Radiographic modalities for diagnosis of caries in a historical perspective: from film to machine-intelligence supported systems

Abstract

Radiographic imaging for the diagnosis of caries lesions has been a supplement to clinical examination for approximately a century. Various methods, and particularly X-ray receptors, have been developed over the years, and computer systems have focused on aiding the dentist in the detection of lesions and in estimating lesion depth. The present historical review has sampled accuracy ex vivo studies and clinical studies on radiographic caries diagnosis that have compared two or more receptors for capturing the image. The epochs of film radiography, xeroradiography, digital intraoral radiography, panoramic radiography and other extraoral methods, TACT analysis, cone-beam CT and artificial intelligence systems aiding in decision-making are reviewed. The author of this review (43 years in academia) has been involved in caries research and contributed to the literature in all the mentioned epochs.

Radiographic diagnosis of caries lesions

Radiography is a widely recommended adjunct method in the diagnosis of caries lesions in clinically inaccessible surfaces. An additional diagnostic value of radiography compared with merely a visual examination has been reported in several studies over the years for the detection of carious demineralisation in approximate surfaces both in adults and children.^1–3 Even though variation among observers’ radiographic interpretation of lesions is well known, radiography is still the most feasible para-clinical method for daily clinical practice.

The lesion observed in a radiographic image is not the caries disease, but a reflection of present or past microbial activity.⁴ The caries lesion is displayed as a demineralised area starting at the tooth surface. The classical shape of approximal surface demineralisation, that is, the radiolucent area, in the enamel is a triangle with its broad base at the tooth surface, but other appearances are known, such as a dot, a notch, a band or lines. Lesions involving approximal surfaces usually occur in the sub-contact area. Lesions involving occlusal surfaces are often seen only after reaching the dentine as a bowl-shaped demineralisation beneath the enamel.⁴ Lesions in other tooth surfaces can usually be diagnosed clinically. Tooth morphology, for example pits and fissures, optical illusions such as the Mach-band effect and phenomena such as the cervical radiolucency (burn-out effect) may be falsely interpreted as caries lesions.⁴ Also other disease entities in the dental hard tissues, for example abrasion, hypoplastic pits and pre-eruptive intracoronal resorption may mimic caries lesions. This should be considered when reporting on caries lesions in order to avoid excessive false-positive observations.

A single radiograph cannot per se distinguish between an active and an inactive caries lesion. An old inactive lesion will still appear as a demineralised area in the hard tissues, a “scar”, since remineralisation takes place only in the outermost surface of the lesion.⁵ Because the image only mirrors the extent of demineralisation, one radiograph alone cannot determine whether activity is ongoing. Moreover, the radiographic appearance of the demineralisation does not allow a distinction between a non-cavitated and a cavitated lesion. Attempts have been made to analyse the grey shade distribution inside the lesion in digital images to distinguish between lesions with and without surface cavitation. Grey shade values were significantly lower (i.e. areas appearing darker in the image) in carious dentine, but no differences could be observed between grey values in cavitated and non-cavitated dentinal lesions.⁶ The outer state of the surface can, therefore only be determined clinically.^7,8

While 40 years ago, any carious demineralisation in the tooth surface observed in a radiograph might be treated with a filling,^9–11 treatment strategies have changed, and the new paradigm encourages the attempt to arrest lesions suspected to have an intact surface, thus sparing operative treatment for cavitated lesions, inaccessible to cleaning.¹² This treatment strategy leaves a challenge for approximal surfaces in teeth with tightly contacting neighbouring tooth surfaces, particularly in the molar regions, where the clinical examination may be insufficient. Thus, in clinically inaccessible approximal surfaces, and also in deep occlusal fissures, the radiographic display of the lesion will often be the method to determine the treatment choice. The risk for cavitation in approximal surfaces is related to lesion depth, and based on studies (although some rather old) of the relationship between radiographic lesion depth and clinical cavitation,^12,13 a threshold for operative treatment decision has been suggested when a lesion is observed radiographically more than one-third into the dentine, or when a previously observed shallow lesion has progressed further into dentine.

Determining the diagnostic accuracy efficacy of an imaging method does not reach the highest level in the hierarchical evidence ladder, but ex vivo studies need to be performed in order to have a solid reference standard for the true state of a lesion.¹⁴ Studies that provide a useful reference method for radiographic caries detection in the clinic are few. Obviously, studies comparing two radiographic methods for lesion detection or lesion behaviour in patients cannot be easily conducted due to patient dose issues. Fortunately, it seems that results from ex vivo experiments may be relied upon, also in the clinical situation. Two studies have assessed the same teeth in vivo and ex vivo: Pre- and post-extraction film radiographs were recorded of third molars with subsequent histological validation of the caries lesions. The accuracy was almost identical in the two situations for both occlusal and approximal lesions.¹⁵ Patients, who had parts of the lower jaw excised owing to a cyst or neoplasm, had radiographs of the involved teeth before and after operation. There were no significant differences in the detection of approximal caries lesions in digital images taken with the same receptor in the two situations.¹⁶

The bitewing (BW) technique, suggested by Raper (1925),¹⁷ has been in use for almost a century with the aim of visualizing interproximal caries, and has continued since with only minor alterations. The vast majority of studies conducted on caries detection with intraoral receptors have used the BW projection or a simulation hereof ex vivo. An appropriate design for ex vivo studies is that extracted, cavitated and/or non-cavitated, non-restored human teeth are positioned in a row simulating the posterior parts of the jaws before exposure. Most ex vivo studies define histological sectioning of the teeth and microscopic analysis of the sections as the reference standard for the true presence/absence of a caries lesion and for estimating lesion depth. This review includes only studies with a solid reference standard,^18–21 and only studies with natural caries lesions in human teeth.

The aim of the review is to provide a historical overview of radiographic modalities that have been investigated over the years, of which some are still in use, for the diagnosis of carious lesions with or without surface cavitation. Examples of first-movers within each epoch are given. A historical overview of X-ray sources and the developments in radiographic technology in dentistry has just been published as an anniversary article in this journal,²² and radiographic equipment will not be discussed in this review.

The review will include studies that have compared various radiographic receptors/modalities used in dentistry for assessing accuracy or validity of detecting caries lesions and estimating lesion depth. The review does not include studies comparing clinical and radiographic methods for caries diagnosis. It will not discuss when a radiographic examination of clinically suspected caries lesions is indicated, or which viewing media and conditions are optimal with the various modalities; nor will it discuss in detail radiation dose or costs in relation to the radiographic modalities.

Intraoral film radiography

It is a well-known fact that the first dental radiographs were available in 1896 on Eastman roll film wrapped in black paper. Eastman Kodak Co. also produced the first packaged dental film with a lead foil for direct exposure in 1919; thereafter, film was the media for capturing the image, viewing the image and archiving the information. The film was developed over the years, consecutively named by capital letters A-, B-, C-, D- (Ultra-speed, introduced 1955), E- (Ektaspeed, introduced 1981, Ektaspeed Plus, introduced 1994), and F-speed (Insight, introduced 2000) after its radiation sensitivity, which means that the dose needed for an optimal exposure was reduced to 1/60 over a hundred years.²³ D- to F-speed films are still commercially available, and over these speed categories, radiation is reduced approximately 60%.²⁴ All types of film need light proof wet chemical processing. Since speed mostly depends on the size and organization of the silver halide crystals influencing spatial resolution, there has been a need to compare the various film speeds for diagnostic detail and validity. Moreover, with the aim to further reduce dose, studies were conducted in the 1980s with screen/film systems in a cassette.

A number of ex vivo studies have thus compared various film speeds assessing their accuracy for detection of caries lesions. D-speed film was found to be more accurate for detection of approximal lesions than the first E-speed film (the Ektaspeed film had a more grainy appearance than the subsequent Ektaspeed Plus) and three screen/film systems.²⁵ Most studies, however, found no differences between various D-speed (Ultraspeed and DV 57) and subsequent E-speed films (Ektaspeed Plus, Agfa M2 Comfort, Flow, and EV 57) for detecting initial, approximal, occlusal or recurrent lesions.^26–33 Also, no differences were observed between these lower film speeds and the more sensitive F-speed film.^33,34 F-speed film was also accurate in estimating the depth of the lesion compared to histology in one study,³⁵ while in another, radiographic diagnosis in F-speed film underestimated the true clinical extent of the lesion.³⁶ In one study, Ektaspeed Plus film obtained significantly higher accuracy than both D- and F-speed films.³⁷

In conclusion, the development of faster film speeds over the years seems not to have significantly influenced the diagnostic accuracy for detecting caries lesions in film radiographs, when processing and viewing are optimal. Screen/film systems with a BW cassette have not entered the clinic for routine use; reasonably because of cumbersome positioning and unpleasantness for the patient. With new receptor types entering the market, research on caries detection comparing different film speeds and types seemed to cease approximately 15 years ago.

Intraoral xeroradiography

Xeroradiography was known to the medical field, before it became available to dentistry. This method uses a xeroradiographic copying process to record images produced by X-rays. It differs from the silver halide film technique, not involving wet chemical processing (xero = dry), nor the use of a dark room/box.³⁸ A new xerographic method was designed for dental use in the late 1970s.³⁹ The surface of a small selenium photoreceptor is electrically charged. After insertion into a light-tight cassette, the photoreceptor is positioned intraorally and exposed like film. The resultant electrostatic charge image is developed in a processor using liquid toner. The toner image is transferred from the photoreceptor and fixed to a white plastic substrate for viewing. After cleaning, the receptor can be reused. Processing time was approximately 20 s.⁴⁰

Most of the research conducted on xeroradiography in the late 1980s originated in University of California, where accuracy for detection of caries lesions was determined for xeroradiographs and film in ex vivo studies. One study found similar accuracies for E-speed film and xeroradiographs, while D-speed film obtained somewhat lower accuracy.⁴¹ Other studies showed that xeroradiographs and D-speed film were superior to E-speed film for detecting approximal caries lesions,⁴² and xeroradiographs showed higher sensitivities than film, but at the same time with higher fractions of false-positive scores.⁴³

In the clinical environment, xeroradiographs showed high anatomic detail of teeth and bone in comparison with film radiographs.⁴⁴ Both high- and low-contrast xeroradiographs were found to be useful in the clinic for the detection of approximal surface caries by experienced as well as inexperienced dentists.⁴⁵ In a clinical setting, xeroradiography and D- and E-speed film were equally useful,⁴⁶ while xeroradiography seemed to be better than film for detection of recurrent caries.⁴⁷

In conclusion, the xeroradiographic method has not been evaluated much since the 1980s, and its clinical use seems to have stagnated. This may be ascribed to the developments of intraoral digital receptors in the late 1980s, and the capacity of the personal computer. Moreover, the xeroradiographic cassette was more difficult to position correctly for full-mouth intraoral radiography than film packets, increasing patient discomfort.⁴⁸

Digital intraoral radiography

In 1987, the first European meeting of dental and maxillofacial radiology was organised in Geneva (now the European Association for Dento-Maxillo-Facial Radiology, EADMFR). A French dentist and inventor here demonstrated the first direct digital intraoral radiography system for dentistry, which became known as RadioVisioGraphy (RVG).⁴⁹ The first RVG system consisted of an exchangeable scintillation screen, fibre optics and a miniature charge-coupled device (CCD) as receptor. Monitor resolution in personal computers was still low in the 1980s, and only a limited number of grey shades was supported at a time. The RVG unit, therefore, utilised a black-and-white, TV-quality stand-alone monitor for image display and video prints for archiving the images.

The first diagnostic ex vivo studies evaluated the performance of this new CCD receptor compared with conventional film. Accuracy of D- and E-speed film, digitised film radiographs, and RVG print images was not significantly different for the detection of dentinal caries in occlusal surfaces.⁵⁰ In another study of occlusal caries, sensitivity and specificity were similar to those of film, while for approximal caries, sensitivity of RVG video prints was slightly lower than film.⁵¹

As soon as S-VGA graphic cards were available, the digital image could be displayed on the computer monitor showing 64 of its 256 shades of grey at a time. Thus in the beginning of the 1990s, many new CCD sensors, and later CMOS (complementary metal oxide semiconductor technology) sensors were brought to the market.^52–54 In 1994, the first photostimulable storage phosphor system (PSP), Digora, was launched for dental imaging based on technology known from medical radiography.⁵⁵ The two types of intraoral digital receptors are now named solid-state sensors, or direct digital systems, and storage phosphor plate systems, or indirect digital systems. Solid-state sensors are available with or without a cord that connects the receptor to the computer (in a cordless sensor, the signal is transferred by radio wave to a receiver), and the image appears at the computer monitor in seconds. In PSP systems, the image must be read out in a scanner, where the stored energy in the emulsion is released by stimulation with visible light. Hereafter, the plate can be reused.

Numerous ex vivo studies have been conducted to compare accuracy of these new receptors with conventional film and with each other for caries diagnosis. In a further development of the CCD RVG receptor, RVGui, the performance in high- and low-resolution modes for approximal caries detection was comparable to that of Ektaspeed Plus film.⁵⁶ D- and E-speed film and CCD receptors, RVG and Visualix, were compared for the detection of enamel approximal caries lesions; and D- and E-speed films and the CCD Visualix receptor for the detection of dentinal occlusal lesions, all with no statistically significant differences among the methods.⁵⁷ The same results were obtained for various other CCD and CMOS sensors compared with D-, E- or F-speed film in detecting approximal caries.^58–68 For detection of occlusal lesions in primary molars, the CCD Dixi3 receptor and insight film likewise obtained similar diagnostic accuracy.⁶⁹

Subsequently, comparisons were done between solid-state sensor and PSP systems. The diagnostic accuracy was identical with four digital systems (Digora PSP, and CCD systems RVG, Sens-A-Ray, and Visualix) and in compressed (JPEG, irreversible compression) and uncompressed images.⁷⁰ Similarly, in two PSP systems (Digora and DenOptix) and two CCD-based sensor systems (Dixi Planmeca and Sidexis), differences in accuracy for approximal caries detection were non-significant, and no effect was found of different software modalities used to assess the images.⁷¹ Insight film and digital systems (Digora, DenOptix, and CygnusRay MPS) showed no significant differences for detection of approximal caries lesions.⁷² Digora Optime PSP system, RVGui CCD, and Ektaspeed film were equally accurate for approximal caries detection in primary molars at 65- and 70-kV tube potentials.⁷³ A CCD sensor (Dixi3), PSP Digora PCT, and E-speed film were similar in the detection of non-cavitated approximal caries.⁷⁴ A CMOS sensor system was comparable to PSP systems and film for approximal caries detection.⁷⁵ Approximal caries lesion depth was accurate, when estimated in images from a PSP system (Vista Scan) compared with stereomicroscopy of the tooth sections.⁷⁶ A CCD-based system (Dixi) and a PSP system (Digora) were more accurate than other CCD- (Sidexis) and PSP- (DenOptix) systems for measurement of caries lesion depth.⁷⁷ With the aim to detect cavitated lesions in primary molars, there was no significant difference between the accuracy of E-speed film and the Digora PSP system.⁷⁸ Also for detection of cavitated lesions, the Sirona, Kodak, and Schick sensors performed equally accurately.⁷⁹ CCD- and PSP receptors and films exposed with 60 and 70 kVp were not significantly different for approximal caries detection.⁸⁰

A number of studies have, on the other hand, observed differences in accuracy among digital systems and between film and digital images. Ektaspeed Plus film was superior to the CCD Sens-A-Ray receptor for detection of approximal caries.⁸¹ Among six CCD-based receptors and two film types, the CCD Dixi sensor and Ektaspeed Plus film were significantly more accurate than the other systems for approximal caries detection.⁸² Also, in one study, a CCD-based Schick sensor was not as accurate as conventional film for diagnosing approximal surface caries in the mixed dentition.⁸³ Two E-speed films (Ektaspeed Plus and Dentus M2 Comfort), two CCD systems (Sidexis Sirona and Visualix), and two PSP systems (Digora and DenOptix) were compared in a study on approximal caries detection, which found lower accuracies for the Visualix and DenOptix system.⁸⁴ DenOptix, Digora (blue and white plates), and Ektaspeed Plus film all had higher accuracies than another PSP system, Cd-dent, for approximal caries detection.⁸⁵ In various scanning resolutions, overall accuracy was lower for PSP Digora “high-resolution mode” than for a CCD-based sensor (Dixi2) and a PSP system (VistaScan) in max-resolution modalities, while there were no differences among the other modalities.⁸⁶ Caries lesions were underestimated in PSP VistaScan images compared with two sensors (Planmeca Dixi 2, Gendex Visualix HDI) and PSP Digora Optime, and the CCD systems had a significant higher rate of correct dentine diagnoses and lower rate of false diagnoses.⁸⁷ The Gendex PSP system seemed to provide higher specificities than E- and F-speed films.⁸⁸ PSP systems (DenOptix and Digora FMX with white and blue plates) and Insight film were equally accurate for detection of approximal enamel caries, except that the Denoptix had somewhat lower specificity than Digora white plates.⁸⁹ In a comprehensive study on nine receptors: three PSP (Digora fmx and Digora Optime and VistaScan), five sensor systems (Dixi2, Schick CDR-APS and CDR wireless, Sidexis, and Dr Suni Plus), and one Insight film, the false-positive rate for detection of approximal lesions was significantly lower for Digora fmx PSP and Dixi2 sensor.⁹⁰

In conclusion, digital intraoral radiography systems are now widespread in general dental practitioners’ offices, and in European countries, more dentists use digital radiography systems than conventional film in the daily routine. Based on ex vivo findings on accuracy, most commercially available receptors for digital radiography have obtained diagnostic accuracies in line with film. Some digital receptors although have consistently shown lower validity than others, and research results may have aided in regulating the market. In the above paragraph, only trade names of the receptors are given since companies have merged (some are obsolete), and names have consistently been altered.

Other factors than accuracy results may, on the other hand, determine the choice of receptor in the clinical situation. There may be challenges when solid-state sensors are used for bitewing examination.⁹¹ First, the radiation-sensitive surface area is smaller than the area of a size 2 film, displaying fewer interproximal tooth surfaces per exposure than with film. Second, the stiffness and increased thickness of these sensors are more uncomfortable for the patient⁹² and may result in more projection errors and more exposures and retakes.^91,93 Recently developed sensors with more ergonomic shapes may overcome most positioning problems and patient discomfort, though.⁹⁴ The PSP plate is operated almost like film, although other errors may occur than with film radio-graphy.⁹⁵ Many advantages in the work flow with digital systems compared to film have been thoroughly described in previous reviews.^96,97

Enhancement algorithms

The introduction of Microsoft Windows^™ loosened the tight relationship between monitor display and software. This broadened the utility of programs for enhancement of the image in real-time, leading to the contrast-, brightness-, inversion-, gamma-curve functions, etc., which are now part of imaging software with the digital receptor. Task-specific enhancement algorithms were also developed, and a number of studies have focused on the effect of these algorithms compared to the original digital image for diagnosis of caries lesions.

In studies on digitized film, a filter applied to the image seemed to improve the overall validity of caries depth measurements,⁹⁸ while others found no effect on dia-gnostic accuracy of histogram-equalization, grey-scale inversion or various grey scale modifications in digitized film.^99–101 For occlusal caries detection, contrast stretch and filtering performed overall better than the original digitized film,^102,103 while another study found no differences between contrast-enhanced CCD (RVG) images or digitized film compared with the original images.⁵⁰ Similarly, enhancement with sharpness, zoom or pseudocolour did not improve accuracy.¹⁰⁴ Contrast enhancement in CCD-based (Dexis) images did not improve accuracy of restorative treatment decision compared with D-speed film and a PSP (Digora) system.¹⁰⁵ Enhancement of PSP images improved detection of approximal caries compared with unenhanced images and E-speed film,¹⁰⁶ while a caries-specific procedure only improved detection of caries lesions in the outer half of the enamel, but not for other caries depths.¹⁰⁷ For inverted grey scale images, there were no accuracy differences compared with original CMOS (Schick) sensor and film images,¹⁰⁸ while in another study, sensitivity for original images in detecting approximal dentinal caries was significantly higher with no significant differences for specificity.¹⁰⁹ A sharpening filter along with magnification seemed to improve accuracy in PSP images,¹¹⁰ while others found that filters in PSP had no effect for the detection of caries lesions or cavitation.¹¹¹ Sensor-based images were not different from Ektaspeed Plus film in detecting approximal caries, while observer-enhanced images exhibited a significantly lower accuracy.^59,63 Diagnostic accuracy was improved after correction for X-ray attenuation.¹¹² An image-enhancement filter had a small effect on the validity of measurements of caries lesion depth.¹¹³ The presence of high-density material and adjustment of brightness and contrast did not significantly affect the diagnosis of approximal caries.¹¹⁴

A number of studies have evaluated “task-specific” enhancement algorithms, that is filters specifically developed to enhance the carious demineralisation. A predefined enhancement filter (“dento-enamel”) in RVG sensor images obtained higher sensitivities, but with more false-positive rates than the original images from RVG and images from other comparable sensors.¹¹⁵ In the VistaScan PSP system, enhancement filters (“Caries1” and “Caries2”) were less accurate than the original images for approximal caries detection.¹¹⁶ Two proprietary caries-specific enhancement algorithms (“K1” and “K2”) had similar accuracies compared with standard PSP images.¹¹⁷ Some authors developed their own enhancement filters: The accuracy of approximal caries detection with pre-enhanced Digora PSP images (“caries-specific Oslo enhancement procedure”) and images individually enhanced by the system’s modalities was similar, though.¹¹⁸

A few clinical studies have compared film and a PSP plate by positioning the film behind the plate during exposure in patient examinations. Changing contrast and brightness in DenOptix PSP images did not alter the dia-gnosis compared with Insight film,¹¹⁹ while caries depth in PSP images seemed to be underestimated compared with film.¹²⁰

In conclusion, enhancement algorithms are part of the software with digital imaging systems. It seems, however, that optimally exposed images gain very little by specific enhancement procedures for accurate detection of caries lesions. Images recorded with suboptimal contrast and brightness may be enhanced to some extent by these features, though. Clinicians seem to be comfortable in using the system’s enhancement features when viewing and interpreting the image. Observers, who were monitored without their knowledge during interpretation of caries lesions, used one or more enhancement features for almost all images recorded with a PSP and three CCD systems.¹²¹

Subtraction radiography

In the beginning of the 1980s, ingenious researchers were first movers in digitizing two film radiographs and subtracting the numeric information from one another to assess whether more information could be extracted from the subtraction image than from comparing the two radiographs in the traditional way.^122,123 The pioneer work included mathematical corrections of distortions in the images to be subtracted.¹²⁴ When two radiographs are recorded with controlled projection angles and thereafter subtracted, theoretically all unchanged background anatomical structures are cancelled, and these areas will be displayed in a neutral grey shade in the subtraction image, while regions that have changed between the radiographic examinations are displayed in darker or lighter shades of grey. Unavoidable projection distortion may be compensated for by mathematical algorithms. Positioning of reference points in the two images to be subtracted provided superior subtraction images than the traditional manual superimpositioning of the images.¹²⁵

A number of ex vivo studies were conducted in the 1990s and onwards comparing the subtraction technique with conventional radiography for caries lesion detection. Adding barium sulphate to the approximal surface before the second exposure, extracted teeth with natural caries lesions were recorded and the images thereafter subtracted. Assessment of lesion depth in the subtraction image was in concordance with histological depth, while the original radiograph showed shallower lesions.¹²⁶ Adding stannous fluoride to natural caries lesions in extracted teeth before the second exposure, subtraction images revealed a density increase in all the demineralised areas and with fewer false-positive scores than when assessing the original images.^127,128 Some studies induced demineralisation to simulate a lesion before the second exposure. Significant gray level changes in the demineralised areas could be detected by subtraction analysis.¹²⁹ It was, moreover, found that linear and logarithmically contrast-enhanced subtraction images did not differ in detecting the demineralised area,¹³⁰ but in another study, contrast-enhanced subtraction images were significantly more accurate in detecting demineralisation than conventional, digital and digitised radiographs.¹³¹

Based on the results from ex vivo studies, the value of subtraction radiography in the clinic was investigated. Thus, studies compared the outcome of subtraction with conventional viewing of two radiographs taken with a time interval. The results were promising and mostly in favour of the subtraction technique. In one study, the average intraobserver reproducibility was significantly higher for the subtraction images than when visually comparing the digitized film bitewing radiographs obtained with approximately two years between recordings.¹³² In opposition, a study found no differences in observer agreement when monitoring approximal caries behaviour in subtraction and two bitewing images; however, more progressing lesions were detected by subtraction.¹³³ Also, a study showed differences between the mean pixel values in subtraction images for demineralised, unchanged and remineralised lesions.¹³⁴ In occlusal surfaces, subtraction radiography was more accurate and reproducible than visual assessment of the two original digital images.¹³⁵

During the next decades, several longitudinal research studies in patients were conducted using subtraction radio-graphy as the outcome method for determining caries lesion progression or arrest. One study assessed the efficacy of sealing approximal lesions in a split-mouth design with both conventional bitewings and subtraction.¹³⁶ Subtraction radiography appeared to be the most sensitive method for assessing lesion progression. Two randomized clinical trials (RCT) were conducted on the efficacy of resin infiltration to arrest progression of approximal non-cavitated caries lesions over three years,^137–139 and another RCT assessed the therapeutic effect of infiltration versus sealing for controlling caries progression in approximal surfaces.¹⁴⁰ A study assessed whether total removal of infected dentine was fundamental for caries arrest after sealing of the surface¹⁴¹; and another studied lesion changes, also after incomplete carious dentine removal and tooth sealing.¹⁴² One RCT assessed the effect of resin infiltration of approximal caries lesions in halting progression of the lesion.¹⁴³ In primary molars, dentine remineralisation was assessed one year after glass-ionomer cement restorations with partial carious dentine removal.¹⁴⁴

In conclusion, the subtraction technique seems not to have entered the clinic in the daily routine when monitoring caries lesions; however, the technique is still widely in use in longitudinal research studies as outcome method for assessing caries lesion behaviour.

Extraoral 2D radiography – Panoramic radiography, bitewing panoramic mode, scanography, linear tomography

The intraoral bitewing technique may be unpleasant for some patient. After launching of panoramic radio-graphy in the 1970s,²² studies were conducted on the suitability of panoramic imaging for caries diagnosis.

In early clinical studies with patients, panoramic film radiographs were ineffective in detecting early interproximal carious lesions compared with BW and periapical exposures.¹⁴⁵ Agreement between a panoramic image and intraoral radiographs was on average 21% for enamel caries diagnosis and 44% for dentinal lesions,¹⁴⁶ and conventional BW radiographs gave greater dia-gnostic yield than the orthogonal and standard panoramic modalities.¹⁴⁷

There are few ex vivo studies with a solid validation that have compared panoramic extraoral BW modalities with traditional BW radiography. Multidirectional tomo-graphy and panoramic radiography performed as well as intraoral D-speed film for the combined assessment of approximal and occlusal caries.¹⁴⁸ When approximal surfaces were evaluated alone, D-speed film had significantly higher accuracy. Intraoral BWs and periapical radiographs also seemed to perform better than unfiltered and filtered digital panoramic images for approximal lesion detection.¹⁴⁹

More recent studies found no significant differences in detecting enamel approximal lesions with an intraoral sensor and various panoramic modes,¹⁵⁰ and an intraoral digital receptor with a post-processed sharpening filter showed similar diagnostic accuracy compared with an extraoral panoramic BW mode.¹⁵¹ Another study demonstrated that intraoral BW radiography with F-speed film was superior to extraoral BW and full panoramic radiography in diagnosing approximal caries.¹⁵² No differences were seen in posterior approximal caries detection between intraoral BWs, extraoral BW projection and a full panoramic view, while extraoral panoramic BWs were better than panoramic radiographs in visualizing open posterior interproximal contacts.¹⁵³ Recent developments in the panoramic technique have resulted in a novel program that displays 41 multilayer images from the buccal to the lingual aspects. When the multilayer imaging program was used, interobserver and intraobserver agreement improved for approximal caries lesion detection in patients compared with traditional views.¹⁵⁴ Most of the clinical studies are de facto on observer variation with no reference standard for true caries lesions, though.¹⁴

Another extraoral modality that existed in the 1990s was scanography, a radiographic technique that moves an orthogonal beam along the length of the examined structure. Enhanced digital scanograms were not different from F-speed film for approximal caries detection, while unenhanced scanograms obtained lower diagnostic accuracy.¹⁵⁵

Film and digital radiographs were compared with conventional linear tomograms, and no significant differences were found among the three imaging modalities for detection of approximal caries lesions.¹⁵⁶

In conclusion, it seems that bitewing radiography with intraoral receptors is still the most accurate technique for caries lesion detection. However, when intraoral recording is difficult, for example in patients with mouth-opening limitations and gag reflexes, current digital panoramic units have a fair diagnostic validity, and new modalities are available that may better display free interproximal spaces. This also means that when a panoramic image is indicated for other diagnostic tasks, caries lesions can be reported with reasonable accuracy.

Tuned-aperture computed tomography (TACT) – simulated 3D

As early as 1985, a pioneer researcher explored the idea of 3D dental radiography, and a model was developed based on the principle of tomosynthesis.¹⁵⁷ With eight-to-nine original exposure projections, information on tooth structure was extracted and expressed in three dimensions. About a decade later, a refined development of the tomosynthesis principle, now known as tuned-aperture computed tomography (TACT), was launched.¹⁵⁸

A number of studies were performed ex vivo on detection of caries lesions, comparing TACT with traditional bitewing-like film and digital image projections. TACT and film had similar diagnostic accuracies for caries detection by six observers, while a CCD sensor system obtained lower accuracy.¹⁵⁹ Other studies on approximal lesion detection showed that the TACT and 2D displays had almost similar accuracies.^160–162 For occlusal caries detection, one study found an increased accuracy for TACT.¹⁶³ Significant lower accuracy was observed, when the TACT modality was generated with less than eight image projections.¹⁶⁴ Differences were observed, neither with projection angular disparities between 10 and 40 degrees, nor with various reconstruction algorithms, for detection of approximal and occlusal lesions.^165,166 When observers viewed the original eight images that were recorded with different projection angles and compared with the TACT reconstruction, no differences were found between the two ways of viewing.¹⁶⁷

In conclusion, the TACT method was an innovative development, but it never entered the clinic for diagnosis of caries lesions, probably due to the large number of projections that were needed for reconstruction and the rather discouraging result that accuracy was not much improved. Further, true 3D volumetric methods meant for dentistry were developed and launched in this time period.

Cone-beam computed tomography (CBCT)

CBCT is a 3D radiographic modality, initially developed in Japan and Italy, which has been available for dentistry since the turn of the century.^168,169 A recent publication provides an overview of units over the last two decades.¹⁷⁰ There are obvious diagnostic advantages when an image volume can be sectioned in all planes displaying structural dimensions that are not visible with 2D imaging. In principle, a small field-of-view (FOV) is associated with less radiation than larger FOVs, but also spatial resolution plays a role.

Numerous ex vivo studies have been conducted to assess accuracy parameters of CBCT imaging compared with 2D radiography for detection of caries lesions. CBCT sections are usually performed in the mesio-distal and axial planes of the teeth for approximal caries detection and in the bucco-lingual (cross-sectional) plane for occlusal caries detection. The majority of studies have compared CBCT systems in a small FOV and high resolution examination with one or more intraoral radiographic receptors.

In ten studies of approximal caries detection, it was found that CBCT was no more accurate than intraoral film, solid-state sensors or PSP systems (bitewing-like or periapical exposure) for overall detection of carious demineralisation.^{151,171–179} Further, no significant differences were observed among CBCT units and among various FOVs and resolutions.¹⁷⁹ Two studies showed that CBCT was somewhat more accurate than intraoral receptors for measuring lesion depth,^171,180 while two showed the opposite for lesion detection (one used a large FOV).^181,182 Two studies found that CBCT had higher sensitivity for detecting dentinal lesions than a CCD sensor or a PSP system.^179,183 A CBCT examination with large voxel size was less accurate than a unit with smaller voxel size.¹⁷³ Assessment of section thickness showed that the highest accuracy was found for thinner sections (1 and 2 mm) with filtration modes than for 5 mm sections.¹⁸⁴

Eight studies assessed accuracy of occlusal caries detection. One study showed no differences between a CBCT examination (large FOV) and conventional radiographs,¹⁸⁵ while another found higher accuracy for CBCT than for bitewing exposure.¹⁸⁶ Some studies showed that CBCT performed better than intraoral digital receptors for deep lesion detection,^187,188 while there were no general differences when all lesions were considered among CBCT systems, film or digital receptors.^188,189 When measuring lesion depth, CBCT showed a significantly higher correlation with histology than did film and CCD sensor-based radiographs.¹⁹⁰ One study stated that an increase in cross-sectional CBCT slice thickness resulted in lower diagnostic accuracy, while accuracy was highest in a panoramic view.¹⁹¹

More recent studies have assessed caries lesions in connection with metal artefacts in the images. Accuracy for caries detection with CBCT under fixed dental prostheses depended significantly on the material for the prosthesis, that is zirconia full metal, metal-ceramic, lithium disilicate full-ceramic, and full ceramic and metal-acrylic.^192,193 Application of an artefact-reduction algorithm increased accuracy of approximal caries detection in connection with amalgam, composite, ceramic-based composite or CAD-CAM zirconia materials.¹⁹⁴ A study on the effect of orthodontic materials and artefact reduction on detection of approximal lesions found higher accuracy without stainless steel bracket and wires, but no effect of artefact reduction.¹⁹⁵

Three ex vivo studies compared CBCT in small FOVs with intraoral systems in a bitewing-like projection for detection of cavitated approximal surfaces in non-filled permanent teeth.^151,196,197 Significantly higher sensitivities were obtained with CBCT (Accuitomo and Orthophos Sirona) than with the intraoral receptors, which were not compromised by lower specificities. The difference in sensitivity between CBCT and the intraoral receptors was 20–50% and interpreted as clinically relevant.¹⁹⁷

It seems that only one clinical study has been conducted for detection of cavitated approximal lesions in teeth without restorations.¹⁹⁸ Patients suspected to have caries lesions after a visual clinical and a bitewing examination participated in a CBCT examination, and lesions with and without cavitation were recorded. Orthodontic separators were placed interdentally between the lesion-suspected surfaces. When the separator was removed, the surface was clinically recorded as cavitated or intact. Sensitivity was significantly higher for CBCT than for bitewings (average difference 33%), while specificity was not significantly different between the methods.¹⁹⁸

In conclusion, little is gained by using CBCT for detection of carious demineralisation compared with traditional intraoral radiographic receptors, and due to the relatively high radiation dose^199,200 and costs,^201,202 CBCT examination cannot at current be advocated as a the primary radiographic examination for diagnosis of caries lesions. In the above literature review, focus was not on dose measures or specific CBCT units. Recent CBCT units, however, work with ultra-low doses, thus in the coming decades more studies should focus on caries detection, not least diagnosis of surface cavitation, in small FOV low-dose examinations. Also, imaging artefacts exist, the primary ones being metal artefacts and patient movement artefacts.^203,204 Older restorative dental materials such as amalgam may affect image quality and even plast and glass-ionomer materials may contain barium, lead or zinc to lend radiopacity in bitewing radiographs, but which may disturb correct CBCT image reconstruction. The effect of artefact reduction algorithms has been assessed only sparsely. On the other hand, when the CBCT volume in reasonably high resolution is sectioned in thin slices, it could be imagined that a radiographic method for the first time is able to show whether an approximal surface is cavitated or intact, thus determining treatment choice. The available literature indicates that when a CBCT examination is performed for other diagnostic tasks, caries lesions can be reported with accuracies similar to intraoral exposures in teeth without fillings.

Machine-intelligence supported systems

In early publications, it has been well described that the reproducibility and accuracy of radiographic diagnosis of caries lesions depend on the skills of the observers and that great variation may exist among clinicians.^205,206 Automated interpretation of the image with the aim to standardise diagnosis and optimise accuracy has been a research object in dentistry for at least four decades.

Initially named “Computer-aided image analysis systems” or “Decision support systems”, programs were developed for detection of approximal caries lesions and published in the beginning of the 1980s.²⁰⁷ A program was consistent in detecting lesions and estimating lesion size in both patient radiographs and in images of extracted teeth.²⁰⁸ Ex vivo studies concluded that computer analysis was able to reproducibly detect and measure lesions^209,210 and more sensitive, but less specific, than human observers.²¹¹ The computer-analysis system was also useful in monitoring lesions in children.²¹² Using a system named “CariesFinder”, dental practitioners significantly increased their ability to diagnose caries correctly and their ability to recommend restoration for cavitated surfaces.^213,214

One system from the late 1990s, Logicon Caries Detector (LCD), was widely marketed to aid the clinician in detecting caries lesions in approximal surfaces. Independent studies found the system to be little more successful in finding lesions and ruling out sound surfaces than were trained clinicians. One clinical study assessed the reproducibility of detecting approximal lesions by repeating the automated analysis 10 times for each surface, and concluded that the LCD program was not consistent and provided different outcomes for the same surface in the same image.²¹⁵ Also, observer agreement did not improve using the program.²¹⁵ In an ex vivo study, specificities for the outcome of the LCD program were significantly lower than when the observers themselves assessed the RVG images.²¹⁶ Sensitivity was also lower for two observers on the diagnostic threshold caries in dentine.²¹⁶ Other studies found that LCD appeared to be more reliable in ruling out (both enamel and dentine) caries than in detecting caries with extremely low sensitivities for enamel lesions,²¹⁷ while sensitivity was around 70% for dentinal lesions in another study.²¹⁸ Also, accuracies were not significantly different between observers when working with and without the LCD system, when caries lesions of all depths were assessed. Accuracy was significantly higher although for LCD, when deep caries lesions were considered separately.²¹⁹ More recent software has likewise been tested with the outcome that dentinal lesions were reasonably well detected, while the system had limited ability in detecting enamel lesions.²²⁰

In recent years, more complex artificial intelligence (AI) systems have been developed within dental radiology.^221,222 Systems using texture analysis have been suggested for detection and characterisation of dental caries and various other maxillo-facial pathologies in panoramic images.²²³ Deep convolutional neural networks (CNN) is a rapidly expanding area of machine intelligence research. One study evaluated the efficacy of deep CNN algorithms for detection of dental caries in periapical radiographs with rather high accuracy.²²⁴ Very recently, CNN have been explored in detection of caries lesions of various depths in bitewings. A study showed an overall accuracy of the system of 80%, while dentists’ mean accuracy was lower (71%). The neural network was significantly more sensitive than dentists, while its specificity was not significantly lower.²²⁵ One study applied an artificial intelligence model, a multilayer perceptron neural network, which was shown to significantly improve accuracy of radiographic diagnosis of approximal caries compared to clinical observers.²²⁶

In a clinical trial, a neural network was able to predict the development of caries in patients with a relatively high probability.²²⁷ Two clinical studies using recently developed deep learning software found that both dental students and dentists performed better in detecting enamel approximal caries lesions in BW radiographs with the use of the deep learning assistive software than without the use of the software.^228,229

In conclusion, none of the computer-aided diagnostic systems from the 1980s and 1990 s have entered the clinic for use in the daily routine. Today, computer hardware can process large data sets, and recent developments and research of CCN illustrate the potential of the AI technology to help improve caries diagnosis. In the future, deep CNN algorithms may be expected to be more effective and efficient for detecting caries lesions.

Conclusions

The ultimate goals for a radiographic method that intends to diagnose caries lesions are (1) the method is reproducible and accurate in detecting lesions and in estimating lesion depth and (2) the method is able to distinguish between cavitated and non-cavitated lesions. Intraoral bitewing radiography is still the state-of-the-art method, despite its inability to accurately fulfil the second goal. Moreover, it should be born in mind that intraoral radiography is potentially an aerosol-generating procedure, and particularly in times of a pandemic, the choice of radiographic technique should be balanced against the likelihood of inducing gagging or coughing. A recent review offers a strategy for the practice of oral and maxillofacial radiology during pandemics.²³⁰

While the bitewing technique has not changed much over the years, the receptor and image display have undergone a transition from film-based to digital image acquisition and display in the clinical environment. Herein, most intraoral methods are fairly accurate for detection of the demineralisation in approximal surfaces and in occlusal surfaces in dentine. The deeper the lesion, the more accurately can it be detected. Still, interobserver variation exists in caries diagnosis, and new AI systems may aid the clinician in establishing more reproducible diagnoses and treatment decisions. It may be doubtful although, whether AI systems can help in reaching the second goal, to determine whether or not a surface is cavitated, when based on 2D radiographic modalities. 3D modalities may be more accurate in distinguishing between cavitated and non-cavitated lesions, but are not yet suitable as the primary diagnostic method, when caries lesions are clinically suspected, due to dose and artefact issues, expenses and workload.