Академический Документы
Профессиональный Документы
Культура Документы
Purpose: One of the major factors governing implant success is the quantity and density of the host bone. The
aim of this work was to determine whether mandibular bone texture and cortical width measurements on plain
radiographs could be associated with implant failure and/or marginal bone loss. Materials and Methods: A
statistical model was built to predict implant failure; it incorporated several radiographic features of cortical
and cancellous bone texture, cortical width, and patient smoking habits. Cortical width measurements and
texture measurements of cortical and cancellous bone were made on the panoramic radiographs of 460
subjects. These were used to predict implant failure and marginal bone loss after 5 years. Receiver operating
characteristic (ROC) curve analysis and area under the curve (AUC) were used to determine the diagnostic
accuracy of the variables in predicting implant failure and marginal bone loss. Additionally, for 91 of 460
subjects with periapical radiographs, marginal bone levels around implants were measured over a 5-year
period. Results: Of the 460 patients assessed for implant failure, 29 had failed implants (93.7% success
rate). The ROC curve built from this model had a sensitivity of 62.1% and specificity of 67.5%. The AUC
from the model was 0.690 (95% confidence interval [CI] 0.597 to 0.783). A model was also built to predict
marginal bone loss. The ROC curve had 78.6% sensitivity and 74.6% specificity (AUC = 0.880, 95% CI = 0.810
to 0.953). Mandibular cortical width was not a significant predictor of either implant failure or bone loss.
Conclusion: In a retrospective analysis 5 years after implant placement, features of cancellous and cortical
bone of the mandible were significant in predicting implant failure and marginal bone loss in a sample of 460
patients. Int J Oral Maxillofac Implants 2015;30:372–377. doi: 10.11607/jomi.3604
Key words: bone density, bone loss, cancellous bone, cortical bone, dental implants, failure, mandible
© 2015 BY QUINTESSENCE PUBLISHING CO, INC. PRINTING OF THIS DOCUMENT IS RESTRICTED TO PERSONAL USE ONLY.
NO PART MAY BE REPRODUCED OR TRANSMITTED IN ANY FORM WITHOUT WRITTEN PERMISSION FROM THE PUBLISHER.
Merheb et al
© 2015 BY QUINTESSENCE PUBLISHING CO, INC. PRINTING OF THIS DOCUMENT IS RESTRICTED TO PERSONAL USE ONLY.
NO PART MAY BE REPRODUCED OR TRANSMITTED IN ANY FORM WITHOUT WRITTEN PERMISSION FROM THE PUBLISHER.
Merheb et al
This method is described briefly here. A co-occurrence represented the same statistical property of the co-oc-
matrix is a two-dimensional histogram that shows the currence matrix but at different orientations or scales,
frequency of occurrence of pairs of grey values of pix- the features were averaged. Where correlations were
els separated by a specific pixel distance in a specified found between different co-occurrence matrix mea-
direction. Entries to the histogram are accumulated sures, only one was retained in the model. Correlations
from all such pixel pairs in the region of interest. Typi- were assessed by means of a biplot based on a factor
cally, several such matrices are built to correspond to a analysis. In addition to cortical width and 35 texture
range of pixel separations and directions. This method features, smoking status was included in the model, so
is described briefly in Appendix 1, which also defines that 37 variables were available for the stepwise selec-
the texture measures used in this study; the reader is tion procedure. Measures from the left and right sides
referred to Haralick et al14 or to a number of secondary of the mandible were averaged. Each texture feature
sources (eg, Tucervan and Jain15 and Petrou and Garcia can be determined at a specific pixel separation and
Sevila16) for a complete description and a definition of orientation. The definitions of these specific features
the texture features. are found in Haralick et al.14
Roberts et al9 evaluated 13 of these features for their Because there is a danger of “overtraining” when
effectiveness in identifying bone loss in the context validating a model that is then validated on the same
of osteoporosis. In cortical bone, these features were dataset, a validation-training experiment was done.
derived from histograms calculated with pixel separa- Therefore, a model was developed using half of the
tions of 2, 3, 4, 5, and 6 in the direction perpendicu- data, and this model was subsequently tested on the
lar to the superior cortical border. In cancellous bone, remaining data. The training-validation experiment
pixel separations of 2 and 4 pixels were used in four di- was repeated and randomized. Because there were
rections: perpendicular and parallel to the direction of few failed implants in comparison to the number of
the cortical border and in the two diagonal directions successful implants, this model was only tested for
between these. The same features were evaluated here bone loss analysis.
as potential predictors of bone loss around implants
and implant failure. These features were derived, in the
case of cancellous bone, from the regions bilaterally RESULTS
between the mental foramen and the gonial angle,
and between the superior cortex and the mandibular Implant Loss
canal. This region provided an area with minimal over- Of the 460 patients assessed, 29 of them had man-
lay of other structures. The measurement area in the dibular implants that had failed (93.7% success rate)
cortex was defined by the superior and inferior bound- after 5 years. The mean mandibular cortical width of
aries of the inferior mandibular cortex, located by the the patients with successful implants was 3.04 mm
automated software12,13 between the mental foramen (standard deviation [SD] = 0.51) and that of those with
and antegonion. The width of the inferior mandibular failed implants was 3.21 mm (SD = 0.50). There was a
cortex was measured at a point near antegonion on small but nonsignificant difference between the mean
each side of the radiograph, as described in Allen et of the two groups (95% confidence interval [CI], –0.61
al.12 The cortical widths were the average value of the to –0.02).
measurements from both sides of the mandible. Table 1 lists those variables that predicted implant
loss over 5 years of the study. On a receiver operating
Statistical Analysis characteristic (ROC) curve built from this model, the
A linear model was built to classify outcomes of im- point of quasiequality of sensitivity and specificity was
plant failure and bone loss at 5 years after implant 62.1% sensitivity and 67.5% specificity. The area un-
placement. The model was constructed using the der the curve (AUC) from the model was 0.69 (95% CI,
complete set of features via stepwise variable selec- 0.60 to 0.78). Whether the patient smoked was not a
tion. Linear models were built using all data, and for significant predictor of implant loss in the model, but
some experiments, the data were split into training smoking did show an odds ratio of 2.24 for implant loss
and validation sets. In constructing the model, a num- versus those who did not smoke.
ber of features were found to be strongly correlated
with each other. In the trabecular bone, for example, Bone Loss Around Implants
it is unlikely that image textures will show strong di- Table 2 lists those variables in the general linear model
rectional properties; this leads to strong correlations from the trabecular and cortical analysis that signifi-
between the texture features measured in different di- cantly predicted bone loss around the implants after
rections. Likewise, some features may be similar at dif- 5 years. Whether the patient smoked also significantly
ferent pixel separations. When the correlating features predicted pathologic bone loss around the implants
© 2015 BY QUINTESSENCE PUBLISHING CO, INC. PRINTING OF THIS DOCUMENT IS RESTRICTED TO PERSONAL USE ONLY.
NO PART MAY BE REPRODUCED OR TRANSMITTED IN ANY FORM WITHOUT WRITTEN PERMISSION FROM THE PUBLISHER.
Merheb et al
Table 1 Variables in Trabecular and Cortical Bone Predicting Implant Failure After 5 Years
Variable Pixel separation Orientation* P value
Cancellous variance 4 0123 .0001
Cancellous angular second moment 4 0 .0001
Cancellous difference entropy 4 123 .0053
Cortical angular second moment 4 1 .0001
Cortical inverse difference moment 4 1 .0001
Cancellous sum entropy 4 0123 .0475
*0 = parallel to the upper border of the mandibular cortex; 1 = perpendicular to this border; 2,3 = diagonal directions between 0 and 1;
0123 = average of measures at all four orientations; 123 = average of orientations other than that tangential to the cortical border.
Table 2 Variables in Trabecular and Cortical Bone Predicting Pathologic Bone Loss
Around Implants After 5 Years
Variable Pixel separation Orientation* P value
Smoking status – – .0017
Cortical sum entropy 2 1 .0001
Cortical sum entropy 3 1 .0001
Cortical sum average 4 1 .0053
*0 = parallel to the upper border of the mandibular cortex; 1 = perpendicular to the upper border of the mandibular cortex.
(P = .0017). For this model, the point of quasiequality of The use of textural features based on co-occurrence
sensitivity and specificity on the ROC curve was 78.6% matrices for assessment of bone quality has been re-
sensitivity and 74.6% specificity (AUC = 0.88; 95% CI, ported previously. Kolacinski et al20 used the entropy
0.81 to 0.95). This model successfully predicted 20 of measure as a quantitative texture feature in intraoral
the 28 cases with pathologic peri-implant bone loss radiographs for assessing bone healing following sur-
(sensitivity = 71.4%) and 56 of the 63 cases with no gery. Dudek21 and Kozakiewicz et al22 used a different
pathologic bone loss (specificity = 88.9%). Mandibular co-occurrence matrix feature—sum of squares of vari-
cortical width was not a significant predictor variable ance—to assess damage to bone in response to dif-
of bone loss. ferent implant designs. Interestingly, the same feature
The validation-training experiment showed that the arose from the present analysis as one of the predictors
model developed from half the data and tested on the of bone loss around implants in panoramic radiographs.
remainder gave an AUC of 0.67 (95% CI, 0.51 to 0.83). In this study, the patients selected had panoramic
radiographs taken immediately before or after implant
placement. Although the aim of the study was to deter-
DISCUSSION mine the predictive value of panoramic radiographs, it
was hypothesized that cancellous and cortical bone
Implants tend to fail more frequently in areas of poor features would not vary significantly over such a short
bone quality because they depend on the surround- period of time. The algorithm was not optimized to
ing bone for support and retention.17 It can be difficult distinguish between failed and successful implants,
for the implant surgeon to determine noninvasively but numerous textural features of the mandibular can-
whether there is sufficient bone quality in the man- cellous bone were significantly predictive of implant
dible prior to implant placement and thereby offer a failure, indicating that cancellous bone plays a major
realistic treatment prognosis for the patient. Lettry et role in resisting the functional loading exerted by the
al18 found a weak correlation between the mechanical implant. The same conclusion was reached using fi-
properties of fresh mandibular bone and CT Hounsfield nite element analysis models of implants placed in the
numbers, but this was insufficient to accurately predict mandible.23 Experiments using high-resolution mag-
the mechanical properties in an individual given the netic resonance imaging have shown that the trabecu-
variability in this relationship. Some authors recom- lae are preferentially oriented in the long axis of the
mend cone beam CT to assess bone volume prior to tooth24 and therefore functionally adapted to resist oc-
implant placement, as these machines are optimized clusal loads. Cortical and cancellous bone quality influ-
for imaging of hard tissue,19 but the low contrast reso- ences the primary stability of implants and therefore
lution of the technique may limit the ability to distin- the success of treatment,25 and the textural features
guish between tissue types. of both were important in predicting implant success.
© 2015 BY QUINTESSENCE PUBLISHING CO, INC. PRINTING OF THIS DOCUMENT IS RESTRICTED TO PERSONAL USE ONLY.
NO PART MAY BE REPRODUCED OR TRANSMITTED IN ANY FORM WITHOUT WRITTEN PERMISSION FROM THE PUBLISHER.
Merheb et al
ACKNOWLEDGMENTS
Increment element (i, j) The authors reported no conflicts of interest related to this study.
Grey level j
Grey level i
tion meant that the CIs for the diagnostic ability (AUC Px+y (i)log Px+y (i)
{ }
i=2
in ROC analysis) were large. Difference entropy:
G 1
Px y (i)log Px y (i)
{ }
CONCLUSION where
i=0
G G
with features of the cortical bone. The cortical width the probability that the difference between co-occurrence matrix
did not have a significant relationship with implant coordinates is x-y.
© 2015 BY QUINTESSENCE PUBLISHING CO, INC. PRINTING OF THIS DOCUMENT IS RESTRICTED TO PERSONAL USE ONLY.
NO PART MAY BE REPRODUCED OR TRANSMITTED IN ANY FORM WITHOUT WRITTEN PERMISSION FROM THE PUBLISHER.
Merheb et al
REFERENCES 16. Petrou M, Garcia Sevila P. Image Processing, Dealing with Texture.
Chichester, UK: John Wiley and Sons, 2006.
17. Jemt T, Lekholm U. Implant treatment in edentulous maxillae:
1. Lambert FE, Weber HP, Susarla SM, Belser UC, Gallucci GO. Descrip- A 5-year follow-up report on patients with different degrees of jaw
tive analysis of implant and prosthodontic survival rates with fixed resorption. Int J Oral Maxillofac Implants 1995;10:303–311.
implant-supported rehabilitations in the edentulous maxilla. 18. Lettry S, Seedhom BB, Berry E, Cuppone M. Quality assessment of
J Periodontol 2009;80:1220–1230. the cortical bone of the human mandible. Bone 2003;32:35–44.
2. Jaffin RA, Berman CL. The excessive loss of Brånemark fixtures in 19. Fuster-Torres MA, Penarrocha-Diago M, Penarrocha-Oltra D. Rela-
type IV bone: A 5-year analysis. J Periodontol 1991;62:2–4. tionships between bone density values from cone beam computed
3. Merheb J, Van Assche N, Coucke W, Jacobs R, Naert I, Quirynen M. tomography, maximum insertion torque, and resonance frequency
Relationship between cortical bone thickness or computerized analysis at implant placement: A pilot study. Int J Oral Maxillofac
tomography-derived bone density values and implant stability. Implants 2011;26:1051–1056.
Clin Oral Implants Res 2010;21:612–617. 20. Kolacinski M, Kozakiewicz M, Materka A. Textural entropy as a
4. Adell R, Eriksson B, Lekholm U, Brånemark PI, Jemt T. Long-term potential feature for quantitative assessment of jaw bone healing
follow-up study of osseointegrated implants in the treatment of to- process. Arch Med Sci 2014. doi 10.5114/aoms.2013.33557 [Epub
tally edentulous jaws. Int J Oral Maxillofac Implants 1990;5:347–359. ahead of print]
5. Lekholm U, Zarb G. Patient selection and preparation. In: 21. Dudek D. The Evaluation of Dental Implant Loaded Remodelling in
Brånemark P-I, Zarb G, Albrektsson T (eds). Tissue-Integrated Patients with the Application of Digital Analysis of Radiological Im-
Prostheses: Osseointegration in Clinical Dentistry. Chicago: age Texture [thesis]. Lodz, Poland: Medical University of Lodz, 2012.
Quintessence, 1985. 22. Kozakiewicz M, Dudek D, Materka A. Influence of dental implant de-
6. Aranyarachkul P, Caruso J, Gantes B, et al. Bone density assessments sign on jaw bone structure. Presented at the 19th Congress of the
of dental implant sites: 2. Quantitative cone-beam computerized European Association for Cranio-Maxillo Facial-Surgery, Bologna,
tomography. Int J Oral Maxillofac Implants 2005;20:416-424. Italy, 2008.
7. Shahlaie M, Gantes B, Schulz E, Riggs M, Crigger M. Bone density 23. Matsunaga S, Shirakura Y, Ohashi T, et al. Biomechanical role of
assessments of dental implant sites: 1. Quantitative computed peri-implant cancellous bone architecture. Int J Prosthodont 2010;
tomography. Int J Oral Maxillofac Implants 2003;18:224–231. 23:333–338.
8. Devlin H, Karayianni K, Mitsea A, et al. Diagnosing osteoporosis by 24. Choel L, Last D, Duboeuf F, et al. Trabecular alveolar bone microar-
using dental panoramic radiographs: The Osteodent project. Oral chitecture in the human mandible using high resolution magnetic
Surg Oral Med Oral Pathol Oral Radiol Endod 2007;104:821–828. resonance imaging. Dentomaxillofac Radiol 2004;33:177–182.
9. Roberts M, Graham J, Devlin H. Image texture in dental panoramic 25. Bass SL, Triplett RG. The effects of preoperative resorption and
radiographs as a potential biomarker of osteoporosis. IEEE Trans jaw anatomy on implant success. A report of 303 cases. Clin Oral
Biomed Eng 2013;60:2384–2392. Implants Res 1991;2:193–198.
10. Albrektsson T, Zarb GA, Worthington P, Eriksson AR. The long-term 26. Devlin H, Allen PD, Graham J, et al. Automated osteoporosis
efficacy of currently used dental implants: A review and proposed risk assessment by dentists: A new pathway to diagnosis. Bone
criteria of success. Int J Oral Maxillofac Implants 1986 Summer;1(1):1 2007;40:835–842.
1–25. 27. Amorim MA, Takayama L, Jorgetti V, Pereira RM. Comparative
11. Horner K, Karayianni K, Mitsea A, et al. The mandibular cortex on ra- study of axial and femoral bone mineral density and parameters
diographs as a tool for osteoporosis risk assessment: The Osteodent of mandibular bone quality in patients receiving dental implants.
project. J Clin Densitom 2007;10:138–146. Osteoporos Int 2006;17:1494–1500.
12. Allen PD, Graham J, Farnell DJ, . Detecting reduced bone mineral 28. Friberg B, Ekestubbe A, Mellström D, Sennerby L. Brånemark im-
density from dental radiographs using statistical shape models. plants and osteoporosis: A clinical exploratory study. Clin Implant
IEEE Trans Inf Technol Biomed 2007;11:601–610. Dent Relat Res 2001;3:50–56.
13. Roberts M, Graham J, Devlin H. Improving the detection of osteo- 29. Bain CA, Moy PK. The association between the failure of dental
porosis from dental radiographs using active appearance models. implants and cigarette smoking. Int J Oral Maxillofac Implants
In: Niessen W, Meijering E (eds). The IEEE International Symposium 1993;8:609–615.
on Biomedical Imaging. Rotterdam, The Netherlands: IEEE, 2010: 30. Le BT, Follmar T, Borzabadi-Farahani A. Assessment of short dental
440–443. implants restored with single-unit nonsplinted restorations. Im-
14. Haralick RM, Shanmugam K, Dinstein I. Textural features for image plant Dent 2013;22:499–502.
classification. IEEE Trans Syst Man Cybern 1973;3:610–621.
15. Tucervan M, Jain A. Pattern Recognition and Computer Vision, ed 2.
Singapore: World Scientific Publishing Co, 1998.
© 2015 BY QUINTESSENCE PUBLISHING CO, INC. PRINTING OF THIS DOCUMENT IS RESTRICTED TO PERSONAL USE ONLY.
NO PART MAY BE REPRODUCED OR TRANSMITTED IN ANY FORM WITHOUT WRITTEN PERMISSION FROM THE PUBLISHER.