Filtered back projection vs. iterative reconstruction for CBCT: effects on image noise and processing time

Abstract

Objectives:

To assess the effect of standard filtered back projection (FBP) and iterative reconstruction (IR) methods on CBCT image noise and processing time (PT), acquired with various acquisition parameters with and without metal artefact reduction (MAR).

Methods:

CBCT scans using the Midmark EIOS unit of a human mandible embedded in soft tissue equivalent material with and without the presence of an implant at mandibular first molar region were acquired at various acquisition settings (milliamperages [4mA-14mA], FOV [5 × 5, 6 × 8, 9 × 10 cm], and resolutions [low, standard, high] and reconstructed using standard FBP and IR, and with and without MAR. The processing time was recorded for each reconstruction. ImageJ was used to analyze specific axial images. Radial transaxial fiducial lines were created relative to the implant site. Standard deviations of the gray density values (image noise) were calculated at fixed distances on the fiducial lines on the buccal and lingual aspects at specific axial levels, and mean values for FBP and IR were compared using paired t-tests. Significance was defined as p < 0.05.

Results:

The overall mean for image noise (± SD) for FBP was 198.65 ± 55.58 and 99.84 ± 16.28 for IR. IR significantly decreased image noise compared to FBP at all acquisition parameters (p < 0.05). Noise reduction among different scanning protocols ranged between 29.7% (5 × 5 cm FOV) and 58.1% (5mA). IR increased processing time by an average of 35.1 s.

Conclusions:

IR significantly reduces CBCT image noise compared to standard FBP without substantially increasing processing time.

Introduction

Dental radiographic images provide useful diagnostic information about pathologies and diseases of the teeth and jaws, contribute to clinical diagnosis, and contribute to treatment planning. ¹ This clinical contribution depends on the image quality of the radiographic modality. ² For cone-beam computed tomography (CBCT), image quality, in terms of contrast resolution and noise, varies greatly between different CBCT units and acquisition settings with a wide range in patient radiation dose. ³ One method of improving image quality is to use iterative reconstruction (IR) rather than the more commonly employed standard filtered back projection (FBP) algorithms to reconstruct data. ^4,5

Image reconstruction in tomographic medical imaging refers to the mathematical process of applying complex algorithms to raw data obtained from multiple projections to create cross-sectional images. FBP reconstruction is the most common medical image reconstruction method. In CBCT, Feldkamp-Davis-Kress algorithm (FDK) is the most commonly used FBP technique for image reconstruction ⁴ and is computationally relatively simple and fast. The image is formed from the back projection of multiple attenuation profiles at different projections. A mathematical filter is applied to the data prior to back projection. The ramp filter suppresses low spatial frequency components of the attenuation profiles, correcting the blur intrinsic to the projection. ^4,6

IR techniques use repeated comparisons of the projection data and the current back-projected image estimate. The process of comparing the two is done until it reaches a specified level of acceptability defined through a specific stopping criterion. ⁴ This method requires more computation time. ^4,7

The benefits and disadvantages of IR compared to FBP in medical computed tomography (CT) are firmly established. ⁸ Optimization of CT image reconstruction using IR was initially introduced as a dose reduction strategy. At the same acquisition exposure, IR can improve CT image quality by reducing image noise and minimizing artifacts from high-density materials from beam hardening. Beam hardening is more pronounced with low-energy beams, such as those used in CBCT, and denser materials, like metallic restorative materials and implants. ^9–12

Most dental CBCT systems use FBP reconstruction algorithms for processing raw data. ^4,13 The use of IR potentially allows a reduction in exposure parameters and, therefore, a lower patient radiation dose to produce images of comparable image quality. To the authors’ knowledge, currently, only one manufacturer (Midmark EIOS, Midmark, Dayton, OH, USA) allows users to choose either a “Standard” reconstruction (FBP) or “Enhanced Processing” (IR) after each image acquisition. Enhanced Processing is an IR mode that aims to reduce image noise, particularly with low-dose protocols, and can be applied with or without metal artefact reduction (MAR). Other manufacturers also offer various reconstruction modes for metal or motion artefact and/or noise reduction – these may employ IR algorithms; however, this information is proprietary.

This study aimed to compare the effects of IR and standard FBP CBCT reconstruction modes on image quality by measuring image noise and assessing if MAR, the presence of an implant, and acquisition parameters modulated this effect. A secondary aim was to determine the practical effect of IR on processing time.

Methods and materials

Phantom

A human dentate mandible was used as the phantom (Figure 1). The lower right first molar was extracted, and an initial CBCT scan was acquired. A single titanium implant was placed, and a second CBCT scan was obtained. The mandible was embedded in ballistic gelatin made by mixing 48 g of unflavored and uncolored gelatin, 200 ml of glycerin and 500 ml of water. ¹⁴ This ballistic gelatin was used as a homogenous soft tissue equivalent material and ensured that the mandible was in the same position for repeated imaging. A block of Sil-Tech special condensation silicone (Ivoclar Vivadent Inc, Amherst, NY, USA) was made utilizing the manufacturer’s instructions. It was placed on the buccal cortex of the anterior ridge of the center of the mandibular alveolar ridge. This allowed for proper orientation and selection of an axial slice during the image analysis process.

Figure 1.

Lateral (a) and superior (b) views of the mandible with the right first molar extracted are shown embedded in ballistic gelatin. Right lateral view of the mandible (c) with the titanium implant inserted in the lower right molar region.

Image acquisition and reconstructions

The Midmark EIOS Model V8201 CBCT unit (Midmark, Dayton, OH, USA) was used because it provides a choice of either standard (FBP) or enhanced (IR) reconstruction methods after image acquisition, both with and without MAR. Image acquisition was performed in two phases: before and after implant placement. The operating features of this unit are displayed in Table 1. Twenty-four scans were acquired in each phase with various combinations of acquisition parameters (Table 2).

Table 1.

Specifications of the Midmark EOIS Model V8201 and computer

Specification
X-ray tube voltage	60–84 kV (fixed at 84 kV for CBCT scans)
X-ray tube current	4–14 mA
Nominal focal spot	0.5 mm
Rotational arc	220o
Exposure time	3.5–5.1 s
Field of View available (cm)	Adult:5 × 5, 6 × 8, 9 × 10 Child: 5 × 5, 5 × 8, 8 × 8
Voxel sizes available	78 µm, 156 µm, 195 µm
Computer specifications	HP Z4 G4 Workstation (Intel Xeon W-2123, 32 gb RAM, NVIDIA Quadro P400, SSD Storage)

Table 2.

Details of exposure parameters used for the acquisition protocols of CBCT scans

Field of view	Resolution	mA	kVp
9 × 10 cm	LD	4, 5, 6	84
	SD	6, 7, 8
	HQ	10, 12, 14
6 × 8 cm	LD	4, 5, 6	84
	SD	6, 7, 8
	HQ	10, 12, 14
5 × 5 cm	SD	6, 7, 8	84
	HQ	10, 12, 14

HQ, high quality; LD, low dose; SD, standard.

All scans were reconstructed under four conditions: FBP and MAR, FBP without MAR, IR with MAR, and finally, IR without MAR, providing a total of 192 CBCT volumes for analysis. Computational processing time was recorded, in seconds, for each reconstruction.

Image analysis

Image noise was considered as a metric for image quality in this study, and it was measured as the standard deviations of gray density values on multiple fiducial lines within a homogenous soft-tissue density material (i.e., ballistic gelatin). Images for each exposure and condition were exported in DICOM format. ImageJ (National Institute of Health, Bethesda, MD, USA) was used to visualize and analyze specific axial images. Radial transaxial fiducial lines were created relative to either the right mandibular first molar site (initial scans) or an implant placed at this site (second scans). Fiducial lines were located entirely within the ballistic gelatin and did not include bone, tooth, or other material. Standard deviations of the gray density values (image noise) were calculated at fixed distances on the fiducial lines on the buccal and lingual aspects at specific axial levels. Mean values for IR and FBP were compared using paired t-tests.

An axial reference slice was selected at the level where the top of the buccal cortex block was observed. A line was drawn following the angulation of the alveolar process and passing in the center of the implant site. Six parallel fiducial lines were then added to the image as regions of interest at 0.5 cm, 1 cm, and 1.5 cm buccally and lingually from the center of the implant site (Figure 2). The standard deviation of gray values was recorded for each fiducial line. The procedure was repeated on five axial slices above and below the selected initial slice. The average standard deviation of gray values along these lines was recorded for each reconstruction. The fiducial lines were stored in the ROI manager tool on ImageJ to standardize their length and position among the images within each field of view.

Figure 2.

Matrix of six representative images acquired at 10mA, HQ resolution, and reconstructed with MAR at the level of the axial reference plane images showing the location of fiducial lines. Each row represents one of two different reconstruction algorithms: filtered back projection (FBP) or iterative reconstruction (IR). (a) FBP/9 × 10 cm; (b) IR/9 × 10 cm; (c) FBP/6 × 8 cm; (d) IR/6 × 8 cm; (e) FBP/5 × 5 cm; (f) IR/5 × 5 cm.

Statistical analysis

For each acquisition parameter, the means of image noise obtained from both FBP and IR were compared using paired t-tests. All analyses were conducted in SPSS (Version 28, IBM, USA), with significance defined as p < 0.05.

Results

Figure 3 displays the mean noise obtained from different acquisition protocols reconstructed using the FBP and IR methods. The overall estimated image noise (mean ± SD) for IR was 99.84 ± 16.28 and 198.65 ± 55.58 for FBP, representing an average of 49.7% reduction in image noise when IR is used.

Figure 3.

Histogram comparing the effect of reconstruction algorithm (FBP or IR) on mean image noise (mean standard deviation of gray values) ± standard error for each condition, including the presence of an implant, FOV, resolution, application of MAR, and increases in milliamperage (mA). All comparisons of FBP vs IR were significantly different (p < 0.001).

When no implant was present in the field of view, IR significantly (p < 0.05) decreased image noise by 52.7%, on average, compared to FBP. When an implant was in the field of view, IR decreased image noise (p < 0.05) by 47.1% on average. For all other acquisition parameters considered (field of view, resolution, MAR, milliamperage), IR also significantly reduced image noise compared to FBP (all p < 0.05).

Noise reduction with IR ranged between 29.7 and 53.9% among different fields of view. The 5 × 5 cm FOV had the least influence of the selected reconstruction method. Regarding the resolution settings, IR showed a greater reduction in noise for scans in LD resolution (56%) compared to those in SD resolution (47.5%) and HQ resolution (46.3%). For reconstructions with or without MAR, the reduction in image noise was 47 and 52.5%, respectively.

Low milliamperage settings (4mA, 5mA) had the highest impact of IR on image noise decrease (56.4 and 58.1%, respectively), while high mA settings (12mA and 14mA) had lower noise reductions with IR (45.2 and 44.5%, respectively).

Table 3 shows the recorded time for the computational processing of each reconstruction. The additional time required to process IR, compared to FBP reconstruction, ranged between 21 s and 49 s, with an average of 35.1 s (27.3%).

Table 3.

Computational processing time (in seconds) recorded for filtered back-projection and iterative reconstructions from different acquisition parameters

Field of view	Resolution	MAR	Filtered back projection	Iterative reconstruction	Additional processing time (%)
9 × 10	LD	-	59	93	34 (58%)
		+	104	133	29 (28%)
	SD	-	109	153	44 (40%)
		+	201	249	48 (24%)
	HQ	-	104	150	46 (44%)
		+	199	248	49 (25%)
6 × 8	LD	-	60	81	21 (35%)
		+	96	121	25 (26%)
	SD	-	60	81	21 (35%)
		+	99	121	22 (22%)
	HQ	-	62	83	21 (34%)
		+	97	123	26 (27%)
5 × 5	SD	-	156	194	38 (24%)
		+	248	297	49 (20%)
	HQ	-	152	193	41 (27%)
		+	248	296	48 (19%)
Mean			128.4	163.5	35.1 (27%)

-, off; +, on.

Discussion

The results of this study showed that IR decreases image noise when compared to the FBP reconstruction method for all the acquisition parameters tested, both with and without the application of MAR. Image noise was measured via the standard deviation of gray values along the utilized fiduciary line markers drawn on areas of homogenous density (ballistic gelatin). Smaller numerical values of this uniformity parameter ultimately indicate more uniform gray values. More uniformity between gray values produces less image noise, producing a higher quality image. ¹⁵

Parameters such as field of view, resolution, and mA impact the difference in image quality between IR and FBP. Noise reduction was less expressive among scans acquired with 5 × 5 cm FOV (29.7%), compared to those with larger FOV (53.3% for 6 × 8 cm and 53.9% for 9 × 10 cm). The lower the resolution, the greater the noise reduction achieved with IR (56% vs  46.3% for LD and HQ resolutions, respectively). Lower mA values yielded greater noise reduction when IR was applied (44.5% reduction at 14mA vs 58% reduction at 5mA). These results demonstrate that the positive impact of IR on image quality tends to be more pronounced among the acquisition parameters that are usually selected when aiming at lower radiation doses.

An implant placed in the field of view causes known image artifacts due to beam hardening and streaking effects. These effects can be quantified by determining the ratio of the signal difference or the contrast to the noise level. ^16,17 The fiduciary locations used in this study were determined by lines drawn at fixed distances from the implant site and positioned such that dark and bright streaks generated on the labial and lingual aspects of the implant in the extraction socket of the bone were intersected by these lines. Therefore, decreases in gray level values may have been caused by a reduction in noise and an overall decrease in these streaking effects.

Iterative reconstruction algorithms help to enhance image quality by utilizing more precise modeling of the acquisition process. This process works by completing many repetition scans of an object to synthesize a higher-quality scan. However, with this increased precision, an increased computational time is anticipated, and on average, the processing time for IR increased by 27% compared to FBP processing. However, the actual time was about half a minute, which is clinically insignificant compared to the overall processing time. ¹⁸

Our study focused on the image quality measured by noise between two different reconstruction methods. By averaging noise at specific distances, we were able to make broad conclusions that image quality is increased with the use of IR. Other studies have been conducted that focus on the intensity of the noise and artefact at specific distances away from the implant. Fontenele et al ¹⁹ evaluated the magnitude of artifacts, as measured using contrast-to-noise ratio, for both titanium and zirconium implants at various distances and angulations to image quality with CBCT scans using a similar methodology. They found that implant-related artifacts are greater closer to the implant region and can affect regions as far as 3.5 cm from the high-density object. They also found that artifacts are most prominent at a 90 degree angle in relation to the mandibular axis.

Dental professionals rely on diagnostic imaging, including CBCT, to guide diagnosis and plan treatment. Therefore, factors affecting image quality are essential in accurately depicting structures and interpretation. The results of this study indicate that IR methods improve image quality in the posterior mandible region. The tendency of image noise and other artifacts to decrease with IR at other regions of interest and on diagnostic performance for different tasks should also be investigated.

Moreover, our study shows that the selection of IR does not add appreciable clinical wait time to the patient experience. However, reconstruction times are computer specific, and the Midmark unit incorporates a dedicated video graphics card to minimize the expected processing time with IR algorithms. This has been successful in this instance in reducing overall computational times overall to about 35 s.

Conclusion

The use of iterative reconstruction can reduce image noise by approximately 40–70% compared to filtered back-projection. Iterative reconstruction does not add significant clinical time to the patient experience.