Can advanced edge enhancement software improve image quality to visualise tubes, catheters and wires in digital chest radiographs?

Introduction: This study aimed to test whether Advanced Edge Enhancement (AEE) software could improve the localisation of tubes, catheters or wires, while also affecting the overall image quality in chest x-rays (CXR). Methods: In total, 50 retrospective CXRs were included. All images were obtained utilising the Canon X-ray system (CANON/Arcoma Precision T3 DR System, Canon Europe, Amsterdam, NL) with a CXDI-810C wireless detector. A clinical image, plus three additional AEE algorithms were applied using post processing (two intensity variations 1 and 4) on all CXRs totalling 350 different images. Three radiologists evaluated the images using a subjective Absolute Visual Grading Analysis (VGA). The clinical images used in post processing were not applied as reference in the analysis. Each radiologist graded the images separately in a randomized order, with a score of three indicating suitability for diagnostic assessment. Results: The three AEE algorithms contributed to an overall improvement (average 16 e 49%) in visualisation of tube, catheter or wire on CXR images. The Mann e Whitney U tests showed a statistically signi ﬁ cant (p < 0.05) improvement in contrast resolution and sharpness, indicating an increased ability to differentiate tubes, wires or catheters tips from surrounding tissues. For the noise criterion, not applying anyAEEalgorithmshowedasigni ﬁ cantlyhigherhomogeneityinsofttissue(p < 0.001),reducingtheability to visualise soft tissue. The high-intensity catheter algorithm was the only algorithm to achieve a statistically signi ﬁ cant (p ¼ 0.017) increase in the ability to differentiate pulmonary tissues of similar density. Conclusion: An overall improvement in the visualisation of tube, catheter and wire placement was obtained using the three AEE-algorithms. The bone and catheter algorithms showed the highest consistency, with the small structure algorithm underperforming in resolution and low contrast resolution. In general, image noise increased regardless of algorithm type or applied intensity. The AEE-algorithms should therefore be seen as a supplementary tool to the clinical image protocol, while having the potential to improve image quality to speci ﬁ c clinical situations. Implications for practice: AEE ﬁ ltered images appear to be a supplement to the current practice of using CXRs in the diagnosis in placement of catheters, tubes and wires in the chest region. The use of AEE-algorithms has the potential to improve the daily work in clinical practice, which serves the basis for further investigation of its effect on radiographic practices.


Introduction
Digital Radiography (DR) is the standard approach for determining the correct placement of tubes, catheters and wires in the chest region. 1,2Incorrect placement can lead to complications and therefore chest X-rays (CXR) are important in ensuring correct placement. 3Venous catheterization, intercostal -or nasal tubing are examples of cases being dependent on radiological evaluation, to ensure proper pleural drainage, administration of chemotherapy or nasal gastric feeding. 4,5Tubes, catheters and wires often include plastic like materials, with a small metallic component, that can be difficult to distinguish from anatomical structures on CXR.Therefore, chest images have to be of a high standard to be considered a helpful tool for clinicians.The implementation of enhancement software for DR systems has improved the visual characteristics of X-ray images in other anatomical areas.6e8 Enhancement software enables radiation dose reduction, complying with the ALARA principle.6e9 Even though the enhancement software is present in clinical settings and is being used to optimise CXR, the evaluation of its perceived usefulness in patients with diagnostic challenges, such as venous catheter or pleural drainage placement, are lacking in the scientific literature.From a clinical standpoint, this type of software could potentially provide the radiologist with enhanced image quality, and eventually improve patient outcome.
In 2019 Canon introduced an addition to the Canon CXDI control software (Advanced Edge Enhancement (AEE), CXDI Control software NE V.217, Canon Europe, Amsterdam, NL), offering the opportunity to enhance the visual quality of multiple anatomical structures, applying three different user algorithms (catheter, bone and small structure). 10This software gives radiologists optimised images for clinical decision making.However, to our knowledge, no previous studies have explored the usefulness of applying AEE software particularly to chest radiographs.
This study examines how AEE software affects the visual image quality of CXR in patients with tubes, catheters or wires within the thorax.

Patient study
A retrospective study was performed with ethical approval from hospital management (Acadre number: 20/1718).Data was collected during an 11-month period from January to December 2020.Informed patient consent was waived for inclusion in this study, and collected images were anonymised.
Fifty consecutive Anterior-Posterior (AP) CXRs from patients with either a tube, catheter or wire present, were included in this study (see Table 1).Due to the need for observation of subacute patients participating in CXR examinations, the radiology department recommends using an AP positioning.All included patients had their CXR's performed at the radiology department, either standing or lying down depending on their physical condition.Each of the 50 CXRs were post-processed with three AEE algorithms, Catheter, Bone and Small structure abbreviated as (cat, bon, sms).The three AEE algorithms were applied at two intensity levels 1 (low intensity (L)) and 4 (high intensity (H)), as described in a recently published pilot study. 6ach CXR had six AEE images and one additional image without AEE algorithm applied (clinical image), resulting in a total of 350 images.In total, 50 patients were included in this study, with 48% being male, see (Table 2).Due to missing registration of demographic data from ten of the included patients, the exposure parameters used for the CXR, were evaluated to get an estimate of the different patient sizes.

DR system and technical settings
A Canon/Arcoma Precision T3 DR System (Canon Europe, Amsterdam, NL) with a Canon CXDI-810C wireless detector was used for all CXR. 11Radiographers created images with clinical exposure settings based on each patient.Table 3 summarises the technical exposure data.
A physical Arcoma JPI (Arcoma AB, V€ axj€ o, Sverige) anti-scatter grid, focused to 180 cm for standing and 110 cm for supine CXR, with a ratio of 10:1 was applied followed by post-processing using the CXDI control software NE Version 217 (Canon Europe, Amsterdam, NL).For information about parameters, (see Table 4).

Visual grading analyses
Visual image quality was evaluated using an absolute VGA. 12,13ach CXR was rated by three certified radiologists, with 4, 12 and 21 years' of experience, respectively.The three radiologists were introduced to the algorithm and the aim of the study in the initial phase.Thirty images were randomly duplicated and used to determine intra-observer agreement.This calculation increased the number of images per radiologist to 385.
The 385 images were evaluated randomly and the radiologists were blinded to the applied algorithm.During the image quality analysis, the clinical image was not used as a mean of reference.Rather it was used to evaluate the effect when applying the different algorithms.The VGA image criteria were based on the European Guidelines but altered to suit a more clinical approach, including the visibility of tubes, catheters or wires. 2,14,15More specifically, image quality was evaluated according to various parameters in the CXR, (see Table 5).Criteria 1 and 2 evaluated all types of tubes, catheters and wires in the study.Criterion 1.1 was created to supplement criterion 1 in relation to materials inserted into the lung cavity.If there was more than one tube present, the radiologists were explicitly instructed which tube to rate.A Likert rating scale from 1 to 5 was applied to quantify each VGA criteria. 16 rating of 3 suggests a clinically acceptable or improved image quality, while a score <3 suggests a reduced image quality.All images were evaluated using an EIZO diagnostic monitor (RX 360 colour LCD monitor, Eizo, Hakusan, Japan). 17Each image was presented in Viewdex 2.0, with unlimited evaluation time. 18The window-width (WW) and windowelevel (WL) could be altered during the VGA; the Zoom and Pan function could also be used.A 1h training session was given to each radiologist before the VGA, to explain the criteria, rating scale, ViewDex 2.0 functions and get an impression about what kind of images to expect in the analysis.

Statistical analysis
A Visual Grading Analysis Score (VGAS) was used to analyse the differences in VGA score for all five criteria to estimate whether or not the different algorithms had an effect on image quality (see Fig. 1). 12,13ll statistical analyses were performed using SPSS 26 (IBM, New York, USA), and Microsoft Excel was used to visualise the data (Microsoft, Washington, USA). 19A ManneWhitney U test for nonparametric and unpaired data was used to evaluate each VGA criteria, comparing the score of the AEE images to that of the clinical image, using a significance level of 5%.The VGA results were visualised using bar charts and presented as mean of the scoring.A linear weighted Cohen's kappa-test with agreement intervals evaluated the intra-observer agreement.To interpret the degree of agreement between the observers, Landis and Kochs guideline for interpreting kappa values were used. 20The guideline is divided into the following intervals: 0.0 to 0.20 indicating slight agreement, 0.21 to 0.40 indicating fair agreement, 0.41 to 0.60 indicating moderate agreement, 0.61 to 0.80 indicating substantial agreement and 0.81 to 1.0 indicating a perfect agreement.Finally, the inter-observer agreement was performed using the Fleiss's kappa test. 21,22

Results
Overall, the VGAS suggested improved image quality when the AEE algorithms were applied, (see Fig. 2).
Criteria 1 and 2 were statistically significantly higher for all AEE options.In addition, the high-intensity level for all algorithms was rated higher in criteria 1 and 2 than the low-intensity level algorithms.Interestingly, VGA scores for criterion 4, was low for the small structure algorithm at high-intensity compared to the bone, catheter and clinical images.The high-intensity catheter algorithm was the only algorithm to show a significant difference in low contrast resolution compared to the clinical image.In criterion 5, not applying AEE algorithms resulted in a statistically significant amount of image noise.Furthermore, low-intensity AEE algorithms were preferred over high-intensity in criterion 5.
The ManneWhitney U test showed a significant difference (pvalue ¼ 0.001) for the VGAS of the AEE images in criteria 1, 2 and 5 (see Table 6).For criterion 3, the low-intensity catheter and both small structure AEE images showed no statistically significant difference, while they're counterparts did.For criterion 4, all AEE images except for the high-intensity catheter algorithm, showed no statistically significant difference.
The intra-observer agreement analysis showed a Cohen's linear weighted kappa of (0.42) for radiologist 1 and 2, (0.29) for radiologist 1 and 3 and (0.23) for radiologist 2 and 3, respectively, indicating a fair to moderate agreement between radiologists. 22The total inter-observer agreement between all observers was calculated using Fleiss' kappa and showed no to slight agreement (see Table 7).

Discussion
Fifty retrospective clinical AP CXRs of tubes, catheters or wires inserted were included based on one clinical image (without the AEE software added) and six AEE images.Results from the ManneWhitney U test showed an improvement in image quality for criteria 1 and 2, when using the AEE algorithms.For criteria 3 and 4, improvements were seen for only a finite number of algorithms.
Overall, the AEE images showed the highest VGA grading, for all VGA criteria except for criterion 5 (noise).Clinicians reported improved image quality when images were enhanced with AEE algorithms.The results from this study correlate well with an earlier study that investigated the effect of using the AEE algorithms on hand radiographs. 6This study reported that the AEE algorithm improved both technical and visual image quality. 6To our knowledge, there is no other published research focusing on this specific software.The closest comparison is novel edge enhancement software that increases sharpness, spatial resolution and  image quality under different contrast levels in pneumothorax and dental DR. 23e25 The strength of the AEE algorithms (cat, bon and sms) could be adjusted between 1 and 10, with one being the lowest level of enhancement.This study only examined intensity levels 1 (L) and 4 (H), as higher intensity levels had previously reported significantly increased image noise. 6Furthermore, application specialists (Canon Medical Systems) had advised adamantly against maximum intensity levels as noise levels of images could affect the clinical evaluation process.
The ManneWhitney U test for criteria 1 (contrast) and 2 (sharpness) indicated a significant improvement in VGA gradings for high and low algorithm intensity levels.In addition, the highintensity level was graded better than the low-intensity for all three algorithms.Previous studies have indicated that low and high-intensity level algorithms increase image quality in both contrast resolution and sharpness. 6However, results from this study indicate that high intensity levels outperform low-intensity algorithms within these exact image criteria.This result indicates that the use of AEE algorithms can improve the radiologist's ability to evaluate the placement of inserted medical devices in relation to the surrounding anatomy.
For criterion 5 (noise), the ManneWhitney U test showed that the gradings were significantly higher when not applying the AEE algorithms.In addition, the low-intensity levels performed better than the high-intensity algorithms, which can be explained by the lower amount of noise added by the algorithm's mathematical calculations.A previous paper by Nabavi et al., 25 reported improved visual acuity when using MATLAB scripts to enhance the edge of optotypes in background images.Jefferson et al. 23 reported similar results by demonstrating how image sharpening software affected noise in dental DR images.Application specialists from Canon medical have previously explained that higher intensity levels will increase image noise and therefore the results for criterion 5 (noise), were anticipated.It was assumed that the use of AEE algorithms would improve sharpness of small structures.However, criterion 3 (spatial resolution) confirmed that only the high and low-intensity bone, and high-intensity catheter algorithms significantly improved spatial resolution.Criterion 4 (low contrast), indicated that the highintensity catheter algorithm was the only algorithm to produce better low contrast resolution, compared through the ManneWhitney U test.The remaining algorithm intensity levels were slightly better or worse than the clinical image.In contrast, a previous study showed algorithms of corresponding intensity producing significantly better low contrast resolution. 6here are several limitations in this study.Firstly, a limited number of patients were included, and as a consequence, a limited number of tube, catheter and wire types are present for evaluation.Therefore, only general conclusions can be drawn from the statistical analysis.Secondly, demographic data was missing for 10 patients, which to some extent will affect the evaluation of the effect of the AEE algorithm.To accommodate the missing data, the exposure parameters were used as a measure for patient size.Thirdly, the included CXR images had different exposure parameters applied due to the study's retrospective nature.However, clinical practise and the ALARA principle dictates adjusted parameters for different patient types.These differing parameters potentially affected the radiologist's VGA grading.Last, the intra and inter-observer agreement varied from fair to moderate and none to slight.As the VGA is a subjective evaluation of image quality each radiologist will perceive the quality differently.Hence, the outcome of this study is not unexpected and may have several causes, such as differences in age, diagnostic experience, preferences, etc., among the radiologists.Independent reading by three radiologists was performed exactly for this reason.Time is also an interesting parameter in psychometric experiments and influenced the agreement and changes throughout the analysis.Though the study illustrates a clinically realistic scenario, the observed inconsistency in the agreement analysis weakens the validity of the perceived algorithm performance, another article will look into these details.

Clinical implication
Based on the results found in this study, the use of AEE algorithms have the potential to support radiographers when examining subacute patients.Working with these types of patients requires a good understanding of optimising exposure parameter to produce CXR's of good image quality.The use of AEE algorithms has the potential to reduce the number of retakes in this patient group, by improving visualisation of tubes catheters and wires on initial radiographs, rather than requiring modification of exposure parameters and irradiation of the patient.Therefore, AEE algorithms have the potential to reduce dose and save time.
Future studies are required to further investigate which AEE settings are most effective for different patient sizes, medical devices, sexes and patient positioning.Furthermore, the algorithm should also be evaluated in clinical practice to measure its effect during a clinical workday.

Conclusion
This study indicated that the three AEE algorithms increased the perception of the tubes, catheters and wires in the included patients.Therefore, AEE algorithms are a feasible tool for evaluating the placement of medical devices in CXRs.The bone and catheter algorithms have shown to be helpful for the general screening.The small structure algorithm performed equally or worse than not applying the AEE algorithm in criteria 3 (spatial resolution) and 4 (low contrast resolution), making it applicable for specific situations only.With this in mind the AEE algorithm has the potential to optimise the evaluation and localisation of medical devices in CXR without requiring any retakes.The algorithm automatically post-processes images with the different settings, making the algorithm a supplement to the clinical image protocols Disclosures None.

Figure 1 .
Figure 1.Formula for calculation of VGAS.G (abs) is the absolute grading of a specific image (i), structure (s) and observer (o).I, S and O are the numbers of images, structures and observers respectively.

Figure 2 .
Figure 2. VGAS comparison for all radiologists distributed as criteria and algorithm intensity levels.

Table 1
Tube, catheter and wire types present in patient images.

Table 3
Technical exposure data used for the 50 CXR's, with Dose Area Product (DAP), Reached Exposure value (REX) and Exposure index (EI).

Table 4
Software setup for the 50 CXR's and the clinical images.

Table 5
The VGA image criteria, definition and connection to technical image quality parameters.

Table 6
Results of ManneWhitney U test representing rank-sum and p-value.All statistically significant results are marked with (*).

Table 7
Fleiss kappa-and Cohen*s linear weighted results.