Efficacy of denoising and enhancement filters for detection of approximal and occlusal caries on digital intraoral radiographs.

BACKGROUND AND OBJECTIVES: Image processing and enhancement filters can significantly improve the diagnostic value of digital radiographs. Evidence shows that increasing the contrast and filtering improve the diagnostic accuracy for caries detection. This study sought to assess the diagnostic accuracy of original and enhanced digital radiographs for the detection of approximal and occlusal caries. SUBJECTS AND METHODS: In this experimental study, incipient carious lesions were artificially created on 120 proximal and occlusal surfaces of human extracted permanent molar and premolar teeth. After mounting the teeth in wax, digital radiographs were obtained using photostimulable phosphor plates and enhanced by enhancement filters 1, 2, and 3 with/without denoising. Two oral and maxillofacial radiologists blinded to filtering viewed the radiographs and filled out a chart. A definite diagnosis was made by a pathologist by observing the samples under a stereomicroscope (gold standard). Data were analyzed using SPSS software version 16. Interobserver agreement was calculated using kappa statistics. Chi-square test was used to assess the correlation between qualitative variables.
RESULTS: Assessment of sensitivity, specificity, accuracy, positive predictive value, and negative predictive value showed that enhancement filter 2 without denoising was the most efficient and original radiographs (filter free) were the least efficient radiographs for the detection of carious and sound surfaces. Application of filters significantly increased the accuracy of caries detection on digital radiographs. The lowest diagnostic accuracy was noted for the detection of enamel lesions on original radiographs (52%). Enhancement filter 2 plus denoising was the best filter for the detection of these lesions (79.25%). No significant difference was noted among different filters for detection of carious and sound surfaces but enhanced, and original radiographs were significantly different in visualization and detection of caries (P < 0.05).
CONCLUSION: Application of enhancement filters, particularly enhancement filter 2 with/without denoising, increases the accuracy of caries detection on digital radiographs.

INTRODUCTION

Dental caries is an infectious, multifactorial, and communicable disease, which occurs due to the complex interactive effect of cariogenic oral flora and fermentable carbohydrates on dental surfaces over time.[1]

Carious lesions have a progressive trend and often lead to pulp tissue inflammation requiring more complex, costly treatments. Thus, accurate and early detection of caries is clinically important.[2]

Since posterior teeth have larger proximal surfaces and mineral loss is insignificant in incipient caries, incipient approximal carious lesions in posterior teeth are often hard to detect.[3] Thus, radiographic examination is extremely important for the evaluation of proximal surfaces of posterior teeth.

Evidence shows that a considerable percentage of incipient occlusal caries remain undetected and detection of caries in occlusal pits and fissures is much more difficult than detection of other forms of carious lesions and requires high expertise and skills as well as advanced tools.[456] Morphology of occlusal grooves and occurrence of hidden caries at the depth of these grooves further highlight the significance of early detection of carious lesions.[7] Clinical examination, radiography, and transillumination are among the available caries detection modalities.[8]

Pathologies and abnormalities are not often easily detectable on intraoral digital images.[2] Digital radiographs usually have high noise, low contrast, and blurred edges.[9] The purpose of sharpening (enhancing) and smoothing (denoising) filter is to improve image quality and diagnostic accuracy by removing blur or noise. Noise represents random intensity variation.[210] Despite the advantages of digital system, the efficacy of some software enhancement features is still questionable because in some cases, software manipulation of digital radiographs may result in false-positive caries detection and lead to incorrect diagnosis and treatment.[1112] Thus, this study aimed to assess the efficacy of denoising and enhancement filters for approximal caries detection on digital radiographs.

SUBJECTS AND METHODS

This in vitro, experimental study was performed on 120 proximal surfaces of human extracted molar and premolar teeth. The samples were washed and disinfected with 5% sodium hypochlorite solution and were then stored in distilled water. The teeth were visually inspected and examined by a dental explorer to ensure the absence of caries or enamel cracks.

The exclusion criteria were the presence of white spot lesions or caries and presence of defects such as cracks or fracture.

A total of 120 teeth, which met the inclusion criteria, were selected and two layers of nail varnish were applied on their surfaces. Nail varnish was then removed from some of the randomly selected proximal and occlusal surfaces in the form of a square measuring 2 mm × 2 mm. To artificially induce caries, ten cate cariogenic solution with a pH of 4 was used. To induce caries-like lesions, the teeth were subjected to pH cycling such that the teeth were immersed in demineralizing agent with a pH of 4 for 18 h and were then placed in remineralizing agent with a pH of 7 for 6 h. The composition of demineralizing agent was as follows:

0.05 mM CaCl2

2.2 mM NaH2 PO4

50 mM acetic acid.

The composition of remineralizing solution was as follows:

20 mM HEPES

1.5 mM Ca2+ as CaCl2

0.9 mM phosphate as KH2 PO4

1 ppm fluoride as NaF.[1112]

Periodically, after 30 days, the teeth were radiographed using MinRay intraoral radiography unit (Soredex, Tuusula, Finland) with exposure settings of 60 kVp, 7 mA, and 0.1 s. Presence of incipient caries was confirmed both visually and radiographically.

The process of pH cycling to induce incipient caries was adjusted such that incipient carious lesions were limited to the enamel thickness and were visible on radiographs. During this period, pH of the solution was measured periodontally and the solution was replaced if any change was detected. After formation and radiographic confirmation of caries, the teeth were coded and randomly mounted in wax (three/piece) such that carious proximal surfaces were positioned at the contact areas between teeth. To obtain bitewing radiographs, photostimulable phosphor (PSP) plates (Optime; Soredex, Tuusula, Finland) were used. Size 2 PSP plate was placed in appropriate position and radiographs were obtained using MinRay radiographic unit with exposure settings of 60 kVp, 7 mA, and 0.1 s. Original radiographic images (without filtering) were saved in a computer. Using Scanora software, denoising and enhancement filters 1, 2, and 3 were applied until the histogram curve of each image remained unchanged and further filtering had no effect on image. Denoising filters smooth an image by removing high-frequency noise (small-scale intensity variations). Enhancing filters sharpen an image by removing low-frequency noise (gradual or large-scale intensity variations) or enhance boundaries between regions with different intensities (edge enhancement filters).[2] By increasing the degree of enhancing filters, the effect of them, increase. It should be noted that, after applying each filter, the image was saved and then the image was converted to its original form and the next filter was applied. Thus, we had one original (without filtering) and six filtered versions of each image for evaluation Figure 1.

Figure 1

Seven versions of radiographs evaluated in this study (original and with different combinations of filters)

First version – No denoising filter, enhancement filter 1

Second version – No denoising filter, enhancement filter 2

Third version – No denoising filter, enhancement filter 3

Fourth version – Application of denoising filter plus enhancement filter 1

Fifth version – Application of denoising filter plus enhancement filter 2

Sixth version – Application of denoising filter plus enhancement filter 3.

Two oral and maxillofacial radiologists, blinded to the type of filters applied, viewed the images, and filled out a checklist for each image. Observers were also blinded to the location of carious lesions and randomly observed the radiographs. Each observer viewed the images twice in an adequately lit room with no light reflection on a liquid crystal display monitor (L40 Satellite 15, Toshiba) with 768 × 1367 resolution. Observers scored each image using the following four-point scale:

0: No caries

1: Enamel caries

2: Caries at the dentinoenamel junction (DEJ) and outer half of dentin

3: Caries in inner half of dentin.

The teeth were sectioned by a diamond disc parallel to the longitudinal axis of the tooth in mesiodistal direction along the central groove [Figure 2] and evaluated under a stereomicroscope (Olympus, Hamburg, Germany) at ×20 to determine definite presence or absence of caries. The results of stereomicroscopic examination served as the gold standard [Figures 3 and 4]. It should be noted that four samples were damaged during sectioning for stereomicroscopic assessment and were excluded from the study.

Figure 2

Sectioning of teeth using CNC machine

Figure 3

Microscopic view of noncarious teeth

Figure 4

Microscopic view of carious teeth

Data of each checklist were entered into SPSS software version 16 (SPSS Inc., Chicago, IL, USA). Interobserver agreement was assessed using kappa statistics. The correlation between qualitative variables was assessed using Chi-square test. Level of significance was set at P = 0.05.

RESULTS

A total of 120 teeth were evaluated by each of the two observers including 60 molar and 60 premolar teeth; 30 occlusal and 30 proximal surfaces of molar and premolar teeth were evaluated. In total, 60 occlusal and 60 proximal surfaces were evaluated. The status of 60 occlusal and 60 proximal surfaces in terms of presence/absence of caries is shown in Table 1.

Table 1

Status of occlusal and proximal surfaces in terms of presence/absence of caries

Interobserver agreement was evaluated for original and filtered radiographs. The highest agreement was noted in the use of enhancement filter 3 without denoising (0.889) and the lowest agreement was noted in the absence of filter (0.670).

Diagnostic accuracy for occlusal and proximal caries was assessed in the presence of filters and absence of them, and the least accuracy was noted for radiographs without filters for both occlusal (0.551) and proximal (0.650) caries.

It should be mentioned that since the diagnostic accuracy for occlusal and proximal caries was close and no statistically significant difference was found between them (P > 0.05), we considered occlusal and proximal surfaces as one entity (altogether) in the next statistical analysis.

Table 2 summarizes that enhancement filters 1 and 2 without denoising had higher sensitivity but lower specificity for caries detection compared to enhancement filters 1 and 2 with denoising; this difference was greater for enamel caries. Furthermore, diagnostic accuracy of enhancement filter 1 without denoising was higher than that of enhancement filter 1 with denoising.

Table 2

Diagnostic sensitivity, specificity, positive predictive value, and negative predictive value of digital radiographs with different filters for different depths of carious lesions

Diagnostic accuracy of enhancement filter 2 with denoising was higher for enamel caries and those at the DEJ and outer half of dentin. However, enhancement filter 2 without denoising had higher accuracy than enhancement filter 2 with denoising for caries in inner half of dentin.

Enhancement filter 3 with and without denoising had similar sensitivity for detection of caries, but enhancement filter 3 with denoising had superior specificity compared to enhancement filter 3 without denoising particularly for enamel caries.

Enhancement filter 3 with denoising had higher accuracy for detection of enamel caries compared to enhancement filter 3 without denoising.

As seen in Table 2, evaluation of sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) of original (nonfiltered) images showed that, in general, sensitivity, specificity, and accuracy of caries detection improved with an increase in depth of carious lesions.

As seen in Table 3, evaluation of diagnostic sensitivity, specificity, accuracy, PPV, and NPV for carious and noncarious surfaces revealed that enhancement filter 3 without denoising had the highest and original (nonfiltered) radiographs had the lowest sensitivity. Enhancement filter 1 with denoising had the highest and enhancement filter 3 with denoising had the lowest specificity. Enhancement filter 2 without denoising had the highest and original radiographs had the lowest accuracy.

Table 3

Sensitivity, specificity, accuracy, positive predictive value, and negative predictive value of digital radiographs with different filters for detection of caries (irrespective of depth of carious lesions)

Table 4 summarizes that no significant difference existed between filters in the detection of caries, but a significant difference was noted in caries detection between original and filtered radiographs.

Table 4

Difference between filters in caries detection using Chi-square test

DISCUSSION

Selection of the best diagnostic modality for detection of caries has always been a highly debated topic among researchers.[13] Bitewing radiography is commonly requested for detection of caries. Digital radiography has lower patient radiation dose because digital sensors have higher sensitivity to radiation exposure compared to analog films.[14] Evidence shows that digital radiography has no superior diagnostic value to analog radiography for detection of superficial carious lesions.[1415] Thus, enhancement of digital radiographs was attempted to improve caries detection.[1516] Choi et al. confirmed that enhancement of digital radiographs affects the quality of images depending on the diagnostic goal.[17] Although some studies showed that enhancement of digital radiographs may improve caries detection accuracy,[13] most studies have shown poor detection of incipient enamel caries with both analog and digital radiography.[18]

In our study, different enhancement filters and denoising were applied to enhance the detection of caries in occlusal and proximal surfaces. Based on the current results, diagnostic accuracy of filtered images was significantly higher than that of original images. Kajan et al. showed that image sharpening with 1:3 magnification improved diagnostic accuracy for detection of noncavitated proximal caries.[19]

Assessment of diagnostic accuracy of filtered and original images for detection of occlusal and proximal caries revealed that original images had the lowest diagnostic accuracy for occlusal and proximal caries. We concluded that, by an increase in the degree of enhancement, sensitivity increased but specificity decreased. It should be noted that, in caries detection, high specificity of a diagnostic test/tool is more important than high sensitivity to prevent unnecessary treatment (considering the slow trend of caries progression, false-negative diagnosis is preferred to false-positive diagnosis).

In the study by Belém et al., similar to our study, the highest sensitivity and accuracy were noted when “sharpen” filter was used.[13] Thus, our study confirmed the findings of Belém et al. We compared the presence and absence of denoising filter and concluded that application of denoising filter decreased sensitivity and diagnostic accuracy and increased specificity. A possible explanation for this finding is that application of denoising filter decreases the sharpness of images to some extent, which decreases the efficacy of enhancement filter and thus decreases sensitivity and increases specificity. Kositbowornchai et al.[20] reported that application of enhancement filter had no efficacy for detection of occlusal caries, which was in contrast to our results and those of Belém et al.[13]

In our study, sensitivity of filtered digital radiographs for detection of caries was 70%–83%, which is higher than the value reported in other studies.[13] A possible explanation for this difference is that naturally developed carious lesions are irregular and have relatively low contrast and are not easily detectable. In this in vitro study, the teeth were sound, and only demineralizing solution was used to induce caries. As a result, induced carious lesions were more regular than those naturally occurring in the oral environment; such regular borders enhance caries detection on radiographs.

Belém et al. reported that use of “sharpen” filter resulted in high percentage of accurate detection of the presence (83%) and absence (79%) of enamel subsurface demineralization.[21] Abesi et al. and Castro et al. reported that diagnostic accuracy increases with an increase in depth of carious lesions. Our findings confirmed those of Abesi et al. and Castro et al., in this respect.[1822]

Enhancement filter 3 without denoising had the highest sensitivity (83.75%) and original images had the lowest sensitivity (55.82%) in our study. Furthermore, no significant difference was noted in caries detection between different filters, but original and filtered radiographs were significantly different in terms of caries detection. Berkhout et al.[23] showed that different spatial resolution did not improve the diagnostic accuracy of caries detection. This controversy in the results of the two studies may be due to different types of filters used in the two studies.

Our study results were in agreement with those of Castro et al., in 2007 in that they also concluded that the use of different types of filters would yield comparable results.[22]

Shi and Li[24] in 2009 used color-coded and black and white digital radiographs for detection of approximal caries and found no significant difference between the two, which may be attributed to different types of filters used in their study compared to ours.

Haak and Wicht[25] in 2005 concluded that the use of reverse contrast filter yielded higher diagnostic validity. Thus, it appears that caries detection can be enhanced, especially for enamel lesions using the software and filters applied in our study. Dove et al.[16] in 1992 compared original digital and analog radiographs and concluded that the two modalities were not significantly different in detection of approximal caries. However, in our study, we concluded that, by changing the enhancement level, higher diagnostic accuracy may be achieved.

In our study, the lowest diagnostic accuracy for enamel caries belonged to original images (58.06%), and enhancement filter 2 with denoising was the best filter for detection of these lesions (79.25%).

In the current study, difference in accuracy of caries detection between filtered and original images was mainly seen in superficial carious lesions and by an increase in depth of carious lesions, the difference in diagnostic accuracy decreased. Therefore, the efficacy of application of enhancement filters is greater for superficial caries. Such an increase in diagnostic accuracy of digital images by applying enhancement filters is promising considering the significance of early detection of superficial carious lesions since in case of early detection; these lesions can be treated by noninvasive and nonrestorative procedures. Therefore, further studies are required to better elucidate this topic.

CONCLUSION

Use of enhancement filters can improve the quality of digital radiographs and increase their diagnostic value for detection of caries. Among filters, enhancement filter 2 without denoising had the highest accuracy for caries detection. Although different filters increased caries detection accuracy, detection of enamel lesions is still challenging.

Financial support and sponsorship

Nil.

Conflicts of interest

There are no conflicts of interest.