Commonly Missed Findings on Chest Radiographs: Causes and Consequences

Underestimation of lung regions on chest X-ray segmentation masks assessed by comparison with total lung volume evaluated on computed tomography

INTRODUCTION

The lung regions on chest X-ray segmentation masks created according to the current gold standard method for AI-driven applications are underestimated. This can be evaluated by comparison with computed tomography.

METHODS

This retrospective study included data from non-contrast chest low-dose CT examinations of 55 individuals without pulmonary pathology. Synthetic X-ray images were generated by projecting a 3D CT examination onto a 2D image plane. Two experienced radiologists manually created two types of lung masks: 3D lung masks from CT examinations (ground truth for further calculations) and 2D lung masks from synthetic X-ray images (according to the current gold standard method: following the contours of other anatomical structures). Overlapping and non-overlapping lung regions covered by both types of masks were analyzed. Volume of the overlapping regions was compared with total lung volume, and volume fractions of non-overlapping lung regions in relation to the total lung volume were calculated. The performance results between the two radiologists were compared.

RESULTS

Significant differences were observed between lung regions covered by CT and synthetic X-ray masks. The mean volume fractions of the lung regions not covered by synthetic X-ray masks for the right lung, the left lung, and both lungs were 22.8 %, 32.9 %, and 27.3 %, respectively, for Radiologist 1 and 22.7 %, 32.9 %, and 27.3 %, respectively, for Radiologist 2. There was excellent spatial agreement between the masks created by the two radiologists.

CONCLUSIONS

Lung X-ray masks created according to the current gold standard method significantly underestimate lung regions and do not cover substantial portions of the lungs.

IMPLICATIONS FOR PRACTICE

Standard lung masks fail to encompass the whole range of the lungs and significantly restrict the field of analysis in AI-driven applications, which may lead to false conclusions and diagnoses.

Collaboration between clinicians and vision-language models in radiology report generation

Automated radiology report generation has the potential to improve patient care and reduce the workload of radiologists. However, the path toward real-world adoption has been stymied by the challenge of evaluating the clinical quality of artificial intelligence (AI)-generated reports. We build a state-of-the-art report generation system for chest radiographs, called Flamingo-CXR, and perform an expert evaluation of AI-generated reports by engaging a panel of board-certified radiologists. We observe a wide distribution of preferences across the panel and across clinical settings, with 56.1% of Flamingo-CXR intensive care reports evaluated to be preferable or equivalent to clinician reports, by half or more of the panel, rising to 77.7% for in/outpatient X-rays overall and to 94% for the subset of cases with no pertinent abnormal findings. Errors were observed in human-written reports and Flamingo-CXR reports, with 24.8% of in/outpatient cases containing clinically significant errors in both report types, 22.8% in Flamingo-CXR reports only and 14.0% in human reports only. For reports that contain errors we develop an assistive setting, a demonstration of clinician-AI collaboration for radiology report composition, indicating new possibilities for potential clinical utility.

AI Applications for Thoracic Imaging: Considerations for Best Practice

Artificial intelligence (AI) technology is rapidly being introduced into thoracic radiology practice. Current representative use cases for AI in thoracic imaging show cumulative evidence of effectiveness. These include AI assistance for reading chest radiographs and low-dose (1.5-mSv) chest CT scans for lung cancer screening and triaging pulmonary embolism on chest CT scans. Other potential use cases are also under investigation, including filtering out normal chest radiographs, monitoring reading errors, and automated opportunistic screening of nontarget diseases. However, implementing AI tools in daily practice requires establishing practical strategies. Practical AI implementation will require objective on-site performance evaluation, institutional information technology infrastructure integration, and postdeployment monitoring. Meanwhile, the remaining challenges of adopting AI technology need to be addressed. These challenges include educating radiologists and radiology trainees, alleviating liability risk, and addressing potential disparities due to the uneven distribution of data and AI technology. Finally, next-generation AI technology represented by large language models (LLMs), including multimodal models, which can interpret both text and images, is expected to innovate the current landscape of AI in thoracic radiology practice. These LLMs offer opportunities ranging from generating text reports from images to explaining examination results to patients. However, these models require more research into their feasibility and efficacy.

Artificial intelligence-driven automated lung sizing from chest radiographs

Lung size measurements play an important role in transplantation, as optimal donor-recipient size matching is necessary to ensure the best possible outcome. Although several strategies for size matching are currently used, all have limitations, and none has proven superior. In this pilot study, we leveraged deep learning and computer vision to develop an automated system for generating standardized lung size measurements using portable chest radiographs to improve accuracy, reduce variability, and streamline donor/recipient matching. We developed a 2-step framework involving lung mask extraction from chest radiographs followed by feature point detection to generate 6 distinct lung height and width measurements, which we validated against measurements reported by 2 radiologists (T.A. and W.B.G.) for 50 lung transplant recipients. Our system demonstrated <2.5% error (<7.0 mm) with robust interrater and intrarater agreement compared with an expert radiologist review. This is especially promising given that the radiographs used in this study were purposely chosen to include images with technical challenges such as consolidations, effusions, and patient rotation. Although validation in a larger cohort is necessary, this study highlights artificial intelligence's potential to both provide reproducible lung size assessment in real patients and enable studies on the effect of lung size matching on transplant outcomes in large data sets.

The blind spots on chest computed tomography: what do we miss

Chest computed tomography (CT) is the most frequently performed imaging examination worldwide. Compared with chest radiography, chest CT greatly improves the detection rate and diagnostic accuracy of chest lesions because of the absence of overlapping structures and is the best imaging technique for the observation of chest lesions. However, there are still frequently missed diagnoses during the interpretation process, especially in certain areas or "blind spots", which may possibly be overlooked by radiologists. Awareness of these blind spots is of great significance to avoid false negative results and potential adverse consequences for patients. In this review, we summarize the common blind spots identified in actual clinical practice, encompassing the central areas within the pulmonary parenchyma (including the perihilar regions, paramediastinal regions, and operative area after surgery), trachea and bronchus, pleura, heart, vascular structure, external mediastinal lymph nodes, thyroid, osseous structures, breast, and upper abdomen. In addition to careful review, clinicians can employ several techniques to mitigate or minimize errors arising from these blind spots in film interpretation and reporting. In this review, we also propose technical methods to reduce missed diagnoses, including advanced imaging post-processing techniques such as multiplanar reconstruction (MPR), maximum intensity projection (MIP), artificial intelligence (AI) and structured reporting which can significantly enhance the detection of lesions and improve diagnostic accuracy.

Bridging human and machine intelligence: Reverse-engineering radiologist intentions for clinical trust and adoption

In the rapidly evolving landscape of medical imaging, the integration of artificial intelligence (AI) with clinical expertise offers unprecedented opportunities to enhance diagnostic precision and accuracy. Yet, the "black box" nature of AI models often limits their integration into clinical practice, where transparency and interpretability are important. This paper presents a novel system leveraging the Large Multimodal Model (LMM) to bridge the gap between AI predictions and the cognitive processes of radiologists. This system consists of two core modules, Temporally Grounded Intention Detection (TGID) and Region Extraction (RE). The TGID module predicts the radiologist's intentions by analyzing eye gaze fixation heatmap videos and corresponding radiology reports. Additionally, the RE module extracts regions of interest that align with these intentions, mirroring the radiologist's diagnostic focus. This approach introduces a new task, radiologist intention detection, and is the first application of Dense Video Captioning (DVC) in the medical domain. By making AI systems more interpretable and aligned with radiologist's cognitive processes, this proposed system aims to enhance trust, improve diagnostic accuracy, and support medical education. Additionally, it holds the potential for automated error correction, guiding junior radiologists, and fostering more effective training and feedback mechanisms. This work sets a precedent for future research in AI-driven healthcare, offering a pathway towards transparent, trustworthy, and human-centered AI systems. We evaluated this model using NLG(Natural Language Generation), time-related, and vision-based metrics, demonstrating superior performance in generating temporally grounded intentions on REFLACX and EGD-CXR datasets. This model also demonstrated strong predictive accuracy in overlap scores for medical abnormalities and effective region extraction with high IoU(Intersection over Union), especially in complex cases like cardiomegaly and edema. These results highlight the system's potential to enhance diagnostic accuracy and support continuous learning in radiology.

Large-Scale Study on AI's Impact on Identifying Chest Radiographs with No Actionable Disease in Outpatient Imaging

RATIONALE AND OBJECTIVES

Given the high volume of chest radiographs, radiologists frequently encounter heavy workloads. In outpatient imaging, a substantial portion of chest radiographs show no actionable findings. Automatically identifying these cases could improve efficiency by facilitating shorter reading workflows.

PURPOSE

A large-scale study to assess the performance of AI on identifying chest radiographs with no actionable disease (NAD) in an outpatient imaging population using comprehensive, objective, and reproducible criteria for NAD.

MATERIALS AND METHODS

The independent validation study includes 15000 patients with chest radiographs in posterior-anterior (PA) and lateral projections from an outpatient imaging center in the United States. Ground truth was established by reviewing CXR reports and classifying cases as NAD or actionable disease (AD). The NAD definition includes completely normal chest radiographs and radiographs with well-defined non-actionable findings. The AI NAD Analyzer1 (trained with 100 million multimodal images and fine-tuned on 1.3 million radiographs) utilizes a tandem system with image-level rule in and compartment-level rule out to provide case level output as NAD or potential actionable disease (PAD).

RESULTS

A total of 14057 cases met our eligibility criteria (age 56 ± 16.1 years, 55% women and 45% men). The prevalence of NAD cases in the study population was 70.7%. The AI NAD Analyzer correctly classified NAD cases with a sensitivity of 29.1% and a yield of 20.6%. The specificity was 98.9% which corresponds to a miss rate of 0.3% of cases. Significant findings were missed in 0.06% of cases, while no cases with critical findings were missed by AI.

CONCLUSION

In an outpatient population, AI can identify 20% of chest radiographs as NAD with a very low rate of missed findings. These cases could potentially be read using a streamlined protocol, thus improving efficiency and consequently reducing daily workload for radiologists.

A survey of bridging bone on chest radiography shows a greater than expected prevalence of marginal syndesmophytes

BACKGROUND

The recognition of thin marginal spinal syndesmophytes is important, in part due to their association with non-traumatic or mildly traumatic vertebral fractures.

PURPOSE

To determine a lower limit on the prevalence of marginal spinal syndesmophytes using chest radiographs.

MATERIAL AND METHODS

We conducted a retrospective analysis of 500 chest radiographs, assessing the prevalence of thin marginal syndesmophytes, bridging or near-bridging osteophytes, and flowing paravertebral ossifications in the thoracic intervertebral discs among individuals aged 16 years and older in a North American city.

RESULTS

Among the 500 participants, we observed that thin vertical marginal syndesmophytes were present in 17 (3.4%) cases, bridging or near-bridging osteophytes were present in 126 (25.2%) cases, and flowing paravertebral ossifications were present in 37 (7.4%) cases. Out of the 17 participants with thin marginal syndesmophytes, 10 exhibited a bamboo-like spine appearance, defined as the presence of ≥4 contiguous levels of bridging marginal syndesmophytes. Analysis of syndesmophyte distribution per vertebral level indicated a higher frequency of involvement in the mid to lower thoracic spine, maximal at T9/10.

CONCLUSIONS

The presence of thin marginal syndesmophytes in the thoracic spine on routine chest radiographs is substantially more prevalent than would be anticipated based on the existing literature. The feasibility of reliably identifying these syndesmophytes in the spine and the impact of this on morbidity should be further investigated due to their association with advanced ankylosing spondylitis and their susceptibility to fractures.

Assessment of pulmonary function in COPD patients using dynamic digital radiography: A novel approach utilizing lung signal intensity changes during forced breathing

Objectives

To investigate the association of lung signal intensity changes during forced breathing using dynamic digital radiography (DDR) with pulmonary function and disease severity in patients with chronic obstructive pulmonary disease (COPD).

Methods

This retrospective study included 46 healthy subjects and 33 COPD patients who underwent posteroanterior chest DDR examination. We collected raw signal intensity and gray-scale image data. The lung contour was extracted on the gray-scale images using our previously developed automated lung field tracking system and calculated the average of signal intensity values within the extracted lung contour on gray-scale images. Lung signal intensity changes were quantified as SImax/SImin, representing the maximum ratio of the average signal intensity in the inspiratory phase to that in the expiratory phase. We investigated the correlation between SImax/SImin and pulmonary function parameters, and differences in SImax/SImin by disease severity.

Results

SImax/SImin showed the highest correlation with VC (rs = 0.54, P < 0.0001), followed by FEV1 (rs = 0.44, P < 0.0001), both of which are key indicators of COPD pathophysiology. In a multivariate linear regression analysis adjusted for confounding factors, SImax/SImin was significantly lower in the severe COPD group compared to the normal group (P = 0.0004) and mild COPD group (P=0.0022), suggesting its potential usefulness in assessing COPD severity.

Conclusion

This study suggests that the signal intensity changes of lung fields during forced breathing using DDR reflect the pathophysiology of COPD and can be a useful index in assessing pulmonary function in COPD patients, potentially improving COPD diagnosis and management.

Correlation Between Clinical Assessment and Postmortem Lung Biopsy in Patients With Pulmonary Infiltrates and Respiratory Failure

BACKGROUND

Pulmonary complications commonly cause mortality in critically ill patients with acute and chronic liver diseases in the intensive care unit. Pneumonia is the most common clinical diagnosis supported by radiological and microbiological results, which are subjective and often with poor yield. The study aimed to correlate clinical diagnosis and postmortem lung histology in patients with liver disease with respiratory failure.

METHODS

Records of acute and chronic liver disease patients with respiratory failure-associated mortality and postmortem lung biopsy from September 2009 to March 2020 were analyzed. Clinical diagnosis supported by radiological and/or microbiological data was compared with histology.

RESULTS

One hundred eight patients (age 46.83±12.96 years), males 80 (74.1%), 63 (58.3%) cirrhosis of the liver, 30 (27.8%) acute-on-chronic-liver-failure, and 9 (8.3%) acute liver failure, were analyzed. Of the 76 patients (70.37 % of the total) with pneumonia, 33 (43.4 %) had histological evidence of pneumonia. Other histological diagnoses in these patients were normal or nonspecific changes in 27 (35.5 %) and alveolar hemorrhage in 13 (17.1 %). In the remaining 32 patients, histological diagnosis of pneumonia was evident in nine patients (28.1%). Using postmortem histology as the gold standard, the sensitivity, specificity, positive predictive value, and negative predictive value for clinical diagnosis of pneumonia were found to be 78.57%, 34.85%, 43.42%, and 71.88% respectively. The kappa statistics for agreement between the two was 0.12 (95% C.I. -0.04 to 0.27) suggesting poor agreement. Age and histological pneumonia predicted significant missed diagnosis.

CONCLUSION

There is poor agreement between clinical diagnosis and postmortem histology. Postmortem lung biopsy helps with the unexplained cause of death.

Artificial intelligence-assisted double reading of chest radiographs to detect clinically relevant missed findings: a two-centre evaluation

OBJECTIVES

To evaluate an artificial intelligence (AI)-assisted double reading system for detecting clinically relevant missed findings on routinely reported chest radiographs.

METHODS

A retrospective study was performed in two institutions, a secondary care hospital and tertiary referral oncology centre. Commercially available AI software performed a comparative analysis of chest radiographs and radiologists' authorised reports using a deep learning and natural language processing algorithm, respectively. The AI-detected discrepant findings between images and reports were assessed for clinical relevance by an external radiologist, as part of the commercial service provided by the AI vendor. The selected missed findings were subsequently returned to the institution's radiologist for final review.

RESULTS

In total, 25,104 chest radiographs of 21,039 patients (mean age 61.1 years ± 16.2 [SD]; 10,436 men) were included. The AI software detected discrepancies between imaging and reports in 21.1% (5289 of 25,104). After review by the external radiologist, 0.9% (47 of 5289) of cases were deemed to contain clinically relevant missed findings. The institution's radiologists confirmed 35 of 47 missed findings (74.5%) as clinically relevant (0.1% of all cases). Missed findings consisted of lung nodules (71.4%, 25 of 35), pneumothoraces (17.1%, 6 of 35) and consolidations (11.4%, 4 of 35).

CONCLUSION

The AI-assisted double reading system was able to identify missed findings on chest radiographs after report authorisation. The approach required an external radiologist to review the AI-detected discrepancies. The number of clinically relevant missed findings by radiologists was very low.

CLINICAL RELEVANCE STATEMENT

The AI-assisted double reader workflow was shown to detect diagnostic errors and could be applied as a quality assurance tool. Although clinically relevant missed findings were rare, there is potential impact given the common use of chest radiography.

KEY POINTS

• A commercially available double reading system supported by artificial intelligence was evaluated to detect reporting errors in chest radiographs (n=25,104) from two institutions. • Clinically relevant missed findings were found in 0.1% of chest radiographs and consisted of unreported lung nodules, pneumothoraces and consolidations. • Applying AI software as a secondary reader after report authorisation can assist in reducing diagnostic errors without interrupting the radiologist's reading workflow. However, the number of AI-detected discrepancies was considerable and required review by a radiologist to assess their relevance.

Artificial Intelligence in Cardiovascular Disease Prevention: Is it Ready for Prime Time?

PURPOSE OF REVIEW

This review evaluates how Artificial Intelligence (AI) enhances atherosclerotic cardiovascular disease (ASCVD) risk assessment, allows for opportunistic screening, and improves adherence to guidelines through the analysis of unstructured clinical data and patient-generated data. Additionally, it discusses strategies for integrating AI into clinical practice in preventive cardiology.

RECENT FINDINGS

AI models have shown superior performance in personalized ASCVD risk evaluations compared to traditional risk scores. These models now support automated detection of ASCVD risk markers, including coronary artery calcium (CAC), across various imaging modalities such as dedicated ECG-gated CT scans, chest X-rays, mammograms, coronary angiography, and non-gated chest CT scans. Moreover, large language model (LLM) pipelines are effective in identifying and addressing gaps and disparities in ASCVD preventive care, and can also enhance patient education. AI applications are proving invaluable in preventing and managing ASCVD and are primed for clinical use, provided they are implemented within well-regulated, iterative clinical pathways.

Imaging of the Diaphragm: A Primer

The diaphragm is a critical musculotendinous structure that contributes to respiratory function. Disorders of the diaphragm are rare and diagnostically challenging. Herein, the author reviews the radiologic options for the assessment of the diaphragm.

Misinterpretation of a skin fold artifact as pneumothorax on the chest x-ray of a trauma patient in Korea: a case report

Misinterpreting radiographic findings can lead to unnecessary interventions and potential patient harm. The urgency required when responding to the compromised health of trauma patients can increase the likelihood of misinterpreting chest x-rays in critical situations. We present the case report of a trauma patient whose skin fold artifacts were mistaken for pneumothorax on a follow-up chest x-ray, resulting in unnecessary chest tube insertion. We hope to help others differentiate between skin folds and pneumothorax on the chest x-rays of trauma patients by considering factors such as location, shape, sharpness, and vascular markings.

Spontaneous Pneumothorax in a Healthy Young Woman: Discussion About Treatment Options

A spontaneous pneumothorax, a potentially life-threatening condition, is a disease process in which air enters the space between the visceral and parietal pleural of the lung, thus increasing the pressures in that space. It can be diagnosed by both physical exam and radiographic testing. In this case, we present a 21-year-old, otherwise healthy woman who presented with sudden, sharp shoulder pain and chest tightness and was diagnosed with her first, spontaneous pneumothorax. We further discuss the diagnosis and treatment options for a first-time spontaneous pneumothorax.

Incidental finding of double aortic arch and tract from noncoronary sinus to left atrium: A case report

The double aortic arch with vascular ring is a rare but documented aortic arch variant, traditionally presenting with difficulty swallowing or breathing due to extrinsic compression. Tracts from the noncoronary sinus to left atrium are very rare, with limited case reports to compare against. We report an incidental finding of double aortic arch in an elderly woman who underwent a cerebral angiogram for symptoms of a right-sided stroke, with a further anomaly identified on subsequent CT gated aortogram of a possible tract between the non-coronary sinus and left atrium. It is worth noting that the aortic arch abnormality was missed on previous plain radiographs, which can happen even among experienced radiologists. This case illustrates the need for a thorough, systematic approach to interpreting chest radiographs to avoid missing mediastinal lesions, such as aortic abnormalities.

Generative Artificial Intelligence for Chest Radiograph Interpretation in the Emergency Department

Importance

Multimodal generative artificial intelligence (AI) methodologies have the potential to optimize emergency department care by producing draft radiology reports from input images.

Objective

To evaluate the accuracy and quality of AI-generated chest radiograph interpretations in the emergency department setting.

Design, Setting, and Participants

This was a retrospective diagnostic study of 500 randomly sampled emergency department encounters at a tertiary care institution including chest radiographs interpreted by both a teleradiology service and on-site attending radiologist from January 2022 to January 2023. An AI interpretation was generated for each radiograph. The 3 radiograph interpretations were each rated in duplicate by 6 emergency department physicians using a 5-point Likert scale.

Main Outcomes and Measures

The primary outcome was any difference in Likert scores between radiologist, AI, and teleradiology reports, using a cumulative link mixed model. Secondary analyses compared the probability of each report type containing no clinically significant discrepancy with further stratification by finding presence, using a logistic mixed-effects model. Physician comments on discrepancies were recorded.

Results

A total of 500 ED studies were included from 500 unique patients with a mean (SD) age of 53.3 (21.6) years; 282 patients (56.4%) were female. There was a significant association of report type with ratings, with post hoc tests revealing significantly greater scores for AI (mean [SE] score, 3.22 [0.34]; P < .001) and radiologist (mean [SE] score, 3.34 [0.34]; P < .001) reports compared with teleradiology (mean [SE] score, 2.74 [0.34]) reports. AI and radiologist reports were not significantly different. On secondary analysis, there was no difference in the probability of no clinically significant discrepancy between the 3 report types. Further stratification of reports by presence of cardiomegaly, pulmonary edema, pleural effusion, infiltrate, pneumothorax, and support devices also yielded no difference in the probability of containing no clinically significant discrepancy between the report types.

Conclusions and Relevance

In a representative sample of emergency department chest radiographs, results suggest that the generative AI model produced reports of similar clinical accuracy and textual quality to radiologist reports while providing higher textual quality than teleradiologist reports. Implementation of the model in the clinical workflow could enable timely alerts to life-threatening pathology while aiding imaging interpretation and documentation.