed to cope with complex images by applying sequential linked strategies based on advanced imaging segmentation processes. Also it allows variations on sizes and locations in ROI definitions, giving it high flexibility. Significant points and lines are identified,

Declaration of Competing Interest

The authors declare no conflict of interest

Acknowledgments

This research is part of the multidisciplinary activity of G. Pajares at the Instituto de Conocimiento (Knowledge Institute) in the Complutense University. Special thanks to referees for their help, constructive criticism, and suggestions on the original version of this paper.

References (75)

  • H. Ohtsuki et al.

    Risk of osteoporosis in elderly individuals attending a dental clinic

    Int Dent J

    (2017)
  • E. Jagelavičienė et al.

    Relationship between the mandibular cortical index and calcaneal bone mineral density in postmenopausal women

    Medicine

    (2016)
  • H. Bay et al.

    SURF: speeded up robust features

    Comput Vis Image Underst

    (2008)
  • S. Hinojosa et al.

    Unassisted thresholding based on multi-objective evolutionary algorithms

    Knowl. Based Syst

    (2018)
  • J. Damilakis et al.

    Have panoramic indices the power to identify women with low BMD at the axial skeleton?

    Phys Med

    (2011)
  • A. Al-Dam et al.

    Mandibular cortical shape index in non-standardised panoramic radiographs for identifying patients with osteoporosis as defined by the German Osteology Organization

    J Cranio-maxillofacial Surg

    (2013)
  • S. Geary et al.

    Identifying dental panoramic radiograph features for the screening of low bone mass in postmenopausal women

    Int J Oral Maxillofac Surg

    (2015)
  • A.F. Leite et al.

    Correlations between seven panoramic radiomorphometric indices and bone mineral density in postmeno-pausal women

    Oral Surg Oral MedOral Pathol Oral Radiol Endod

    (2010)
  • NIH Consensus Development Panel on Osteoporosis Prevention, Diagnosis, and Therapy

    Osteoporosis prevention, diagnosis, and therapy

    JAMA J Am Med Assoc

    (2001)
  • Y. Ariji et al.

    Factors affecting observer agreement in morphological evaluation of mandibular cortical bone on panoramic radiographs

    Oral Radiol

    (2017)
  • A. Taguchi et al.

    Use of dental panoramic radiographs in identifying younger postmenopausal women with osteoporosis

    Osteoporos Int

    (2006)
  • ISBI

    Grand challenges in dental X-ray image analysis. IEEE international symposium on biomedical imaging 2015

    (2015)
  • L.G. Chandak et al.

    Correlation of periodontitis with mandibular radiomorphometric indices, serum calcium and serum estradiol in postmenopausal women: a case–control study

    Indian J Dent Res

    (2017)
  • M. Tofangchiha et al.

    Diagnosis of osteoporosis using cortex mandibular indices based on cortex thickness and morphology in comparison with visual assessment of the cortex

    J Craniomax Res

    (2017)
  • E. Calciolari et al.

    Panoramic measures for oral bone mass in detecting osteoporosis: a systematic review and meta-analysis

    JDR Clin. Res. Suppl.

    (2015)
  • S. Parlani et al.

    Role of panoramic radiographs in the detection of osteoporosis

    Oral Hyg Health

    (2014)
  • E. Hastar et al.

    Evaluation of mental index, mandibular cortical index and panoramic mandibular index on dental panoramic radiographs in the elderly

    Eur J Dent

    (2011)
  • J.S. Passos et al.

    Women with low bone mineral density and dental panoramic radiography

    Menopause

    (2012)
  • J. López-López et al.

    Early diagnosis of osteoporosis by means of orthopantomograms and oral x-rays: a systematic review

    Med Oral Patol Oral Cir Bucal

    (2011)
  • T.S. Tounta

    Diagnosis of osteoporosis in dental patients

    J Frailty Sarcopenia Falls

    (2017)
  • C. Muramatsu et al.

    Quantitative assessment of mandibular cortical erosion on dental panoramic radiographs for screening osteoporosis

    Int J Comput Assist Radiol Surg

    (2016)
  • E. Klemetti et al.

    Pantomography in assessment of the osteoporosis risk group

    Scand J Dent Res

    (1994)
  • E. Klemetti et al.

    Morphology of the mandibular cortex on panoramic radiographs as an indicator of bone quality

    Dentomaxillofac Radiol

    (1997)
  • Y.U. Sindhu et al.

    Digital panoramic radiography: an aid in the early detection of osteoporotic signs

    Arch Med Health Sci

    (2016)
  • I. Barngkgei et al.

    IDIOS: an innovative index for evaluating dental imaging-based osteoporosis screening indices

    Imaging Sci Dent

    (2016)
  • L. Munhoz et al.

    The use of panoramic radiographs modified by an open access software to determine Mandibular Cortical Index

    Rev Odonto Cienc

    (2017)
  • A. Taguchi

    Triage screening for osteoporosis in dental clinics usingpanoramic radiographs

    Oral Dis

    (2010)
  • Cited by (25)

    • Dependence of multifractal analysis parameters on the darkness of a processed image

      2022, Chaos, Solitons and Fractals
      Citation Excerpt :

      Gabriel-Guzman [30] have investigated image binarization threshold of a roasted coffee beans and confirms that fractal analysis can be used for indirect monitoring of the roasting degree. In order to facilitate the work of dentists and reduce the risk of osteoporosis patients, Aliaga [31] offers an automated imaging method by applying sequential linked strategies based on advanced imaging segmentation processes. Maugeri [32] used threshold-based pre-processing approaches for mouse spinal cord image analysis and was able to discriminate injured tissue from uninjured, in a manner almost independant of the chosen threshold of binarized images.

    • Gradient-adaptive Nonlinear Sharpening for Dental Radiographs Original Scientific Pap

      2023, International Journal of Electrical and Computer Engineering Systems
    View all citing articles on Scopus
    View full text