RESUMO
OBJECTIVES: This study aims to develop a deep learning algorithm, Pneumonia-Plus, based on computed tomography (CT) images for accurate classification of bacterial, fungal, and viral pneumonia. METHODS: A total of 2763 participants with chest CT images and definite pathogen diagnosis were included to train and validate an algorithm. Pneumonia-Plus was prospectively tested on a nonoverlapping dataset of 173 patients. The algorithm's performance in classifying three types of pneumonia was compared to that of three radiologists using the McNemar test to verify its clinical usefulness. RESULTS: Among the 173 patients, area under the curve (AUC) values for viral, fungal, and bacterial pneumonia were 0.816, 0.715, and 0.934, respectively. Viral pneumonia was accurately classified with sensitivity, specificity, and accuracy of 0.847, 0.919, and 0.873. Three radiologists also showed good consistency with Pneumonia-Plus. The AUC values of bacterial, fungal, and viral pneumonia were 0.480, 0.541, and 0.580 (radiologist 1: 3-year experience); 0.637, 0.693, and 0.730 (radiologist 2: 7-year experience); and 0.734, 0.757, and 0.847 (radiologist 3: 12-year experience), respectively. The McNemar test results for sensitivity showed that the diagnostic performance of the algorithm was significantly better than that of radiologist 1 and radiologist 2 (p < 0.05) in differentiating bacterial and viral pneumonia. Radiologist 3 had a higher diagnostic accuracy than the algorithm. CONCLUSIONS: The Pneumonia-Plus algorithm is used to differentiate between bacterial, fungal, and viral pneumonia, which has reached the level of an attending radiologist and reduce the risk of misdiagnosis. The Pneumonia-Plus is important for appropriate treatment and avoiding the use of unnecessary antibiotics, and provide timely information to guide clinical decision-making and improve patient outcomes. CLINICAL RELEVANCE STATEMENT: Pneumonia-Plus algorithm could assist in the accurate classification of pneumonia based on CT images, which has great clinical value in avoiding the use of unnecessary antibiotics, and providing timely information to guide clinical decision-making and improve patient outcomes. KEY POINTS: ⢠The Pneumonia-Plus algorithm trained from data collected from multiple centers can accurately identify bacterial, fungal, and viral pneumonia. ⢠The Pneumonia-Plus algorithm was found to have better sensitivity in classifying viral and bacterial pneumonia in comparison to radiologist 1 (5-year experience) and radiologist 2 (7-year experience). ⢠The Pneumonia-Plus algorithm is used to differentiate between bacterial, fungal, and viral pneumonia, which has reached the level of an attending radiologist.
Assuntos
Aprendizado Profundo , Pneumonia Bacteriana , Pneumonia Viral , Humanos , Pneumonia Viral/diagnóstico por imagem , Tomografia Computadorizada por Raios X/métodos , Antibacterianos , Pneumonia Bacteriana/diagnóstico por imagem , Estudos RetrospectivosRESUMO
Background: The commercial coronary computed tomographic angiography artificial intelligence (CCTA-AI) platform has made great progress in clinical application. However, research is needed to elucidate the current stage of commercial AI platforms and the role of radiologists. This study compared the diagnostic performance of the commercial CCTA-AI platform with that of a reader based on a multicenter and multidevice sample. Methods: A total of 318 patients with suspected coronary artery disease (CAD) who underwent both CCTA and invasive coronary angiography (ICA) were included in a multicenter and multidevice validation cohort between 2017 and 2021. The commercial CCTA-AI platform was used to automatically assess coronary artery stenosis by using ICA findings as the gold standard. The CCTA reader was completed by radiologists. The diagnostic performance of the commercial CCTA-AI platform and CCTA reader was evaluated at the patient and segment levels. The cutoff values of models 1 and 2 were 50% and 70% stenosis, respectively. Results: It took 20.4 seconds to accomplish post-processing per patient when using the CCTA-AI platform, which was significantly shorter than the time taken to complete this task with the CCTA reader (1,112.1 s). In the patient-based analysis, the area under the curve (AUC) was 0.85 using the CCTA-AI platform and 0.61 using the CCTA reader in model 1 (stenosis ratio: 50%). In contrast, the AUC was 0.78 using the CCTA-AI platform and 0.64 using the CCTA reader in model 2 (stenosis ratio: 70%). In the segment-based analysis, the AUCs of CCTA-AI were slightly better than those of the readers. The negative predictive value (NPV) increased from model 1 to model 2. Furthermore, the diagnostic performance was better for larger-diameter arteries. Conclusions: The commercial CCTA-AI platform may provide a feasible solution for the diagnosis of coronary artery stenosis, and it has a diagnostic performance that is slightly better than that of a radiologist with a moderate level of experience (5-10 years of experience).
RESUMO
BACKGROUND: A double superior vena cava (DSVC) may cause technical difficulties in some cardiovascular procedures. However, no quantitative data exist to describe the morphological features of this anomaly. METHODS: From January 2015 to January 2019, the data of 128 consecutive patients diagnosed with DSVC on computed tomography (CT) images were retrospectively analyzed. We proposed an easy and rational method for DSVC classification based on the presence or absence of the left brachiocephalic vein (LBCV), the presence or absence of an anastomotic vein bridging the bilateral superior vena cava (SVC), and the drainage pattern of the left superior vena cava (LSVC). The following classifications were established: type I, LBVC absent, LSVC drainage into the right atrium via the coronary sinus; type II, LBCV present, LSVC drainage into the right atrium via the coronary sinus; type III, LBCV absent, LSVC drainage into the right atrium via the anastomosis; type IV, LBCV present, LSVC drainage into the right atrium via the anastomosis. The length, diameter, and area of the bilateral SVC and the coronary sinus were carefully measured across the 4 types. RESULTS: Type I was the most frequently occurring type (66 of 128, 51.6%), followed by type II (43 of 128, 33.6%), then type III (15 of 128, 11.7%), and type IV (4 of 128, 3.1%). The LSVC was significantly longer than the right SVC (RSVC) in all 4 types, and the diameters of the LSVC were significantly larger in types without the LBCV (i.e., types I and III) (P<0.0001 for all). Additionally, the diameter of the coronary sinus in types I and II was triple that in types III and IV (P<0.0001), which was thought to be due to increased venous blood reflux through the coronary sinus. CONCLUSIONS: The anatomical features of DSVC can be satisfactorily depicted on CT. The quantitative measurement of this anomaly by the reporting radiologists could assist clinicians to minimize the procedure-associated risks.
RESUMO
Background: Coronary microvascular dysfunction (CMD) is an early character of type 2 diabetes mellitus (T2DM), and is indicative of adverse events. The present study aimed to validate the performance of the stress T1 mapping technique on cardiac magnetic resonance (CMR) for identifying CMD from a histopathologic perspective and to establish the time course of CMD-related parameters in a rabbit model of T2DM. Methods: New Zealand white rabbits (n = 30) were randomly divided into a control (n = 8), T2DM 5-week (n = 6), T2DM 10-week (n = 9), and T2DM 15-week (n = 7) groups. The CMR protocol included rest and adenosine triphosphate (ATP) stress T1-mapping imaging using the 5b(20b)3b-modified look-locker inversion-recovery (MOLLI) schema to quantify stress T1 response (stress ΔT1), and first-pass perfusion CMR to quantify myocardial perfusion reserve index (MPRI). After the CMR imaging, myocardial tissue was subjected to hematoxylin-eosin staining to evaluate pathological changes, Masson trichrome staining to measure collagen volume fraction (CVF), and CD31 staining to measure microvascular density (MVD). The associations between CMR parameters and pathological findings were determined using Pearson correlation analysis. Results: The stress ΔT1 values were 6.21 ± 0.59%, 4.88 ± 0.49%, 3.80 ± 0.40%, and 3.06 ± 0.54% in the control, T2DM 5-week, 10-week, and 15-week groups, respectively (p < 0.001) and were progressively weakened with longer duration of T2DM. Furthermore, a significant correlation was demonstrated between the stress ΔT1 vs. CVF and MVD (r = -0.562 and 0.886, respectively; p < 0.001). Conclusion: The stress T1 response correlated well with the histopathologic measures in T2DM rabbits, indicating that it may serve as a sensitive CMD-related indicator in early T2DM.