Validation of a Recently Developed Fluoroscopic Video Analysis Workstation (Radwisp™) for the Reliable Diagnosis of Acute Pulmonary Thromboembolism

Abstract

Objective

Radwisp™ is a fluoroscopic video analysis workstation recently developed to evaluate pulmonary circulation, thereby obviating the need for contrast medium or breath-holding. This study validated Radwisp as a diagnostic tool for acute pulmonary embolism (APE) and evaluated its potential utility in patients with symptoms of suspected APE.

Methods

The study included 10 patients (mean age, 69±16 years old) who were admitted to our hospital for suspected APE based on symptoms and physical examination findings between January 2020 and April 2021. Contrast-enhanced computed tomography (CT) and cineradiography, based on standard radiographs for the creation of a Radwisp image, were performed on the same day. Of the 10 cases of suspected APE, 7 were definitively diagnosed by CT with APE, and 3 were definitively diagnosed as not having APE. Fifty physicians (25 cardiologists and 25 residents) were blinded to patient information and CT images and asked to diagnose the presence of APE based solely on the Radwisp images.

Results

A total of 250 diagnoses were made by cardiologists and 250 by residents. Among the cardiologists, the sensitivity and specificity of the Radwisp-based analysis were 91% and 48%, respectively, and the positive and negative predictive values were 80% and 69%, respectively. Among the residents, the sensitivity and specificity were 88% and 35%, respectively, and the positive and negative predictive values were 76% and 55%, respectively.

Conclusion

This study showed an initial validation of Radwisp for diagnosing APE, revealing a high sensitivity but not yet achieving a high specificity. Further studies with a larger number of cases are needed to thoroughly evaluate the diagnostic performance.

Introduction

Pulmonary thromboembolism and deep vein thrombosis (DVT) are manifestations of a pathological entity known as venous thromboembolism (VTE). Although VTE was once considered rare in Japan, its incidence is now steadily increasing, mainly due to aging of the population and to association of VTE with cancer, which is also on the rise (1-3).

Acute pulmonary embolism (APE) is the most serious presentation of VTE and is one of the leading causes of death worldwide (3), thus requiring a timely diagnosis. Contrast-enhanced chest computed tomography (CT) is the most important imaging tool for such a diagnosis (4). However, there are drawbacks, such as the need for contrast medium and patient breath-holding to obtain high-quality images. A fluoroscopic video analysis workstation (Radwisp™; Paramevia, The Central, Singapore) was developed to allow for the evaluation of the pulmonary circulation without the use of contrast medium or the need for breath holding. The respiratory function and pulmonary blood flow were assessed by subjecting non-filtered and filtered fluoroscopic images for a frequency analysis (5).

We used this system for the simultaneous evaluation of the respiratory function and pulmonary blood flow in two cases of APE (5,6). If clinically realized, this technology will facilitate primary screening for patients presenting with suspected APE by clarifying the presence of the pathophysiological characteristics of APE. Therefore, we assessed the accuracy of a Radwisp-based diagnosis compared to that of a CT-based diagnosis in patients presenting with suspected APE.

Materials and Methods

Study design and patient enrollment

The study included 10 patients in whom APE was either suspected or could not be ruled out based on symptoms such as sudden dyspnea or echocardiographic or electrocardiographic findings suggestive of APE at the time of presentation between January 2020 and April 2021. Of the 10 cases of suspected APE, 7 were confirmed based on contrast-enhanced CT and clinical findings to be cases of APE. APE was ruled out in the remaining three cases, and the final diagnoses in these individuals were heart failure, non-ST-elevation myocardial infarction with heart failure, and chronic obstructive pulmonary disease (one each).

All patients met the following eligibility criteria: 1) a positive latex agglutination test for D-dimer (LSI Medience, Tokyo, Japan), 2) already scheduled for diagnostic contrast-enhanced CT, 3) age >20 years old, and 4) provision of written informed consent to participate in the study. Exclusion criteria were renal dysfunction (glomerular filtration rate <30 mL/min/1.73 m²) and circulatory instability. Once enrolled, each of the 10 patients underwent fluoroscopy for the creation of a Radwisp image, and immediately thereafter, each underwent diagnostic contrast-enhanced CT. As a validation test, Radwisp images of all 10 patients were reviewed by 50 physicians (25 cardiologists and 25 residents) on a subsequent day to diagnose the presence of APE based solely on the Radwisp images. Residents had graduated from an accredited medical school, were licensed to practice medicine, and had been practicing medicine for less than two years. All of the physicians were blinded to how many of the 10 patients had APE.

The study was approved by the Nihon University Itabashi Hospital Clinical Research Judging Committee (RK-211214-10) and was carried out in accordance with the ethical standards of the institutional research committee and the 1964 Declaration of Helsinki. The study was registered with the UMIN Clinical Trials Registry (UMIN000052843).

The diagnosis of APE

APE was diagnosed by a specialist physician in the coronary care unit. The diagnosis was based on the clinical presentation and findings of contrast-enhanced CT, which is considered the diagnostic gold standard. The contrast-enhanced CT images were interpreted by radiologists. The pre-test probability of APE was estimated based on the Wells scoring system (7). A Simplified Pulmonary Embolism Severity Index (sPESI) assessment was also performed (8). PEs were categorized as nonmassive, submassive, or massive and by whether they were accompanied by cardiac arrest or circulatory collapse (9).

Radwisp workstation

The Radwisp workstation, a chest X-ray fluoroscopic video analysis system (hereinafter referred to as “Radwisp"), enables a computerized analysis of video output from fluoroscopic images. Using this system, the airflow and blood flow in each part of the lung can be assessed qualitatively.

For the analysis of each Radwisp image created in our study, the region of interest, called the lung field, was first defined. The right lung field was defined by 5 Bézier curves drawn between 10 control points (Fig. 1a), and the left lung field was defined by 6 Bézier curves drawn between 12 control points (Fig. 1b). For the analysis, it is necessary to calculate changes in the brightness values of small regions, called blocks, in the lung fields while maintaining their relative positions. Because the lung field expands and contracts with respiration, it is defined in all frames and then subdivided into squares of approximately 4×4 pixels each, so that the squares are comparable (Fig. 1a, b) For each of the 10 cases, the mean brightness value in each block was calculated and then differentiated. Because the differentiated brightness value includes the periodic change associated with respiration and blood flow, a fast Fourier transform (FFT) and inverse fast Fourier transform (iFFT) were used to approximate the periodic change associated with blood flow (Fig. 2).

Figure 1.

Lung fields defined by Bézier curves drawn between control points. For the analysis, the right lung field (a) was defined by 5 Bézier curves drawn between 10 control points (labeled P1-10), and the left lung field (b) was defined by 6 Bézier curves drawn between 12 control points (labeled P1-12).

Figure 2.

The process of conversion from a plain film chest radiograph to a Radwisp image. (a) A block is used for the analysis. (b, c) A fast Fourier transform is performed on the differentiated value of the mean brightness, and (d) an inverse fast Fourier transform is performed using only the major frequency component corresponding to the heartbeat frequency to approximate the periodic change associated with blood flow. (e) An image is generated for the analysis by projecting the normalized values onto small regions, called blocks.

Assuming that the periodic change associated with blood flow synchronizes with the heartbeat, we first identified the major heartbeat frequency component as follows: the change in the brightness value of the outer edge of the heart was calculated, and a Fourier transform was performed. From among the frequency components in the range of 0.9-1.7 Hz, the highest was taken as the major heartbeat frequency component (Fig. 3). An FFT was performed on the differentiated value of the mean brightness for each block (Fig. 2b, c), and an iFFT was performed using only the major frequency component corresponding to the heartbeat frequency (Fig. 2d). Thus, the changes associated with the blood flow in each block were extracted. Once this had been performed for all blocks, normalization was performed.

Figure 3.

Calculation of heartbeat frequency from the outer edge of the heart. (a) The change in the brightness value of the outer edge of the heart was calculated, and (b) Fourier transform was performed. Of the heartbeat frequency components in the range of 0.9 to 1.7 Hz, the highest was defined as the major frequency component.

To reduce noise, the 95% confidence interval (CI) was captured after iFFT, and the maximum value within the 95% CI was normalized to 100, while the minimum value was normalized to -100. An image used for the analysis was generated by projecting the normalized value onto each square (Fig. 2e). The obtained image included the values taken from -100 to 100 for each pixel, and a pseudo-color was assigned by window-level conversion and then used for the evaluation. Therefore, it was possible to distinguish between the variability in respiration and blood flow occurring during spontaneous inspiration.

Validation testing

The validation testing included a total of 50 participating physicians: 25 cardiologists who were proficient in the diagnosis and treatment of APE and 25 residents without such proficiency. Cardiologists are more skilled than residents in diagnosing APE based on clinical findings. However, we hypothesized that there would be no marked difference between the two groups of physicians in diagnosing APE based on Radwisp images alone. Before the test, each of the 50 physicians took part in a 15-minute instruction session in which they were familiarized with the characteristics of Radwisp images obtained from both patients with and without APE (20 images in total). Thereafter, on the same day, all participating physicians, blinded to the patients' clinical information and CT images, were asked individually to view the Radwisp images obtained for each of the 10 study patients and to diagnose APE as either present or absent on each image.

For validation, we calculated the sensitivity, specificity, positive predictive value, negative predictive value, and diagnostic accuracy of the 25 participating cardiologists and 25 participating residents. For further validation, they were also asked individually to indicate the number representing their continuous rating scale (CRS) for a diagnosis of APE: 100% (definitely APE), 50% (probably APE), <50% (probably not APE), or 0% (definitely not APE) (10,11). We then averaged the CRSs of seven APE and three non-APE images for each physician and calculated the mean CRS in the cardiologist and resident groups.

Sample size determination and statistical analyses

A sample of 10 cases was selected for this pilot study. Because of the small sample size, we chose to include 50 physicians (25 cardiologists and 25 residents) to obtain meaningful data. The patients' clinical characteristics and study results are shown as the mean±standard deviation (SD), median (25th and 75th percentile), or percentage (ratio). Differences (between patients with and without APE or between cardiologists and residents) in continuous variables were analyzed using the t-test, and differences in categorical variables were analyzed using the chi-square test. Receiver operating characteristic (ROC) curves were drawn for the CRSs felt by the cardiologists and by the residents regarding their Radwisp-based diagnosis of APE, and the areas under the curve (AUCs) were calculated and compared between the physician groups.

All analyses were performed using the SPSS software program, version 19.0 (SPSS, Chicago, USA); p<0.05 was considered statistically significant.

Results

Of the 10 cases of suspected APE, 7 were confirmed based on contrast-enhanced CT and clinical findings to be cases of APE. APE was ruled out in the remaining three cases, and the final diagnoses in those individuals were heart failure, non-ST-elevation myocardial infarction with heart failure, and chronic obstructive pulmonary disease (one each). There were no marked differences between the APE and non-APE groups in terms of the age, sex, body mass index, or most laboratory values. However, the D-dimer concentration on admission and the Wells score were significantly higher in the APE group than in the non-APE group: 11.6 (9.3, 25.2) μg/mL vs. 1.2 (1.0, 4.0) μg/mL (p=0.04) and 7.5 (3.0, 7.5) vs. 1 (0, 3) (p=0.01), respectively (Table 1).

Table 1.

Patient Characteristics, Per Study Group.

		APE (n=7)		No APE (n=3)		p value
Male sex (%)		3 (43)		1 (33)		0.49
Age (years)		66±14		72±23		0.63
BMI (kg/m2)		27±3		22±8		0.25
Hemodynamic variables
Systolic blood pressure (mmHg)		114±16		124±27		0.51
Respiratory rate (/min)		17±4		30±21		0.41
Heart rate (/min)		92±14		72±15		0.16
SpO2 (%)		96±2		96±2		1.00
Laboratory values
D-dimer (μg/mL)		11.6 (9.3, 25.2)		1.2 (1.0, 4.0)		0.04
SF (μg/mL)		38.7 (5.0, 70.0)		3.0 (3.0, 13.7)		0.05
Troponin I (pg/mL)		0.013 (0.009, 0.05)		0.093 (0.03, 0.47)		0.39
APE risk
Wells score		7.5 (3.0, 7.5)		1 (0, 3)		0.01
Geneva score		6.0 (5.0, 7.0)		4.0 (3.0, 4.0)		0.05
APE severity
Arrest or collapse		0		-
Massive		0		-
Submassive		7 (100%)		-
Nonmassive		0		-
PESI score		69 (60, 108)		126 (50, 156)		0.29

Values are as mean±SD values, median and interquartile range, or as the numbers and/or percentages. APE: acute pulmonary embolism, BMI: body mass index, PESI: Pulmonary Embolism Severity Index, SF: soluble fibrin

Accuracy of the Radwisp-based diagnosis for APE

Radwisp images representative of APE and non-APE cases are shown in Fig. 4, and a contrast-enhanced CT image is shown alongside a corresponding representative Radwisp image of APE in Fig. 5. The contrast-enhanced CT image shows a thrombus in the right pulmonary artery, which corresponds to the impaired blood flow in the circulation area seen on the Radwisp image. Without knowledge of the definitive diagnosis for any of the 7 APE cases or any of the 3 non-APE cases, 25 cardiologists and 25 residents were given 10 Radwisp images. Each group (cardiologists and residents) provided a total of 250 diagnoses, 175 covering APE cases (with 25 cardiologists and 25 residents having evaluated 7 cases each) and 75 covering non-APE cases (with 25 cardiologists and 25 residents having evaluated 3 cases each). Among cardiologists, the sensitivity and specificity of Radwisp-based analysis were 91% (159/175) and 48% (36/75), respectively; the positive and negative predictive values were 80% (159/198) and 69% (36/52), respectively; and the diagnostic accuracy was 78% (195/250). Among residents, the sensitivity and specificity were 88% (154/175) and 35% (26/75), respectively; the positive and negative predictive values were 76% (154/203) and 55% (26/47), respectively; and the diagnostic accuracy was 72% (180/250) (Table 2).

Figure 4.

Representative Radwisp images of acute pulmonary embolism and absence of acute pulmonary embolism and their interpretation. Areas without obstruction of the pulmonary vessels appeared white, and areas of reduced pulmonary blood flow appeared red. Usually, pulmonary blood flow is reduced distally; therefore, the peripheral area often remains red. However, if no white is seen centrally and only red is seen, the findings suggest pulmonary embolism. (A) In this representative case of acute pulmonary embolism, perfusion of the right lung appeared markedly impaired, with the zone of impairment (arrow) being quite extensive. Notably, the zone has no “white,” meaning there is no area without obstruction. (B) In this representative case of absence of acute pulmonary embolism, the white area, indicative of good pulmonary blood flow, extends from the center of both the right and left lungs.

Figure 5.

Radwisp and contrast-enhanced CT images in a case of acute pulmonary embolism. This case is the same as that shown in Fig. 4. (A) The Radwisp image shows a large area of impaired pulmonary blood flow in the right lung (arrow). (B) Contrast-enhanced CT showing a thrombus in the same area (arrow).

Table 2.

Testing Accuracy at Which Cardiologists and Residents Correctly Diagnosed APE and Non-APE Cases by Assessing Radwisp Images.

		Cardiologists		Residents
Sensitivity		91% (159/175)		88% (154/175)
Specificity		48% (36/75)		35% (26/75)
Positive predictive value		80% (159/198)		76% (154/203)
Negative predictive value		69% (36/52)		55% (26/47)
Diagnostic accuracy		78% (195/250)		72% (180/250)

APE: acute pulmonary embolism

CRSs regarding the Radwisp-based diagnosis

The mean CRSs of cardiologists and residents in their Radwisp-based diagnosis of APE were 86±13% and 86±13%, respectively. By evaluating the Radwisp images, many of the participating physicians were strongly convinced of the presence of APE in cases that had been definitively diagnosed as APE. The mean CRSs of cardiologists and residents for diagnosing the absence of APE were 65±23% and 69±19%, respectively, indicating that many of the participating physicians were unable to confidently diagnose the absence of APE from the Radwisp images (Table 3). Overall, the AUC for cardiologists' CRSs was 0.72 (95% CI: 0.63-0.80), and that for residents' CRSs was 0.65 (95% CI: 0.57-0.73), showing no significant difference between the groups (p=0.41) (Fig. 6).

Table 3.

Continuous Rating Scale of Participating Physicians.

Cases		Cardiologists (n=25)		Residents (n=25)		p value
APE (n=7)		86±13%		86±13%		0.96
Non-APE (n=3)		65±23%		69±19%		0.48
Total numbers of correct answers with the “definitely APE”		72% (126/175)		66% (116/175)		0.47
Total numbers of correct answers with the “definitely not APE”		12% (9/75)		11% (8/75)		0.81

APE: acute pulmonary embolism

Continuous rating scale indicates a diagnosis of APE, whether 100% (definitely APE), 50% (probably APE), less than 50% (probably not APE), or 0% (definitely not APE).

Figure 6.

Average receiver operating characteristic curves for cardiologists’ and residents’ continuous rating scale (CRS) regarding the diagnosis of acute pulmonary embolism or absence of acute pulmonary embolism. The area under the curve (AUC) for all cardiologists was 0.72 (95% CI: 0.63-0.80), and that for all residents was 0.65 (95% CI: 0.57-0.73).

Each group of 25 participants provided 175 diagnoses covering the 7 APE cases. There was no significant difference between the groups in terms of the number of correct “Definitely APE” diagnoses [cardiologists vs. residents: 72% (126/175) vs. 66% (116/175); p=0.47] and “Definitely not APE” diagnoses [cardiologists vs. residents: 12% (9/175) vs. 11% (8/75); p=0.81] (Table 3).

Discussion

To our knowledge, this is the first study to validate the use of Radwisp for the objective diagnosis of APE or its absence. As noted above, the positive predictive value of a Radwisp-based analysis for a diagnosis was 80% when performed by cardiologists and 76% when performed by residents. The CRSs for the diagnosis of APE were 86±13% and 86±13%, respectively. A consistently high CRS was thus achieved regardless of whether patients with suspected APE were assessed by cardiologists or residents.

Our data partially support our notion that APE can be distinguished from other conditions and normal pulmonary circulation on Radwisp images. The PIOPED II study showed that pulmonary blood flow scintigraphy had a sensitivity of 77.4% and specificity of 97.7% for the diagnosis of APE (12). Our study showed that the positive and negative predictive values of Radwisp as a diagnostic test for APE were approximately 80% and 40%, respectively. The sensitivity of a Radwisp-based analysis for identifying APE was high (around 90%), whereas specificity was low (below 50%).

We believe that several factors can lead to a misdiagnosis. A false-negative diagnosis can potentially occur when the overall bilateral blood flow is compromised, as a Radwisp analysis detects a loss of pulmonary circulation based on the pulmonary blood flow gradient. The specificity of a Radwisp-based analysis for APE was low, and the physicians' CRS was 65±23% for cardiologists and 69±19% for residents, respectively, indicating a tendency toward a diagnosis of “probably APE” despite the Radwisp analysis being performed in non-APE cases. In addition, the total number of correct answers for “definitely not APE" was very low. These findings suggest a low diagnostic performance in ruling out APE cases from non-APE cases.

One major factor contributing to false-positive diagnoses is the presence of an enlarged heart, where the Radwisp image of the enlarged left ventricle on the chest radiograph may mimic the loss of pulmonary blood flow, even when the flow is normal. As noted above, the non-APE cases in this study turned out to be cases of heart failure, non-ST elevation myocardial infarction with heart failure, and chronic obstructive pulmonary disease. Some of these three cases might have had a small, latent effect on pulmonary circulation, possibly manifesting as a not-so-readily discernible abnormality on the image, leading to false-positive diagnoses. Nonetheless, the Japanese Guidelines for Diagnosis, Treatment, and Prevention of Pulmonary Thromboembolism and Deep Vein Thrombosis make a class I recommendation that APE be ruled out without performing any imaging modality if the D-dimer value is negative in cases with a low or moderate pre-test clinical probability, such as based on the Wells score. In the present study, the non-APE group had a low pre-test clinical probability and no elevated D-dimer levels. Therefore, Radwisp imaging may remain valuable for reinforcing the APE diagnosis when performed in patients with a high pre-test clinical probability (13). Nevertheless, continuous efforts are needed to develop new methods, including the integration of artificial intelligence and scoring systems, in addition to the future accumulation and analyses of Radwisp images for APE, heart failure, and pneumonia, with the aim of further improving diagnostic accuracy.

The use of this novel image analysis system has several advantages over contrast-enhanced chest CT for the timely diagnosis of APE. First, the Radwisp image is a video image captured during natural respiration, i.e., during inspiration and expiration. Usually, patients with APE experience severe respiratory distress and are often tachypneic. Therefore, it can be difficult for patients to hold their breath during CT. Unlike pulmonary scintigraphy for the assessment of blood flow in the lungs, the fluoroscopic imaging required for a Radwisp analysis takes only a few dozen seconds; patients can breathe naturally. Furthermore, the analysis itself does not require much time either. Another important advantage is that a Radwisp image can be obtained without using a contrast medium. Contrast-enhanced CT is the gold standard for the diagnosis of APE and is recommended if APE is suspected (14). However, renal dysfunction is often encountered in patients with APE, with approximately 50% of these patients having at least a moderately impaired renal function on admission (15). Abrupt renal insufficiency, i.e., acute kidney injury, especially in patients with APE, leads to premature death (16). For patients in whom contrast-enhanced CT is contraindicated, non-contrast-enhanced CT can be used to diagnose centrally localized APE (17,18). Of note, overall, the diagnostic performance of non-contrast-enhanced CT in cases of suspected APE is only moderate at best. However, a Radwisp-based analysis can be performed without risking volume overload, even in patients with severe circulatory instability. Dynamic chest radiography, a similar modality, has a very high diagnostic performance and can be used to diagnose chronic thromboembolic pulmonary hypertension (19). The important difference between dynamic chest radiography and Radwisp imaging is that the latter allows for spontaneous breathing; thus, it may be considered preferable for patients with severe dyspnea.

Study limitations

Several limitations associated with the present study warrant mention. First, it had a small number of included cases. It was conducted according to the Clinical Trials Act in Japan, which meant that we were allowed only a little over one year to enroll patients and complete the analysis. In particular, there were only 3 non-APE cases among the 10 validation cases, which might have had a significant impact on our results. This should be considered when interpreting the findings of this study. For example, the relatively high diagnostic accuracy and positive predictive value of >70% might have been strongly influenced by the much larger number of APE than non-APE cases in this study. Second, the preparatory session, while very short (15 min), might also be considered a study limitation because this rendered the participating physicians not naïve to the objective. However, we felt that it was important to expose the participants to both normal Radwisp images and APE Radwisp images immediately before the validation test. Finally, all physicians were asked to diagnose patients' conditions solely based on the Radwisp images. The addition of a complementary scoring system may increase the accuracy of the APE diagnosis.

Conclusion

In this first validation study of Radwisp for APE, the sensitivity was high, while the specificity was not as high. According to the CRS analysis, its diagnostic performance was effective in ruling out non-APE cases from APE cases, but it was less effective in extracting APE cases from non-APE cases. Therefore, this new imaging modality can potentially serve as a complementary diagnostic tool that can reveal the distribution of pulmonary circulation due to thrombi in patients highly suspected of having APE. It can be imaged without the use of contrast medium and without breath-holding, meaning that it can be employed even in patients with severe dyspnea or renal dysfunction. Further studies with a larger number of APE and non-APE cases are needed to evaluate the diagnostic performance.

The authors state that they have no Conflict of Interest (COI).