Evaluation of an Artificial Intelligence Model for Detection of Pneumothorax and Tension Pneumothorax in Chest Radiographs

IMPORTANCE

Early detection of pneumothorax, most often via chest radiography, can help determine need for emergent clinical intervention. The ability to accurately detect and rapidly triage pneumothorax with an artificial intelligence (AI) model could assist with earlier identification and improve care.

OBJECTIVE

To compare the accuracy of an AI model vs consensus thoracic radiologist interpretations in detecting any pneumothorax (incorporating both nontension and tension pneumothorax) and tension pneumothorax.

DESIGN, SETTING, AND PARTICIPANTS

This diagnostic study was a retrospective standalone performance assessment using a data set of 1000 chest radiographs captured between June 1, 2015, and May 31, 2021. The radiographs were obtained from patients aged at least 18 years at 4 hospitals in the Mass General Brigham hospital network in the United States. Included radiographs were selected using 2 strategies from all chest radiography performed at the hospitals, including inpatient and outpatient. The first strategy identified consecutive radiographs with pneumothorax through a manual review of radiology reports, and the second strategy identified consecutive radiographs with tension pneumothorax using natural language processing. For both strategies, negative radiographs were selected by taking the next negative radiograph acquired from the same radiography machine as each positive radiograph. The final data set was an amalgamation of these processes. Each radiograph was interpreted independently by up to 3 radiologists to establish consensus ground-truth interpretations. Each radiograph was then interpreted by the AI model for the presence of pneumothorax and tension pneumothorax. This study was conducted between July and October 2021, with the primary analysis performed between October and November 2021.

MAIN OUTCOMES AND MEASURES

The primary end points were the areas under the receiver operating characteristic curves (AUCs) for the detection of pneumothorax and tension pneumothorax. The secondary end points were the sensitivities and specificities for the detection of pneumothorax and tension pneumothorax.

RESULTS

The final analysis included radiographs from 985 patients (mean [SD] age, 60.8 [19.0] years; 436 [44.3%] female patients), including 307 patients with nontension pneumothorax, 128 patients with tension pneumothorax, and 550 patients without pneumothorax. The AI model detected any pneumothorax with an AUC of 0.979 (95% CI, 0.970-0.987), sensitivity of 94.3% (95% CI, 92.0%-96.3%), and specificity of 92.0% (95% CI, 89.6%-94.2%) and tension pneumothorax with an AUC of 0.987 (95% CI, 0.980-0.992), sensitivity of 94.5% (95% CI, 90.6%-97.7%), and specificity of 95.3% (95% CI, 93.9%-96.6%).

CONCLUSIONS AND RELEVANCE

These findings suggest that the assessed AI model accurately detected pneumothorax and tension pneumothorax in this chest radiograph data set. The model's use in the clinical workflow could lead to earlier identification and improved care for patients with pneumothorax.

Thoracic Imaging Manifestations of Treated Lymphomas: Response Evaluation, Posttherapeutic Sequelae, and Complications

Lymphoma is the most common hematologic malignancy comprising a diverse group of neoplasms arising from multiple blood cell lineages. Any structure of the thorax may be involved at any stage of disease. Imaging has a central role in the initial staging, response assessment, and surveillance of lymphoma, and updated standardized assessment criteria are available to assist with imaging interpretation and reporting. Radiologists should be aware of the modern approaches to lymphoma treatment, the role of imaging in posttherapeutic surveillance, and manifestations of therapy-related complications.

Deep learning in chest radiography: Detection of findings and presence of change

Background

Deep learning (DL) based solutions have been proposed for interpretation of several imaging modalities including radiography, CT, and MR. For chest radiographs, DL algorithms have found success in the evaluation of abnormalities such as lung nodules, pulmonary tuberculosis, cystic fibrosis, pneumoconiosis, and location of peripherally inserted central catheters. Chest radiography represents the most commonly performed radiological test for a multitude of non-emergent and emergent clinical indications. This study aims to assess accuracy of deep learning (DL) algorithm for detection of abnormalities on routine frontal chest radiographs (CXR), and assessment of stability or change in findings over serial radiographs.

Methods and findings

We processed 874 de-identified frontal CXR from 724 adult patients (> 18 years) with DL (Qure AI). Scores and prediction statistics from DL were generated and recorded for the presence of pulmonary opacities, pleural effusions, hilar prominence, and enlarged cardiac silhouette. To establish a standard of reference (SOR), two thoracic radiologists assessed all CXR for these abnormalities. Four other radiologists (test radiologists), unaware of SOR and DL findings, independently assessed the presence of radiographic abnormalities. A total 724 radiographs were assessed for detection of findings. A subset of 150 radiographs with follow up examinations was used to asses change over time. Data were analyzed with receiver operating characteristics analyses and post-hoc power analysis.

Results

About 42% (305/ 724) CXR had no findings according to SOR; single and multiple abnormalities were seen in 23% (168/724) and 35% (251/724) of CXR. There was no statistical difference between DL and SOR for all abnormalities (p = 0.2–0.8). The area under the curve (AUC) for DL and test radiologists ranged between 0.837–0.929 and 0.693–0.923, respectively. DL had lowest AUC (0.758) for assessing changes in pulmonary opacities over follow up CXR. Presence of chest wall implanted devices negatively affected the accuracy of DL algorithm for evaluation of pulmonary and hilar abnormalities.

Conclusions

DL algorithm can aid in interpretation of CXR findings and their stability over follow up CXR. However, in its present version, it is unlikely to replace radiologists due to its limited specificity for categorizing specific findings.

What Patients Want to Know about Imaging Examinations: A Multiinstitutional U.S. Survey in Adult and Pediatric Teaching Hospitals on Patient Preferences for Receiving Information before Radiologic Examinations

Purpose To identify what information patients and parents or caregivers found useful before an imaging examination, from whom they preferred to receive information, and how those preferences related to patient-specific variables including demographics and prior radiologic examinations. Materials and Methods A 24-item survey was distributed at three pediatric and three adult hospitals between January and May 2015. The χ² or Fisher exact test (categorical variables) and one-way analysis of variance or two-sample t test (continuous variables) were used for comparisons. Multivariate logistic regression was used to determine associations between responses and demographics. Results Of 1742 surveys, 1542 (89%) were returned (381 partial, 1161 completed). Mean respondent age was 46.2 years ± 16.8 (standard deviation), with respondents more frequently female (1025 of 1506, 68%) and Caucasian (1132 of 1504, 75%). Overall, 78% (1117 of 1438) reported receiving information about their examination most commonly from the ordering provider (824 of 1292, 64%), who was also the most preferred source (1005 of 1388, 72%). Scheduled magnetic resonance (MR) imaging or nuclear medicine examinations (P < .001 vs other examination types) and increasing education (P = .008) were associated with higher rates of receiving information. Half of respondents (757 of 1452, 52%) sought information themselves. The highest importance scores for pre-examination information (Likert scale ≥4) was most frequently assigned to information on examination preparation and least frequently assigned to whether an alternative radiation-free examination could be used (74% vs 54%; P < .001). Conclusion Delivery of pre-examination information for radiologic examinations is suboptimal, with half of all patients and caregivers seeking information on their own. Ordering providers are the predominant and preferred source of examination-related information, with respondents placing highest importance on information related to examination preparation. ^© RSNA, 2018 Online supplemental material is available for this article.

Bedside Chest Radiographs in the Intensive care Setting: Wireless Direct Radiography Compared to Computed Radiography

OBJECTIVES

To compare image quality, visibility of anatomic landmarks, tubes and lines, and other clinically significant findings on portable (bedside) chest radiographs acquired with wireless direct radiography (DRw) and computed radiography (CR).

METHODS

In a prospective IRB-approved and HIPAA-compliant study, portable DRw (DRX-1C mobile retrofit portable wireless direct radiography, CareStream Inc., Rochester, NY) and portable CR (AGFA CR (DXG) version; NIM2103, AGFA Healthcare, Ridgefield Park, NJ) images of the chest were acquired within 24-hours in 80 patients in the intensive care unit (ICU). Image pairs of 75 patients (37% female) with a mean age of 60.7±16 years were independently compared side-by-side by 7 experienced thoracic radiologists using a five-point scale. When tubes and lines were present, the radiologist also compared an edge-enhanced copy of the DRw image to the CR image.

RESULTS

Most radiologists found significantly fewer artifacts on DRw images compared to CR images and all readers agreed that when present, these artifacts did not significantly preclude the ability to evaluate anatomic landmarks, tubes and lines, or clinically significant findings. None of the radiologists (0/7) reported superior visibility of anatomic structures on CR images compared to DRw images and some radiologists (3/7) found DRw images significantly better for visibility of anatomic landmarks such as the carina (p=0.01-0.001). Most radiologists (6/7) found DRw images to be better or clearly better than CR images for position of tubes and lines, and edge-enhanced DRw images to be especially helpful for evaluation of central venous catheters and esophageal tubes (p=0.027-0.001). None of the radiologists deemed CR images superior for visibility of clinically significant findings.

CONCLUSIONS

Critical care chest radiography with a portable DRw system can provide similar or superior information compared to a CR system regarding clinically significant findings and position of tubes and lines.

The 10 Pillars of Lung Cancer Screening: Rationale and Logistics of a Lung Cancer Screening Program

On the basis of the National Lung Screening Trial data released in 2011, the U.S. Preventive Services Task Force made lung cancer screening (LCS) with low-dose computed tomography (CT) a public health recommendation in 2013. The Centers for Medicare and Medicaid Services (CMS) currently reimburse LCS for asymptomatic individuals aged 55-77 years who have a tobacco smoking history of at least 30 pack-years and who are either currently smoking or had quit less than 15 years earlier. Commercial insurers reimburse the cost of LCS for individuals aged 55-80 years with the same smoking history. Effective care for the millions of Americans who qualify for LCS requires an organized step-wise approach. The 10-pillar model reflects the elements required to support a successful LCS program: eligibility, education, examination ordering, image acquisition, image review, communication, referral network, quality improvement, reimbursement, and research frontiers. Examination ordering can be coupled with decision support to ensure that only eligible individuals undergo LCS. Communication of results revolves around the Lung Imaging Reporting and Data System (Lung-RADS) from the American College of Radiology. Lung-RADS is a structured decision-oriented reporting system designed to minimize the rate of false-positive screening examination results. With nodule size and morphology as discriminators, Lung-RADS links nodule management pathways to the variety of nodules present on LCS CT studies. Tracking of patient outcomes is facilitated by a CMS-approved national registry maintained by the American College of Radiology. Online supplemental material is available for this article.

Accuracy of transmission CT and FDG-PET in the detection of small pulmonary nodules with integrated PET/CT

PURPOSE

The purpose of this study was to determine the accuracy of detection of small pulmonary nodules on quiet breathing attenuation correction CT (CTAC) and FDG-PET when performing integrated PET/CT, as compared with a diagnostic inspiratory CT scan acquired in the same imaging session.

METHODS

PET/CT scans of 107 patients with a history of carcinoma (54 male and 53 female, mean age 57.3 years) were analyzed. All patients received an integrated PET/CT scan including a CTAC acquired during quiet respiration and a contrast-enhanced CT acquired during inspiration in the same session. Breathing CTAC scans were reviewed by two thoracic radiologists for the presence of pulmonary nodules. FDG-PET scans were reviewed to determine accuracy of nodule detection. Diagnostic CT was used as the gold standard to confirm or refute the presence of nodules.

RESULTS

On the CTAC scans 200 nodules were detected, of which 183 were true positive (TP) and 17, false positive. There were 109 false negatives (FN). Overall, 51 (48%) patients had a false interpretation, including 19 in whom CT was interpreted as normal for lung nodules. The average size of the nodules missed was 3.8+/-2 mm (range 2-12 mm). None of the nodules missed on the CTAC scans were detected by PET. In the right lung there were 20 TP, 42 true negative (TN), 11 FP, and 34 FN interpretations with a sensitivity in nodule detection of 37% (CI 24-51%) and a specificity of 79% (CI 66-89%). In the left lungs there were 16 TP, 65 TN, 3 FP, and 23 FN interpretations, with a sensitivity of 41% (CI 26-58%) and a specificity of 96% (CI 88-99%).

CONCLUSION

The detection of small pulmonary nodules by breathing CTAC and FDG-PET is relatively poor. Therefore an additional diagnostic thoracic CT scan obtained during suspended inspiration is recommended for thorough evaluation of those patients in whom detection of pulmonary metastases is necessary for management.

An automated PACS workstation interface: a timesaving enhancement

OBJECTIVE

Interface design is a key element in the efficient use of a picture archiving and communication system (PACS) workstation. In many cases, multiple mouse clicks or keyboard commands are required to open and close a case, to mark it as complete, and to retrieve and allocate screen positions to the next case. We evaluated the work flow effect of software designed for automated image display in which all of these operations are consolidated in a single mouse click.

CONCLUSION

Automated image display increases efficiency in image interpretation and remedies the normally cluttered presentation environment. At our institution, acceptance of automated image display has been overwhelmingly positive. In fact, automated image display has improved radiologist productivity.

Needle localization of peripheral lung nodules for video-assisted thoracoscopic surgery

Video-assisted thoracoscopic surgery provides an alternative to conventional thoracotomy for resection of peripheral lung nodules. To localize small peripheral lung nodules that may not be visible or palpable by the surgeon, we have placed a Kopans hook wire percutaneously into the lung as a guide. The indications for localization included previous nondiagnostic percutaneous needle aspiration biopsy (PNAB) (n = 4), nodules too small for PNAB (n = 2), nodules inaccessible to PNAB (n = 3), and planned resection of a known peripheral tumor less than 1 cm (n = 1). The localization procedure was performed with computed tomographic guidance in all patients. The nodules ranged in size from 2 to 15 mm and were located immediately subpleural to 2-cm deep the pleura. A 20-gauge Greene biopsy needle was used as an introducer for a 35-cm-long Kopans hook wire. Patients were sent directly to the operating room in a dependent position. All ten nodules were successfully resected, including hamartoma (n = 1), carcinoid tumors (n = 2), granulomas (n = 3), adenocarcinoma (n = 1), fibrosis (n = 1), benign metastasizing leiomyoma (n = 1), and lymphoma (n = 1). In two patients, the wire slipped out of the lung. Small focal pneumothoraces developed in five patients. There were no major complications. This procedure can safely and effectively localize nonvisible or nonpalpable pulmonary nodules for thoracoscopic surgery for diagnostic purposes or for resection of small peripheral tumors in patients who cannot tolerate a lobectomy or pneumonectomy.