Artificial Intelligence Tool for Detection and Worklist Prioritization Reduces Time to Diagnosis of Incidental Pulmonary Embolism at CT

Abstract

Purpose

To evaluate the diagnostic efficacy of artificial intelligence (AI) software in detecting incidental pulmonary embolism (IPE) at CT and shorten the time to diagnosis with use of radiologist reading worklist prioritization.

Materials and Methods

In this study with historical controls and prospective evaluation, regulatory-cleared AI software was evaluated to prioritize IPE on routine chest CT scans with intravenous contrast agent in adult oncology patients. Diagnostic accuracy metrics were calculated, and temporal end points, including detection and notification times (DNTs), were assessed during three time periods (April 2019 to September 2020): routine workflow without AI, human triage without AI, and worklist prioritization with AI.

Results

In total, 11 736 CT scans in 6447 oncology patients (mean age, 63 years ± 12 [SD]; 3367 men) were included. Prevalence of IPE was 1.3% (51 of 3837 scans), 1.4% (54 of 3920 scans), and 1.0% (38 of 3979 scans) for the respective time periods. The AI software detected 131 true-positive, 12 false-negative, 31 false-positive, and 11 559 true-negative results, achieving 91.6% sensitivity, 99.7% specificity, 99.9% negative predictive value, and 80.9% positive predictive value. During prospective evaluation, AI-based worklist prioritization reduced the median DNT for IPE-positive examinations to 87 minutes (vs routine workflow of 7714 minutes and human triage of 4973 minutes). Radiologists’ missed rate of IPE was significantly reduced from 44.8% (47 of 105 scans) without AI to 2.6% (one of 38 scans) when assisted by the AI tool (P < .001).

Conclusion

AI-assisted workflow prioritization of IPE on routine CT scans in oncology patients showed high diagnostic accuracy and significantly shortened the time to diagnosis in a setting with a backlog of examinations.

Keywords: CT, Computer Applications, Detection, Diagnosis, Embolism, Thorax, Thrombosis

Supplemental material is available for this article.

© RSNA, 2023

See also the commentary by Elicker in this issue.

Summary

Artificial intelligence–based prioritization of radiologist reading worklists significantly reduced the time to diagnosis of incidental pulmonary embolism at CT in patients with cancer.

Key Points

• ■ Artificial intelligence (AI) software for detecting incidental pulmonary embolism (IPE) at chest CT in patients with cancer showed high diagnostic accuracy in a large sample of 11 736 scans (sensitivity, 91.6%; specificity, 99.7%; negative predictive value, 99.9%).

• ■ In a practice with a backlog of unreported examinations, AI-based worklist prioritization reduced the median detection and notification time of IPE in flagged scans from several days to 1.0 hour.

• ■ The missed rate of IPE was significantly reduced from 44.8% to 2.6% when radiologists were assisted by the AI tool (P < .001).

Introduction

Incidental pulmonary embolism (IPE) is a common comorbidity in oncology patients and is related to a high risk of recurrent venous thromboembolism (1). IPEs can be detected at routine contrast-enhanced CT of the chest that is performed for indications other than thromboembolic disease. The pooled frequency of IPE in oncology patients has recently been reported as 3.4% (2). IPEs are associated with poor outcomes, including disease progression and reduced overall survival (3,4). While an IPE is clinically unsuspected, it can be an urgent and life-threatening finding, depending on the patient’s clinical condition and the extent of the embolism (4). Timely diagnosis and proper management of IPEs are therefore essential.

The workload of radiologists has increased substantially during the past decades, mainly the result of higher volumes and complexity of imaging examinations (5,6). It is therefore not unusual for radiology departments to be confronted with a backlog of unreported examinations, especially when unexpected surges in imaging requests occur and in regions where a shortage of radiologists exists. Consequently, long report turnaround times (TATs) pose a risk for delayed diagnosis of unsuspected critical findings. Artificial intelligence (AI) applications that can automatically detect critical findings at imaging can be used to prioritize scans in the reading worklist of the radiologist, with the aim of shortening time to diagnosis and communication with the treating physician (7). AI-based prioritization tools have been studied for use cases such as intracranial hemorrhage at CT, acute pathologic abnormalities on chest radiographs, and pulmonary embolism (PE) on dedicated CT pulmonary angiograms (CTPAs), with varying results (8–11).

To the best of our knowledge, the clinical benefit of a worklist triage tool for IPE has not been investigated. Detection of IPEs can be challenging because filling defects caused by thrombus need to be identified on CT scans acquired in the venous phase with suboptimal contrast enhancement of the pulmonary arteries. The relative frequency of CT scans that are positive for IPE is also considerably lower when compared with the relative frequency of positive CTPAs. Therefore, a clinically useful AI tool must perform with high sensitivity and an acceptable false-positive rate.

The purpose of this study was to evaluate the diagnostic efficacy of AI software for the detection and prioritization of IPE at chest CT in terms of diagnostic accuracy and its effect on time to diagnosis in a real-world clinical environment.

Materials and Methods

Study Design and Patients

We performed a study to evaluate the diagnostic efficacy of AI software to analyze IPE on chest CT scans in clinical practice by using data from historical controls and prospectively analyzed data. The institutional review board approved this study and waived the requirement for study-specific informed consent. The AI software was provided by Aidoc Medical. The authors are not affiliated with the industry and had control of the data and information submitted.

The study was conducted in the radiology department of a comprehensive cancer center during three periods of 15 weeks each (Fig 1). All adult (≥18 years) patients with cancer who underwent CT of the chest with intravenous contrast agent were included. In total, 11 736 chest CT scans were assessed across the three periods. During the first period (April to July 2019), radiologists received no AI assistance and no special instructions while reporting. During the second period (November 2019 to February 2020), radiologists received no AI assistance but were instructed to screen all newly acquired CT scans for the presence of IPE. The task of human triage was performed by staff radiologists during working hours. Examinations with IPE were immediately reported, and examinations without acute findings were marked and returned to the reading worklist. During the third period (June to September 2020), radiologists were assisted with an AI-based worklist prioritization tool. The 15-week intervals between periods allowed radiologists to get accustomed to the new instructions and implementation of the AI software.

Figure 1:

Flowchart shows study design and data selection per time period. AI = artificial intelligence, CTPA = CT pulmonary angiogram, TP1 = time period 1, TP2 = time period 2, TP3 = time period 3.

Imaging Data

Eligible chest CT scans acquired with intravenous contrast agent were selected using the study description. CT scans were acquired in the venous phase, 60–70 seconds after contrast medium administration. Dedicated CTPAs were not selected and were excluded (n = 3) after data collection. Sixty-four scans were excluded because of incomplete or missing image files.

AI Software

Aidoc Medical provided commercially available AI-based image analysis software that has been cleared by Conformité Européenne and the U.S. Food and Drug Administration. The intended use of the software was specifically the prioritization of IPE. The tool was connected with the picture archiving and communication system to immediately analyze eligible imaging studies after acquisition. After processing, a positive or negative result was transferred to the radiologist’s reading worklist in the radiology information system. Positive examinations were marked with a bright color in the worklist. Additionally, a heatmap highlighting the suspected abnormality was sent to the picture archiving and communication system.

Examinations that were not analyzed by the AI software due to failed data orchestration or technical validation were excluded (n = 378). Failure of AI analysis was possibly related to image quality (eg, insufficient contrast enhancement, incompatible acquisition parameters).

Reference Standard

The radiology report was used to categorize all CT scans for the presence of IPE. Subsequently, two board-certified radiologists (A.N., general radiologist with 3 years of experience; and L.T., subspecialized in oncologic imaging with 5 years of experience) independently reviewed CT scans that were categorized as positive by the radiology report or AI software. The readers reviewed only imaging data, including heatmap results, without any clinical information to label CT scans as positive, negative, or inconclusive. In cases of disagreement between readers, consensus was reached by a third dedicated chest radiologist (A.B.R., with 8 years of experience). CT scans that were categorized as negative by both the radiology report and AI software were not additionally reviewed for the presence or absence of IPE.

The location of the most proximal embolism was identified for each positive CT scan. False-positive detections were classified as technical artifact, flow artifact, or abnormalities adjacent to or within a pulmonary artery.

Diagnostic Accuracy

The performance of the AI software in the detection of IPE was evaluated on both prospectively analyzed and historically collected data to increase the sample size. Sensitivity, specificity, negative predictive value, positive predictive value, and accuracy were calculated.

Time-related End Points

Timestamps were registered electronically to calculate the following time intervals: (a) time to process, defined as the time interval between forwarding of the study to the AI software and availability of the AI analysis result; (b) detection and notification time (DNT), defined as the time interval between availability of the study in the radiologist’s reading worklist and the opening of the study by the radiologist; (c) TAT, defined as the time interval between availability of the study in the worklist and finalization of the report; and (d) time to read (TTR), defined as the time interval between opening of the study and finalization of the report. To avoid changes in reporting behavior, radiologists were not informed about the registration of time-related end points.

Statistical Analysis

The prevalence of IPE in different time periods was compared using a χ² test. CIs for diagnostic accuracy metrics, such as sensitivity and specificity, were computed using a binomial distribution. A Fisher exact test was used to investigate the reduction in missed IPEs by radiologists with AI versus without AI assistance.

Per time variable (DNT, TAT, and TTR), we were interested in comparing values between IPE-positive and -negative CT scans for every time period. A Student t test was used to compare each given log-transformed time variable between groups of positive and negative CT scans. CIs were determined for these differences between positive and negative CT scans, per time period. Subsequently, we compared CIs among the three time periods. Given that different time periods involved different and independent examinations, nonoverlapping CIs led to the conclusion that differences were statistically distinct. In all cases, the null hypothesis was rejected for P < .05. Statistical analyses were performed by a statistician (R.M.) using the R environment for statistical computing (version 3.6.3; https://www.r-project.org) (12).

Results

Patient Demographics

A total of 11 736 chest CT scans from 6447 unique patients (mean age, 63 years ± 12 [SD]; 3367 men [52.2%]) were included. Patient demographics per time period are shown in Table 1. Most imaging studies (11 333 of 11 736 [96.6%]) were performed in outpatients.

Table 1:

Patient Demographics per Time Period

Imaging Data

Chest CT scans were acquired with seven models from three manufacturers: Toshiba (5836 scans of 11 736 [49.7%]), Siemens (5140 of 11 736 [43.8%]), and Philips (760 of 11 736 [6.5%]) (Table S1). CT series reconstructed with a soft-tissue kernel were obtained with a section thickness of 1.0–2.0 mm. In the majority of CT scans (9923 of 11 736 [84.6%]), the chest was imaged in combination with the abdomen and/or neck.

Diagnostic Accuracy

The prevalence of IPE in the sample was 1.2% (143 of 11 736 scans). The prevalence of IPE was similar across the three time periods (χ² test, P = .17). Overall, the AI software returned 131 true-positive, 12 false-negative, and 31 false-positive results (Table 2). Three were inconclusive, and the other 11 559 examinations were categorized as negative by both the radiology report and AI software. The AI software showed a sensitivity of 91.6% (131 of 143 scans; 95% CI: 86.7, 95.8), specificity of 99.7% (11 559 of 11 590 scans; 95% CI: 99.6, 99.8), negative predictive value of 99.9% (11 559 of 11 571 scans; 95% CI: 99.8, 99.9), positive predictive value of 80.9% (131 of 162 scans; 95% CI: 74.7, 86.4), and accuracy of 99.6% (11 690 of 11 733 scans; 95% CI: 99.5, 99.7). Three examinations were marked as inconclusive by the review panel. The diagnostic accuracy per time period is available in Tables S2–S5.

Table 2:

Diagnostic Accuracy in Detection of IPE by the AI Software Alone

Table 3 lists the location of the most proximal filling defect in pulmonary arteries of IPE-positive examinations. In total, 54 of 143 clots (37.8%) were located in the main or lobar pulmonary arteries. Fifty-two of these (96.3%) were detected by the AI software, including all main pulmonary clots (n = 9). Two examples of true-positive detections are shown in Figure 2.

Table 3:

Location of the Most Proximal Filling Defect in Pulmonary Arteries on Positive Incidental Pulmonary Embolism Scans

Figure 2:

True-positive detection of incidental pulmonary embolism (PE) by the artificial intelligence (AI) software. (A, B) Images in a 68-year-old woman who underwent routine CT with intravenous contrast agent for outpatient follow-up of melanoma. (A) Axial CT image shows a large filling defect straddling the bifurcation of the pulmonary trunk (arrow) and extending into both pulmonary arteries, compatible with an incidental saddle PE. (B) Corresponding AI heatmap highlights the detected abnormality (red), thereby prioritizing the case in the radiologists’ worklist. (C, D) Images in a 58-year-old woman with a history of rectal cancer undergoing outpatient follow-up. (C) Axial restaging CT image with intravenous contrast agent shows a small incidental subsegmental PE in the right lower lung lobe (arrow). (D) Corresponding AI heatmap enables the radiologist to localize the finding (red).

False-negative findings were located in the segmental or subsegmental arteries in 10 of 12 examinations (83.3%). The other two false-negative findings (16.7%) were located in lobar arteries; however, the clots were small and most likely chronic (Fig 3).

Figure 3:

False-negative findings of two chronic lobar incidental pulmonary embolisms (PEs) that were not detected by the artificial intelligence (AI) software. (A) Axial CT image with intravenous contrast agent in a 70-year-old man (an outpatient) with urothelial carcinoma shows a small incidental PE (IPE) located against the vessel wall in the right pulmonary artery bifurcation (arrow), compatible with a small chronic IPE. (B) Contrast-enhanced coronal CT image in a 62-year-old man (an inpatient) with lung cancer shows a small eccentric filling defect in the pulmonary artery of the left lower lobe (arrow). These findings were not detected by the AI software.

False-positive findings (n = 31) were consistent with known mimickers of PE (Table 4). Most false-positive findings were categorized as flow artifacts (13 of 31 [41.9%]) (Fig 4A, 4B). Technical artifacts (seven of 31 findings [22.6%]) were predominantly caused by respiratory motion. In 11 of 31 examinations (35.5%), extravascular abnormalities (eg, lymphadenopathy [Fig 4C, 4D]) or intravascular abnormalities (eg, stump thrombus) led to false-positive detections. The number of false-negative and false-positive findings per time period is available in Tables S2–S4.

Table 4:

Causes of False-Positive Detections by the Artificial Intelligence Software

Figure 4:

False-positive detections by the artificial intelligence (AI) software. (A, B) Images in a 59-year-old woman (an outpatient) with melanoma who underwent CT with intravenous contrast agent. (A) Axial CT image shows slightly decreased contrast opacification in a segmental pulmonary artery in the right lower lobe (arrow). This finding was compatible with a flow artifact without any clinical significance. (B) Corresponding AI software heatmap misclassified the finding as a possible incidental pulmonary embolism (IPE), as highlighted in red. (C, D) Images in a 36-year-old woman (an outpatient) with cervical cancer who underwent CT with intravenous contrast agent. (C) Axial CT image shows hilar and mediastinal lymphadenopathy. An enlarged right hilar lymph node shows impression on the right pulmonary arteries (arrow). The finding was misclassified by the AI software as IPE, as shown on (D) the corresponding axial heatmap in red.

In 44.8% (47 of 105) of IPE-positive CT scans from the first two time periods (historically collected data), the finding was missed in the radiology report but correctly identified by the AI software. In comparison, in the third time period when radiologists were assisted by the AI software, only one of 38 studies positive for IPE (2.6%) was missed by the radiologists. This resulted in a 94% reduction (44.8% vs 2.6%) in missed IPE with use of AI assistance (Fisher exact test, P < .001). Categorization of IPEs missed by the radiologists according to the location of the most proximal clot was as follows: lobar in eight of 48 examinations (16.7%), segmental in 28 of 48 examinations (58.3%), and subsegmental in 12 of 48 examinations (25%). No main IPE was missed by the radiologists.

The study design excluded CT scans that were not successfully analyzed by the AI software due to failed data orchestration or technical validation. The scans from the first two time periods in which the AI analysis failed (63 of 7820 [0.8%]) were all reported as negative for IPE. However, during the third time period, the AI software was deployed in the clinical environment, and scans were analyzed in real time. As a result, a larger proportion of CT scans (315 of 4294 [7.3%]) was not processed and thus excluded. Nevertheless, two of the 315 excluded CT scans were reported as positive for IPE, with a segmental and subsegmental location, respectively.

Time-related End Points

The AI analysis of CT scans was performed without human intervention. The median processing time of CT scans by the AI software was 3 minutes, with a maximum of 20 minutes. In 99.5% (3958 of 3979) of studies, the AI result was available to the radiologist at the time of opening the study.

The median DNTs for IPE-positive examinations in all patients were 7714, 4973, and 87 minutes for the respective time periods of routine workflow without AI, human triage without AI, and worklist prioritization with AI. The median DNTs for IPE-positive examinations in outpatients only were 8950, 5454, and 80 minutes for the respective time periods. When only considering true-positive examinations flagged by the AI software in the third time period (n = 34), the median DNT was 62 minutes. The DNT of 29 of the 34 true-positive examinations (85.3%) was less than 6 hours. In comparison, the DNT of the four nonprioritized false-negative examinations in the third time period ranged from 1280 minutes to 12 684 minutes. Figure 5 shows the DNTs of all IPE-negative versus -positive studies for each time period. CIs were calculated for the time differences between negative and positive studies in each time period (Fig 6). Given that the CIs did not overlap between the time period with AI assistance and those without AI assistance, differences were statistically significant between the time periods. In contrast, we found no evidence of a difference between routine workflow without AI and human triage.

Figure 5:

Box plot shows detection and notification times (DNTs) of incidental pulmonary embolism (IPE)–negative versus IPE-positive CT scans per time period: routine workflow without artificial intelligence (AI), human triage without AI, and worklist prioritization with AI. DNT was markedly reduced for positive CT scans during the third time period with AI assistance (median DNT, 87 minutes vs routine workflow DNT of 7714 minutes [5 days]). The horizontal line in each box plot indicates the median, and the box corresponds to the IQR. The whiskers indicate minimum and maximum values in the data. Circles represent outliers. TP1 = time period 1, TP2 = time period 2, TP3 = time period 3.

Figure 6:

Graph shows 95% CIs of detection and notification time (DNT) differences between incidental pulmonary embolism–positive and –negative CT scans per time period: routine workflow without artificial intelligence (AI) (TP1), human triage without AI (TP2), and worklist prioritization with AI (TP3). The DNT difference was largest for the third time period. Given that the CIs of the third time period versus the first and second periods did not overlap, differences were significant between these periods.

Report TAT showed similar results as DNT. The median TATs for IPE-positive examinations were 7772, 4983, and 148 minutes for the three respective time periods. When only considering true-positive examinations flagged by the AI software, the median TAT was 91 minutes.

The median TTR for all examinations was 16 minutes. The median TTR for positive examinations was 21 minutes. We found no evidence of a difference in TTR between time periods.

Discussion

We evaluated the clinical value of AI software for the analysis of IPE on a large sample of chest CT scans (n = 11 736) in oncology patients. The AI tool accurately detected IPE on chest CT scans with intravenous contrast agent, with a high sensitivity of 91.6% (131 of 143 scans), specificity of 99.7% (11 559 of 11 590 scans), and negative predictive value of 99.9% (11 559 of 11 571 scans). False-negative classification occurred in 12 of 143 examinations (8.4%) but was limited to segmental, subsegmental, and small chronic lobar clots. No IPE in the main pulmonary arteries was missed by the software. The number of false-positive detections by the software, 31 of 165 flagged examinations (18.8%), can be considered acceptable because radiologists could easily identify false-positive findings by using the heatmap. In total, only 0.3% of all analyzed examinations (31 of 11 736) were falsely positive. The impact of false-positive alerts on radiologist workflow was therefore limited.

During prospective evaluation, the AI software was deployed in a clinical environment with a backlog of unreported examinations to prioritize IPE on routinely acquired chest CT scans, mostly obtained in outpatients with known primary malignancy for posttreatment follow-up. The AI-based worklist prioritization resulted in a significantly reduced median DNT and TAT for flagged scans with IPE, from several days to 1.0 and 1.5 hours, respectively. In contrast, unassisted triage of CT scans by radiologists did not have a significant effect on the reduction of DNT or TAT when compared with the routine workflow. This is likely a result of the time-consuming nature of this task, contributing to low yield.

To the best of our knowledge, no other published study has investigated the diagnostic performance of AI software for the detection and prioritization of IPE. Previous studies have assessed the diagnostic accuracy of deep learning algorithms in the detection of PE on dedicated CTPAs (13–18). For this task, sensitivities and specificities ranged from 73% to 96% and 77% to 96%, respectively. Although our study focused on the detection of IPE on venous CT scans, which can be considered more challenging, our sensitivity and specificity were comparable with or higher than studies identifying PE on CTPAs. Furthermore, retrospective studies on diagnostic accuracy might not determine the real clinical impact on patient care (18). Schmuelling et al (19) evaluated the clinical implementation of AI software for prioritization of positive CTPAs in the emergency setting; the authors showed good diagnostic accuracy, with 79.6% sensitivity and 95.0% specificity. However, there was no significant reduction in report communication times. This is likely related to the overall short TATs of examinations in an emergency department. Tools to prioritize the reading worklist would provide the most benefit in clinical settings with a high workload and a backlog of unreported examinations, as in our situation. The shorter time to detection of IPE in our study has limited generalizability to practices with short report TATs.

PE is one of the diagnoses that is most commonly missed or delayed by physicians (20). Detection of IPE on routine contrast-enhanced chest CT scans can be especially challenging when the IPE is small and isolated. When analyzing historical CT data, we found that 44.8% of IPEs were missed by radiologists. Low detection rates of IPE by radiologists have also been reported in other studies (21,22). Wildman-Tobriner et al (23) applied a different AI algorithm to retrospectively analyze 11 913 CT examinations for undiagnosed IPE and found 49 missed IPEs (0.41%), leading to a missed rate of 38% (49 of 128 IPEs). In our study, we prospectively evaluated the effect of AI assistance on missed IPEs. The number of missed IPEs was reduced to one scan of 38 (2.6%), thereby demonstrating that AI software can assist radiologists to significantly improve the detection rate of IPE.

The clinical relevance and proper management of IPE remain a subject of debate. It is well known that venous thromboembolism in oncology patients is associated with high morbidity and mortality (24). Observational studies suggest that the prognosis of IPE is similar to that of symptomatic PE with regard to the risk of recurrence and mortality (25). Consequently, treatment guidelines for IPE are similar to those in symptomatic PE (26). Radiologic findings, such as thrombus load and central location, have been associated with adverse clinical outcomes in acute PE (27). These findings can also help determine IPE severity (4). In our study, 37.8% (54 of 143) of IPE-positive scans showed emboli in the main or lobar pulmonary arteries. Therefore, the benefit of AI-based worklist prioritization for timely assessment and treatment is most evident in these patients. The majority of physicians also treat smaller, more distal incidental emboli in patients with cancer (28). In our study, most IPEs that were missed by radiologists but detected by the AI software were segmental (28 of 48 [58.3%]) or subsegmental (12 of 48 [25%]). We must therefore consider the risk of overdiagnosis, specifically of isolated subsegmental IPE, which if left untreated would cause no more harm than treatment complications (29). To our knowledge, no randomized controlled trials have assessed the effectiveness of anticoagulation therapy in patients with subsegmental PE (30). However, recent studies support the use of anticoagulation therapy for subsegmental PE in oncology patients (1,31). Further studies are needed to assess the relevance of diagnosing and treating small incidental emboli.

The intended use of the investigated AI software is limited to workflow triage and not diagnostics (32). Automated worklist prioritization can assist radiologists, who remain responsible for verifying flagged examinations and must be aware of possible false-negative findings. Our study showed that a considerable number of scans (315 of 4294 [7.3%]) were not analyzed by the deployed AI software due to issues with data retrieval and technical validation, resulting in delayed diagnosis of IPE in these patients. This demonstrates the importance of monitoring and improving the yield of scans analyzed by AI software after deployment.

The study had limitations. First, the number of false-negative IPEs may be underestimated because examinations that were categorized as negative by both the radiology report and AI software were not reviewed—this was not feasible due to the high number of included scans. Second, the study sample consisted solely of oncology patients. The frequency of IPE is likely lower in the general patient population, which might impact the clinical relevance of the AI software. Third, the study design was not randomized. Although patient volumes in all time periods were similar and unaffected by the COVID-19 pandemic, TATs in the radiology department can be affected by many factors, such as staffing levels. To account for variations between periods, we calculated the time differences between positive and negative IPE examinations within each period separately and compared CIs of the difference among the periods. Fourth, statistical analysis of the time variables assumed independence of examinations; however, the study included multiple examinations per patient. Fifth, the study focused on diagnostic efficacy; we did not evaluate the value on patient outcomes and cost-effectiveness. Future studies should investigate the effect of early diagnosis of IPE on morbidity and mortality.

In conclusion, we demonstrated that commercially available AI software had high diagnostic accuracy in the detection of IPE on chest CT scans in patients with cancer and was effective in significantly reducing the time to diagnosis of positive examinations compared with the routine workflow in a setting with a backlog of unreported scans.