Article Impact Level: HIGH Data Quality: STRONG Summary of Radiology, 306(2), e221926. https://doi.org/10.1148/radiol.221926 Dr. Márton Kolossváry et al
Points
- Based on chest radiographs in asymptomatic persons, deep learning has also been used to forecast fragility and prolonged prognosis.
- Individuals suffering from acute chest pain syndrome can benefit from more effective screening owing to deep learning examination of plain radiography.
- Initial chest radiography is frequently performed on patients who report to the intensive care unit with severe breast pain syndrome manifestations. They may be advantaged from deep learning investigation.
- Aortic dissection, deep vein thrombosis, and acute coronary syndrome are frequently ruled out in patients who report burning breast pain to the intensive care unit.
Summary
The study aimed to determine if a deep learning assessment of the initial chest radiograph might aid in the effective screening of individuals with severe breast pain syndrome.
Baseline breast images (n = 23,005) of individuals suffering from burning breast pain disease were retrospectively used to train an accessible deep learning system to detect an irregular heartbeat, a collapsed lung, peripheral arterial disease, and all-cause mortality.
Marton Kolossvary investigated whether an in-depth learning assessment of a baseline chest radiograph would aid in treating patients with acute chest pain syndrome. A deep learning model was used on the 23,005 patient chest radiographs to predict the 30-day cumulative endpoint.
Model one consisted of age and gender; the second model incorporated model one with traditional troponin, and the third model consisted of model two with deep learning forecast. These characteristics were used to examine the activities of different models in internal and external databases.
On the experimental testing dataset, the model performed better than one that considered factors such as age, gender, troponin, or D-dimer positivity. In addition, with 99% sensitivity, the model postponed additional cardiovascular or pulmonary tests for many individuals.
Patient outcome findings show that the deep learning technique had excellent test precision for detecting all-cause death. This was consistent with earlier research results showing that deep learning can correctly estimate prolonged mortality. The model’s efficacy as a predictor of acute chest syndrome was the lowest. This was because models including age, gender, and troponin can detect myocardial infarction. The International Classification of Diseases code for acute chest syndrome mostly depends on troponin positivity. Surprisingly, none of the models could accurately identify these individuals with high specificity regarding pulmonary embolism diagnosis.
Considering the bias of deep learning against age, gender, cultural, and phenotypic traits, several issues have been addressed. The disparities in illness between groups and the inadequate electronic medical record information accessible to minority patient groups are mostly to blame for these biases. The secondary studies demonstrate that the deep learning model’s clinical diagnosis was comparable across age, gender, ethnicity, and geographical categories.
The solid evidence of the study concludes that individuals with severe breast pain disorder are prone to adverse outcomes and can be identified by a deep learning review of early critical care unit chest radio images. In addition, future deep learning techniques might facilitate faster triage of acute chest pain syndrome individuals relying on early breast chest radiographs to determine which patients can wait for additional imaging.
Link to the article: https://pubs.rsna.org/doi/10.1148/radiol.221926
References Kolossváry, M., Raghu, V. K., Nagurney, J. T., Hoffmann, U., & Lu, M. T. (2023). Deep learning analysis of chest radiographs to triage patients with acute chest pain syndrome. Radiology, 306(2), e221926. https://doi.org/10.1148/radiol.221926