Validated machine learning tools to distinguish immune checkpoint inhibitor, radiotherapy, COVID-19 and other infective pneumonitis

BACKGROUND

Pneumonitis is a well-described, potentially disabling, or fatal adverse effect associated with both immune checkpoint inhibitors (ICI) and thoracic radiotherapy. Accurate differentiation between checkpoint inhibitor pneumonitis (CIP) radiation pneumonitis (RP), and infective pneumonitis (IP) is crucial for swift, appropriate, and tailored management to achieve optimal patient outcomes. However, correct diagnosis is often challenging, owing to overlapping clinical presentations and radiological patterns.

METHODS

In this multi-centre study of 455 patients, we used machine learning with radiomic features extracted from chest CT imaging to develop and validate five models to distinguish CIP and RP from COVID-19, non-COVID-19 infective pneumonitis, and each other. Model performance was compared to that of two radiologists.

RESULTS

Models to distinguish RP from COVID-19, CIP from COVID-19 and CIP from non-COVID-19 IP out-performed radiologists (test set AUCs of 0.92 vs 0.8 and 0.8; 0.68 vs 0.43 and 0.4; 0.71 vs 0.55 and 0.63 respectively). Models to distinguish RP from non-COVID-19 IP and CIP from RP were not superior to radiologists but demonstrated modest performance, with test set AUCs of 0.81 and 0.8 respectively. The CIP vs RP model performed less well on patients with prior exposure to both ICI and radiotherapy (AUC 0.54), though the radiologists also had difficulty distinguishing this test cohort (AUC values 0.6 and 0.6).

CONCLUSION

Our results demonstrate the potential utility of such tools as a second or concurrent reader to support oncologists, radiologists, and chest physicians in cases of diagnostic uncertainty. Further research is required for patients with exposure to both ICI and thoracic radiotherapy.

Diagnosis and management of pneumonitis following chemoradiotherapy and immunotherapy in stage III non-small cell lung cancer

BACKGROUND

In inoperable stage III NSCLC, the standard of care is chemoradiotherapy and adjuvant durvalumab (IO) for 12 months. Pneumonitis is the commonest toxicity leading to discontinuation of IO. A failure to distinguish between expected radiation-induced changes, IO pneumonitis and infection can lead to unnecessary durvalumab discontinuation. We investigated the use of a structured multidisciplinary review of CT-scans, radiation dose distributions and clinical symptoms for the diagnosis of IO pneumonitis.

METHODS

A retrospective study was conducted at an academic medical center for patients treated for stage III NSCLC with chemoradiotherapy and adjuvant durvalumab between 2018 and 2021. An experienced thoracic radiologist reviewed baseline and follow-up chest CT-scans, systematically scored radiological features suspected for pneumonitis using a published classification system (Veiga C, Radioth Oncol 2018), and had access to screenshots of radiation dose distributions. Next, two experienced thoracic oncologists reviewed each patients' case record, CT-scans and radiation fields. A final consensus diagnosis incorporating views of expert clinicians and the radiologist was made.

RESULTS

Among the 45 included patients, 14/45 (31.1%) had a pneumonitis scored in patient records and durvalumab was discontinued in 11/45 cases (24.4%). Review by the radiologist led to a diagnosis of immune-related pneumonitis only in 6/45 patients (13.3%). Review by pulmonary oncologists led to a diagnosis of immune-related pneumonitis in only 4/45 patients (8.9%). In addition a suspicion of an immune-related pneumonitis was rejected in 3 separate patients (6.7%), after the thoracic oncologists had reviewed the patients' radiation fields.

CONCLUSIONS

In patients treated using the PACIFIC regimen, multidisciplinary assessment of CT-scans, radiation doses and patient symptoms, resulted in fewer diagnoses of immune-related pneumonitis (8.9%). Our study underscores the challenges in accurately diagnosing either IO-related or radiation pneumonitis in patients undergoing adjuvant immunotherapy after chemoradiotherapy and highlights the need for multidisciplinary review in order to avoid inappropriate cessation of adjuvant IO.

Development and Validation of a Machine Learning-Based Model Using CT Radiomics for Predicting Immune Checkpoint Inhibitor-related Pneumonitis in Patients With NSCLC Receiving Anti-PD1 Immunotherapy: A Multicenter Retrospective CaseControl Study

RATIONALE AND OBJECTIVES

This study aimed to develop and evaluate a radiomics-based model combined with clinical and qualitative radiological (semantic feature [SF]) features to predict immune checkpoint inhibitor-related pneumonitis (CIP) in patients with non-small cell lung cancer (NSCLC) treated with programmed cell death protein 1 inhibitors.

MATERIALS AND METHODS

This was a multicenter retrospective casecontrol study conducted from January 1, 2018, to December 31, 2022, at three centers. Patients with NSCLC treated with anti-PD1 were enrolled and randomly divided into two groups (7:3): training (n = 95) and validation (n = 39). Logistic regression (LR) and support vector machine (SVM) algorithms were used to transform features into the models.

RESULTS

The study comprised 134 participants from three independent centers (male, 114/134, 85%; mean [±standard deviation] age, 63.92 [±7.9] years). The radiomics score (RS) models built based on the LR and SVM algorithms could accurately predict CIP (area under the receiver operating characteristics curve [AUC], 0.860 [0.780, 0.939] and 0.861 [0.781, 0.941], respectively). The AUCs for the RS-clinic-SF combined model were 0.903 (0.839, 0.967) and 0.826 (0.688, 0.964) in the training and validation cohorts, respectively. Decision curve analysis showed that the combined models achieved high clinical net benefit across the majority of the range of reasonable threshold probabilities.

CONCLUSION

This study demonstrated that the combined model constructed by the identified features of RS, clinical features, and SF has the potential to precisely predict CIP. The RS-clinic-SF combined model has the potential to be used more widely as a practical tool for the noninvasive prediction of CIP to support individualized treatment planning.

Lung ultrasound and procalcitonin, improving antibiotic management and avoiding radiation exposure in pediatric critical patients with bacterial pneumonia: a randomized clinical trial

BACKGROUND

Pneumonia is a major public health problem with an impact on morbidity and mortality. Its management still represents a challenge. The aim was to determine whether a new diagnostic algorithm combining lung ultrasound (LUS) and procalcitonin (PCT) improved pneumonia management regarding antibiotic use, radiation exposure, and associated costs, in critically ill pediatric patients with suspected bacterial pneumonia (BP).

METHODS

Randomized, blinded, comparative effectiveness clinical trial. Children < 18y with suspected BP admitted to the PICU from September 2017 to December 2019, were included. PCT was determined at admission. Patients were randomized into the experimental group (EG) and control group (CG) if LUS or chest X-ray (CXR) were done as the first image test, respectively. Patients were classified: 1.LUS/CXR not suggestive of BP and PCT < 1 ng/mL, no antibiotics were recommended; 2.LUS/CXR suggestive of BP, regardless of the PCT value, antibiotics were recommended; 3.LUS/CXR not suggestive of BP and PCT > 1 ng/mL, antibiotics were recommended.

RESULTS

194 children were enrolled, 113 (58.2%) females, median age of 134 (IQR 39-554) days. 96 randomized into EG and 98 into CG. 1. In 75/194 patients the image test was not suggestive of BP with PCT < 1 ng/ml; 29/52 in the EG and 11/23 in the CG did not receive antibiotics. 2. In 101 patients, the image was suggestive of BP; 34/34 in the EG and 57/67 in the CG received antibiotics. Statistically significant differences between groups were observed when PCT resulted < 1 ng/ml (p = 0.01). 3. In 18 patients the image test was not suggestive of BP but PCT resulted > 1 ng/ml, all of them received antibiotics. A total of 0.035 mSv radiation/patient was eluded. A reduction of 77% CXR/patient was observed. LUS did not significantly increase costs.

CONCLUSIONS

Combination of LUS and PCT showed no risk of mistreating BP, avoided radiation and did not increase costs. The algorithm could be a reliable tool for improving pneumonia management.

CLINICAL TRIAL REGISTRATION

NCT04217980.

Adhesion molecule-targeted magnetic particle imaging nanoprobe for visualization of inflammation in acute lung injury

PURPOSE

Uncontrolled intra-alveolar inflammation is a central pathogenic feature, and its severity translates into a valid prognostic indicator of acute lung injury (ALI). Unfortunately, current clinical imaging approaches are unsuitable for visualizing and quantifying intra-alveolar inflammation. This study aimed to construct a small-sized vascular cell adhesion molecule-1 (VCAM-1)-targeted magnetic particle imaging (MPI) nanoprobe (ESPVPN) to visualize and accurately quantify intra-alveolar inflammation at the molecular level.

METHODS

ESPVPN was engineered by conjugating a peptide (VHPKQHRGGSK(Cy7)GC) onto a polydopamine-functionalized superparamagnetic iron oxide core. The MPI performance, targeting, and biosafety of the ESPVPN were characterized. VCAM-1 expression in HUVECs and mouse models was evaluated by western blot. The degree of inflammation and distribution of VCAM-1 in the lungs were assessed using histopathology. The expression of pro-inflammatory markers and VCAM-1 in lung tissue lysates was measured using ELISA. After intravenous administration of ESPVPN, MPI and CT imaging were used to analyze the distribution of ESPVPN in the lungs of the LPS-induced ALI models.

RESULTS

The small-sized (~10 nm) ESPVPN exhibited superior MPI performance compared to commercial MagImaging® and Vivotrax, and ESPVPN had effective targeting and biosafety. VCAM-1 was highly expressed in LPS-induced ALI mice. VCAM-1 expression was positively correlated with the LPS-induced dose (R = 0.9381). The in vivo MPI signal showed positive correlations with both VCAM-1 expression (R = 0.9186) and representative pro-inflammatory markers (MPO, TNF-α, IL-6, IL-8, and IL-1β, R > 0.7).

CONCLUSION

ESPVPN effectively targeted inflammatory lungs and combined the advantages of MPI quantitative imaging to visualize and evaluate the degree of ALI inflammation.

Radiographic findings of adenoviral pneumonia in children

OBJECTIVE

Adenovirus pneumonia is a common cause of community-acquired pneumonia in children and can mimic bacterial pneumonia, but there are few publications on its radiographic features. This study has evaluated the chest radiography findings of community-acquired adenovirus pneumonia in children. The frequency of radiological findings mimicking bacterial pneumonia was investigated. The clinical features of patients with adenovirus pneumonia possessing radiological findings mimicking bacterial pneumonia were also evaluated.

MATERIALS AND METHODS

The chest radiographs of patients diagnosed with adenovirus pneumonia were retrospectively reviewed. The chest radiographs were interpreted independently by a pediatric infectious disease specialist and a pediatric radiologist. Chest radiography findings mimicking bacterial pneumonia (bacterial-like) were specified as consolidation +/- pleural effusion. Other findings on chest radiography or a completely normal chest X-ray were specified as findings that were compatible with "typical viral pneumonia".

RESULTS

A total of 1407 patients were positive for adenovirus with respiratory multiplex PCR. The 219 patients who met the study criteria were included in the study. Chest radiographs were normal in 58 (26.5 %) patients. The chest radiograph findings mimicked bacterial pneumonia in 41 (18.7 %) patients.

CONCLUSION

Adenovirus pneumonia occurs predominantly in children aged five years and younger, as with other viral pneumonias. The radiographic findings in adenovirus pneumonia are predominantly those seen in viral pneumonia. Increasing age and positivity for only adenovirus without other viruses on respiratory multiplex PCR were associated with the chest radiograph being more likely to be "bacterial-like". Adenovirus may lead to lobar/segmental consolidation at a rate that is not very rare.

[The diagnostic value of lung ultrasound in children with community-acquired pneumonia]

Objective: To investigate the diagnostic value of lung ultrasound in hospitalized children with community-acquired pneumonia (CAP). Methods: In the cross-sectional study, a total of 422 children with CAP who were hospitalized in the Second Affiliated Hospital and Yuying Children's Hospital of Wenzhou Medical University, from February 2021 to August 2022 and completed lung ultrasound examination within 48 hours after admission were enrolled. The clinical characteristics, lung ultrasound and chest CT were collected. The patients were divided into two groups according to the signs of pneumonia indicated by chest CT, and the signs of lung ultrasound with diagnostic value were screened according to the signs of pneumonia indicated by chest CT by least absolute shrinkage and selection operator (Lasso) regression. According to severity of the disease, the children were divided into the severe group and the mild group, and the differences of lung ultrasound signs between the two groups were compared. Kruskal-Wallis test, Fisher's exact test was selected for comparison between groups. Random forest classifier wes used to evaluate the value of lung ultrasound in the diagnosis of CAP and prediction of severe pneumonia in children. The receiver operating characteristic curve was used to evaluate the prediction effect. Use DeLong test to compare the area under the curve. Results: Among the 422 cases of CAP, there were 258 males and 164 females, and the age of onset was 2.8 (1.3, 4.3) years. The confluent B-line, consolidation and pleural effusion detected by lung ultrasound were 309 cases (73.2%), 232 cases (55.0%) and 16 cases (3.8%), respectively, and the size of consolidation was 3.0 (0, 11.0) mm. One hundred and ten children (26.1%) with CAP completed chest CT. There were 90 cases with signs of pneumonia in chest CT and 20 cases without signs of pneumonia. Lasso was used for feature selection.Lung consolidation (OR=2.46), bilateral lung consolidation (OR=1.16) and confluent B-line (OR=1.34) were the main index. With random forest classifier, the accuracy of models using full variables and Lasso-selected variables were 0.79 (95%CI 0.70-0.86) and 0.79 (95%CI 0.70-0.86), the sensitivity were 0.81 and 0.81, and the specificity were 0.75 and 0.70, and the area under curve were 0.87 (95%CI 0.81-0.94, P<0.001) and 0.84 (95%CI 0.76-0.91, P<0.001), respectively. There were 97 cases in severe group and 325 cases in mild group. Compared with the mild group, the detection rate of consolidation, multiple consolidation, the size of consolidation and the size of consolidation was adjusted by body surface area (consolidation size/body surface area) in severe group were higher (66 cases (68.0%) vs. 166 cases (51.1%), 42 cases (43.3%) vs. 93 cases (28.6%), 8.0 (0, 17.0) vs. 1.0 (0, 9.0) mm, 12.5 (0, 24.6) vs. 2.1 (0, 17.6), χ2=8.59, 9.98, Z=14.40, 12.79, all P<0.05). Using lung ultrasound lung consolidation size and consolidation size/body surface area to predict the severe CAP, the optimal cut-off value were 6.7 mm and 10.2, the accuracy was 0.80 (95%CI 0.75-0.83) and 0.89 (95%CI 0.86-0.92), the sensitivity was 0.99 and 0.99, the specificity was 0.14 and 0.56, respectively, and the area under the curve was 0.66 (95%CI 0.60-0.72, P<0.001) and 0.76 (95%CI 0.70-0.83, P<0.001), respectively. The area under the curve of consolidation size/body surface area was higher than that of consolidation size (Z=5.50, P<0.001). Conclusions: Consolidation and confluent B-line, are important index for lung ultrasound diagnosis of CAP in children. The actual consolidation size adjusted by body surface area is superior to the size of consolidation in predicting severe CAP.

Enhancing pediatric pneumonia diagnosis through masked autoencoders

Pneumonia, an inflammatory lung condition primarily triggered by bacteria, viruses, or fungi, presents distinctive challenges in pediatric cases due to the unique characteristics of the respiratory system and the potential for rapid deterioration. Timely diagnosis is crucial, particularly in children under 5, who have immature immune systems, making them more susceptible to pneumonia. While chest X-rays are indispensable for diagnosis, challenges arise from subtle radiographic findings, varied clinical presentations, and the subjectivity of interpretations, especially in pediatric cases. Deep learning, particularly transfer learning, has shown promise in improving pneumonia diagnosis by leveraging large labeled datasets. However, the scarcity of labeled data for pediatric chest X-rays presents a hurdle in effective model training. To address this challenge, we explore the potential of self-supervised learning, focusing on the Masked Autoencoder (MAE). By pretraining the MAE model on adult chest X-ray images and fine-tuning the pretrained model on a pediatric pneumonia chest X-ray dataset, we aim to overcome data scarcity issues and enhance diagnostic accuracy for pediatric pneumonia. The proposed approach demonstrated competitive performance an AUC of 0.996 and an accuracy of 95.89% in distinguishing between normal and pneumonia. Additionally, the approach exhibited high AUC values (normal: 0.997, bacterial pneumonia: 0.983, viral pneumonia: 0.956) and an accuracy of 93.86% in classifying normal, bacterial pneumonia, and viral pneumonia. This study also investigated the impact of different masking ratios during pretraining and explored the labeled data efficiency of the MAE model, presenting enhanced diagnostic capabilities for pediatric pneumonia.

A deep learning feature extraction-based hybrid approach for detecting pediatric pneumonia in chest X-ray images

Pneumonia is a disease caused by bacteria, viruses, and fungi that settle in the alveolar sacs of the lungs and can lead to serious health complications in humans. Early detection of pneumonia is necessary for early treatment to manage and cure the disease. Recently, machine learning-based pneumonia detection methods have focused on pneumonia in adults. Machine learning relies on manual feature engineering, whereas deep learning can automatically detect and extract features from data. This study proposes a deep learning feature extraction-based hybrid approach that combines deep learning and machine learning to detect pediatric pneumonia, which is difficult to standardize. The proposed hybrid approach enhances the accuracy of detecting pediatric pneumonia and simplifies the approach by eliminating the requirement for advanced feature extraction. The experiments indicate that the hybrid approach using a Medium Neural Network based on AlexNet feature extraction achieved a 97.9% accuracy rate and 98.0% sensitivity rate. The results show that the proposed approach achieved higher accuracy rates than state-of-the-art approaches.

Lung ultrasound: A potential tool in the diagnosis of ventilator-associated pneumonia in pediatric intensive care units

PURPOSE

Ventilator-associated pneumonia (VAP) is a common healthcare-associated infection in pediatric intensive care unit (PICU), increasing mortality, antibiotics use and duration of ventilation and hospitalization. VAP diagnosis is based on clinical and chest X-ray (CXR) signs defined by the 2018 Center for Disease Control (gold standard). However, CXR induces repetitive patients' irradiation and technical limitations. This study aimed to investigate if lung ultrasound (LUS) can substitute CXR in the VAP diagnosis.

METHODS

A monocentric and prospective study was conducted in a French tertiary care hospital. Patients under 18-year-old admitted to PICU between November 2018 and July 2020 with invasive mechanical ventilation for more than 48 h were included. The studied LUS signs were consolidations, dynamic air bronchogram, subpleural consolidations (SPC), B-lines, and pleural effusion. The diagnostic values of each sign associated with clinical signs (cCDC) were compared to the gold standard approach. LUS, chest X-ray, and clinical score were performed daily.

RESULTS

Fifty-seven patients were included. The median age was 8 [3-34] months. Nineteen (33%) children developed a VAP. In patients with VAP, B-Lines, and consolidations were highly frequent (100 and 68.8%) and, associated with cCDC, were highly sensitive (100 [79-100] % and 88 [62-98] %, respectively) and specific (95.5 [92-98] % and 98 [95-99] %, respectively). Other studied signs, including SPC, showed high specificity (>97%) but low sensibility (<50%).

CONCLUSION

LUS seems to be a powerful tool for VAP diagnosis in children with a clinical suspicion, efficiently substituting CXR, and limiting children's exposure to ionizing radiations.

Accuracy and inter-rater agreement among practitioners using quick thoracic ultrasonography to diagnose calf pneumonia

BACKGROUND

Thoracic ultrasonography (TUS) is a commonly used tool for on-farm detection of pneumonia in calves. Different scanning methods have been described, but the performance of novice practitioners after training has not been documented.

METHODS

In this study, 38 practitioners performed quick TUS (qTUS) on 18-23 calves each. Pneumonia was defined as lung consolidation 1 cm or more in depth. Diagnostic parameters (accuracy [Acc], sensitivity [Se] and specificity [Sp]) were compared to those of an experienced operator. Cohen's kappa and Krippendorff's alpha (Kalpha) were determined. The potential effects of training and exam sessions on performance were evaluated.

RESULTS

The average relative Se and Sp were 0.66 (standard deviation [SD] = 0.26; minimum [Min.]-Maximum [Max.] = 0-1) and 0.71 (SD = 0.19; Min.-Max. = 0.25-1), respectively. The average relative Acc was 0.73 (SD = 0.11; Min.-Max. = 0.52-0.96). Over all sessions, Cohen's kappa averaged 0.40 (SD = 0.24; Min.-Max. = 0.014-0.90) and Kalpha was 0.24 (95% confidence interval [CI]: 0.20-0.27), indicating 'fair' agreement. Calf age and housing influenced Se and Sp. Supervised practical training improved Se by 17.5% (95% CI: 0.01-0.34).

LIMITATIONS

The separate effects of calf age and housing could not be determined.

CONCLUSION

This study showed that qTUS, like any other clinical skill, has a learning curve, and variability in performance can be substantial. Adequate training and certification of one's skill are recommended to assure good diagnostic accuracy.

Radiographic Imaging of Community-Acquired Pneumonia: A Case-Based Review

The chest radiograph is the most common imaging examination performed in most radiology departments, and one of the more common indications for these studies is suspected infection. Radiologists must therefore be aware of less common radiographic patterns of pulmonary infection if they are to add value in the interpretation of chest radiographs for this indication. This review uses a case-based format to illustrate a range of imaging findings that can be associated with acute pulmonary infection and highlight findings that should prompt investigation for diseases other than community-acquired pneumonia to prevent misdiagnosis and delays in appropriate management.

Evaluation of the diagnostic performance of physician lung ultrasound versus chest radiography for pneumonia diagnosis in a peri-urban South African cohort

BACKGROUND

Lung ultrasound (US), which is radiation-free and cheaper than chest radiography (CXR), may be a useful modality for the diagnosis of pediatric pneumonia, but there are limited data from low- and middle-income countries.

OBJECTIVES

The aim of this study was to evaluate the diagnostic performance of non-radiologist, physician-performed lung US compared to CXR for pneumonia in children in a resource-constrained, African setting.

MATERIALS AND METHODS

Children under 5 years of age enrolled in a South African birth cohort study, the Drakenstein Child Health Study, who presented with clinically defined pneumonia and had a CXR performed also had a  lung US performed by a study doctor. Each modality was reported by two readers, using standardized methodology. Agreement between modalities, accuracy (sensitivity and specificity) of lung US and inter-rater agreement were assessed. Either consolidation or any abnormality (consolidation or interstitial picture) was considered as endpoints. In the 98 included cases (median age: 7.2 months; 53% male; 69% hospitalized), prevalence was 37% vs. 39% for consolidation and 52% vs. 76% for any abnormality on lung US and CXR, respectively. Agreement between modalities was poor for consolidation (observed agreement=61%, Kappa=0.18, 95% confidence interval [95% CI]: - 0.02 to 0.37) and for any abnormality (observed agreement=56%, Kappa=0.10, 95% CI: - 0.07 to 0.28). Using CXR as the reference standard, sensitivity of lung US was low for consolidation (47%, 95% CI: 31-64%) or any abnormality (5%, 95% CI: 43-67%), while specificity was moderate for consolidation (70%, 95% CI: 57-81%), but lower for any abnormality (58%, 95% CI: 37-78%). Overall inter-observer agreement of CXR was poor (Kappa=0.25, 95% CI: 0.11-0.37) and was significantly lower than the substantial agreement of lung US (Kappa=0.61, 95% CI: 0.50-0.75). Lung US demonstrated better agreement than CXR for all categories of findings, showing a significant difference for consolidation (Kappa=0.72, 95% CI: 0.58-0.86 vs. 0.32, 95% CI: 0.13-0.51).

CONCLUSION

Lung US identified consolidation with similar frequency to CXR, but there was poor agreement between modalities. The significantly higher inter-observer agreement of LUS compared to CXR supports the utilization of lung US by clinicians in a low-resource setting.

Validating the accuracy of deep learning for the diagnosis of pneumonia on chest x-ray against a robust multimodal reference diagnosis: a post hoc analysis of two prospective studies

BACKGROUND

Artificial intelligence (AI) seems promising in diagnosing pneumonia on chest x-rays (CXR), but deep learning (DL) algorithms have primarily been compared with radiologists, whose diagnosis can be not completely accurate. Therefore, we evaluated the accuracy of DL in diagnosing pneumonia on CXR using a more robust reference diagnosis.

METHODS

We trained a DL convolutional neural network model to diagnose pneumonia and evaluated its accuracy in two prospective pneumonia cohorts including 430 patients, for whom the reference diagnosis was determined a posteriori by a multidisciplinary expert panel using multimodal data. The performance of the DL model was compared with that of senior radiologists and emergency physicians reviewing CXRs and that of radiologists reviewing computed tomography (CT) performed concomitantly.

RESULTS

Radiologists and DL showed a similar accuracy on CXR for both cohorts (p ≥ 0.269): cohort 1, radiologist 1 75.5% (95% confidence interval 69.1-80.9), radiologist 2 71.0% (64.4-76.8), DL 71.0% (64.4-76.8); cohort 2, radiologist 70.9% (64.7-76.4), DL 72.6% (66.5-78.0). The accuracy of radiologists and DL was significantly higher (p ≤ 0.022) than that of emergency physicians (cohort 1 64.0% [57.1-70.3], cohort 2 63.0% [55.6-69.0]). Accuracy was significantly higher for CT (cohort 1 79.0% [72.8-84.1], cohort 2 89.6% [84.9-92.9]) than for CXR readers including radiologists, clinicians, and DL (all p-values < 0.001).

CONCLUSIONS

When compared with a robust reference diagnosis, the performance of AI models to identify pneumonia on CXRs was inferior than previously reported but similar to that of radiologists and better than that of emergency physicians.

RELEVANCE STATEMENT

The clinical relevance of AI models for pneumonia diagnosis may have been overestimated. AI models should be benchmarked against robust reference multimodal diagnosis to avoid overestimating its performance.

TRIAL REGISTRATION

NCT02467192 , and NCT01574066 .

KEY POINT

• We evaluated an openly-access convolutional neural network (CNN) model to diagnose pneumonia on CXRs. • CNN was validated against a strong multimodal reference diagnosis. • In our study, the CNN performance (area under the receiver operating characteristics curve 0.74) was lower than that previously reported when validated against radiologists' diagnosis (0.99 in a recent meta-analysis). • The CNN performance was significantly higher than emergency physicians' (p ≤ 0.022) and comparable to that of board-certified radiologists (p ≥ 0.269).

Acute muscle wasting is associated with poor prognosis in older adults with severe community-acquired pneumonia

PURPOSE

To investigate the impact of acute muscle wasting on 90-day mortality in older patients with severe pneumonia using ultrasound and chest computed tomography (CT).

METHODS

Quadriceps muscle layer thickness was measured via ultrasound on days 1, 7, and 14, and cross-sectional area of the erector spinae muscle was assessed using chest CT on days 1 and 14 in patients aged ≥ 65 years old. The primary outcome was all-cause 90-day mortality. Receiver operating characteristic curves were conducted for muscle loss to predict 90-day mortality. Cox proportional hazard models and Kaplan-Meier survival curves were employed to evaluate the association between muscle loss and 90-day mortality.

RESULTS

Sixty-two patients were enrolled with median age of 80.2 years, 29 (46.8%) were men and 28 (45.2%) patients died. Muscle mass measured using ultrasound and CT decreased significantly from baseline to day 14 in the non-survivor group. Muscle loss assessed by ultrasound (with minimum and maximum pressure) and CT independently predicted all-cause 90-day mortality (adjusted hazard ratios = 1.497, 1.400 and 1.082; P < 0.001, P = 0.002, and P = 0.004; respectively), and cutoff values of muscle loss were 0.34 cm, 0.11 cm and 4.92 cm2, correspondingly. A higher muscle loss had an increased risk of 90-day mortality.

CONCLUSIONS

Acute muscle wasting assessed by ultrasound and chest CT persisted for 14 days and was an independent predictor of adverse outcomes in older patients with severe pneumonia. A greater decline in muscle mass was associated with a higher 90-day mortality risk.

Cost of Pediatric Pneumonia Episodes With or Without Chest Radiography

BACKGROUND AND OBJECTIVES

Despite its routine use, it is unclear whether chest radiograph (CXR) is a cost-effective strategy in the workup of community-acquired pneumonia (CAP) in the pediatric emergency department (ED). We sought to assess the costs of CAP episodes with and without CXR among children discharged from the ED.

METHODS

This was a retrospective cohort study within the Healthcare Cost and Utilization Project State ED and Inpatient Databases of children aged 3 months to 18 years with CAP discharged from any EDs in 8 states from 2014 to 2019. We evaluated total 28-day costs after ED discharge, including the index visit and subsequent care. Mixed-effects linear regression models adjusted for patient-level variables and illness severity were performed to evaluate the association between CXR and costs.

RESULTS

We evaluated 225c781 children with CAP, and 86.2% had CXR at the index ED visit. Median costs of the 28-day episodes, index ED visits, and subsequent visits were $314 (interquartile range [IQR] 208-497), $288 (IQR 195-433), and $255 (IQR 133-637), respectively. There was a $33 (95% confidence interval [CI] 22-44) savings over 28-days per patient for those who received a CXR compared with no CXR after adjusting for patient-level variables and illness severity. Costs during subsequent visits ($26 savings, 95% CI 16-36) accounted for the majority of the savings as compared with the index ED visit ($6, 95% CI 3-10).

CONCLUSIONS

Performance of CXR for CAP diagnosis is associated with lower costs when considering the downstream provision of care among patients who require subsequent health care after initial ED discharge.

Interpretable CNN-Multilevel Attention Transformer for Rapid Recognition of Pneumonia From Chest X-Ray Images

Chest imaging plays an essential role in diagnosing and predicting patients with COVID-19 with evidence of worsening respiratory status. Many deep learning-based approaches for pneumonia recognition have been developed to enable computer-aided diagnosis. However, the long training and inference time makes them inflexible, and the lack of interpretability reduces their credibility in clinical medical practice. This paper aims to develop a pneumonia recognition framework with interpretability, which can understand the complex relationship between lung features and related diseases in chest X-ray (CXR) images to provide high-speed analytics support for medical practice. To reduce the computational complexity to accelerate the recognition process, a novel multi-level self-attention mechanism within Transformer has been proposed to accelerate convergence and emphasize the task-related feature regions. Moreover, a practical CXR image data augmentation has been adopted to address the scarcity of medical image data problems to boost the model's performance. The effectiveness of the proposed method has been demonstrated on the classic COVID-19 recognition task using the widespread pneumonia CXR image dataset. In addition, abundant ablation experiments validate the effectiveness and necessity of all of the components of the proposed method.

Efficient pneumonia detection using Vision Transformers on chest X-rays

Pneumonia is a widespread and acute respiratory infection that impacts people of all ages. Early detection and treatment of pneumonia are essential for avoiding complications and enhancing clinical results. We can reduce mortality, improve healthcare efficiency, and contribute to the global battle against a disease that has plagued humanity for centuries by devising and deploying effective detection methods. Detecting pneumonia is not only a medical necessity but also a humanitarian imperative and a technological frontier. Chest X-rays are a frequently used imaging modality for diagnosing pneumonia. This paper examines in detail a cutting-edge method for detecting pneumonia implemented on the Vision Transformer (ViT) architecture on a public dataset of chest X-rays available on Kaggle. To acquire global context and spatial relationships from chest X-ray images, the proposed framework deploys the ViT model, which integrates self-attention mechanisms and transformer architecture. According to our experimentation with the proposed Vision Transformer-based framework, it achieves a higher accuracy of 97.61%, sensitivity of 95%, and specificity of 98% in detecting pneumonia from chest X-rays. The ViT model is preferable for capturing global context, comprehending spatial relationships, and processing images that have different resolutions. The framework establishes its efficacy as a robust pneumonia detection solution by surpassing convolutional neural network (CNN) based architectures.

PLUS-IS-LESS project: Procalcitonin and Lung UltraSonography-based antibiotherapy in patients with Lower rESpiratory tract infection in Swiss Emergency Departments: study protocol for a pragmatic stepped-wedge cluster-randomized trial

BACKGROUND

Lower respiratory tract infections (LRTIs) are among the most frequent infections and a significant contributor to inappropriate antibiotic prescription. Currently, no single diagnostic tool can reliably identify bacterial pneumonia. We thus evaluate a multimodal approach based on a clinical score, lung ultrasound (LUS), and the inflammatory biomarker, procalcitonin (PCT) to guide prescription of antibiotics. LUS outperforms chest X-ray in the identification of pneumonia, while PCT is known to be elevated in bacterial and/or severe infections. We propose a trial to test their synergistic potential in reducing antibiotic prescription while preserving patient safety in emergency departments (ED).

METHODS

The PLUS-IS-LESS study is a pragmatic, stepped-wedge cluster-randomized, clinical trial conducted in 10 Swiss EDs. It assesses the PLUS algorithm, which combines a clinical prediction score, LUS, PCT, and a clinical severity score to guide antibiotics among adults with LRTIs, compared with usual care. The co-primary endpoints are the proportion of patients prescribed antibiotics and the proportion of patients with clinical failure by day 28. Secondary endpoints include measurement of change in quality of life, length of hospital stay, antibiotic-related side effects, barriers and facilitators to the implementation of the algorithm, cost-effectiveness of the intervention, and identification of patterns of pneumonia in LUS using machine learning.

DISCUSSION

The PLUS algorithm aims to optimize prescription of antibiotics through improved diagnostic performance and maximization of physician adherence, while ensuring safety. It is based on previously validated tests and does therefore not expose participants to unforeseeable risks. Cluster randomization prevents cross-contamination between study groups, as physicians are not exposed to the intervention during or before the control period. The stepped-wedge implementation of the intervention allows effect calculation from both between- and within-cluster comparisons, which enhances statistical power and allows smaller sample size than a parallel cluster design. Moreover, it enables the training of all centers for the intervention, simplifying implementation if the results prove successful. The PLUS algorithm has the potential to improve the identification of LRTIs that would benefit from antibiotics. When scaled, the expected reduction in the proportion of antibiotics prescribed has the potential to not only decrease side effects and costs but also mitigate antibiotic resistance.

TRIAL REGISTRATION

This study was registered on July 19, 2022, on the ClinicalTrials.gov registry using reference number: NCT05463406.

TRIAL STATUS

Recruitment started on December 5, 2022, and will be completed on November 3, 2024. Current protocol version is version 3.0, dated April 3, 2023.

Photon Counting Computed Tomography with the Radiation Dose of a Chest X-Ray: Feasibility and Diagnostic Yield

INTRODUCTION

Photon counting (PC) detectors allow a reduction of the radiation dose in CT. Chest X-ray (CXR) is known to have a low sensitivity and specificity for detection of pneumonic infiltrates. The aims were to establish an ultra-low-dose CT (ULD-CT) protocol at a PC-CT with the radiation dose comparable to the dose of a CXR and to evaluate its clinical yield in patients with suspicion of pneumonia.

METHODS

A ULD-CT protocol was established with the aim to meet the radiation dose of a CXR. In this retrospective study, all adult patients who received a ULD-CT of the chest with suspected pneumonia were included. Radiation exposure of ULD-CT and CXR was calculated. The clinical significance (new diagnosis, change of therapy, additional findings) and limitations were evaluated by a radiologist and a pulmonologist considering previous CXR and clinical data.

RESULTS

Twenty-seven patients (70% male, mean age 68 years) were included. With our ULD-CT protocol, the radiation dose of a CXR could be reached (mean radiation exposure 0.11 mSv). With ULD-CT, the diagnosis changed in 11 patients (41%), there were relevant additional findings in 4 patients (15%), an infiltrate (particularly fungal infiltrate under immunosuppression) could be ruled out with certainty in 10 patients (37%), and the therapy changed in 10 patients (37%). Two patients required an additional CT with contrast medium to rule out a pulmonary embolism or pleural empyema.

CONCLUSIONS

With ULD-CT, the radiation dose of a CXR could be reached while the clinical impact is higher with change in diagnosis in 41%.

Pneumonia detection based on RSNA dataset and anchor-free deep learning detector

Pneumonia is a highly lethal disease, and research on its treatment and early screening tools has received extensive attention from researchers. Due to the maturity and cost reduction of chest X-ray technology, and with the development of artificial intelligence technology, pneumonia identification based on deep learning and chest X-ray has attracted attention from all over the world. Although the feature extraction capability of deep learning is strong, existing deep learning object detection frameworks are based on pre-defined anchors, which require a lot of tuning and experience to guarantee their excellent results in the face of new applications or data. To avoid the influence of anchor settings in pneumonia detection, this paper proposes an anchor-free object detection framework and RSNA dataset based on pneumonia detection. First, a data enhancement scheme is used to preprocess the chest X-ray images; second, an anchor-free object detection framework is used for pneumonia detection, which contains a feature pyramid, two-branch detection head, and focal loss. The average precision of 51.5 obtained by Intersection over Union (IoU) calculation shows that the pneumonia detection results obtained in this paper can surpass the existing classical object detection framework, providing an idea for future research and exploration.

The correlation between lung ultrasound scores and outcomes of high-flow nasal cannula therapy in infants with severe pneumonia

OBJECTIVE

The study aimed to explore the effectiveness of bedside lung ultrasound (LUS) combined with the PaO2/FiO2 (P/F) ratio in evaluating the outcomes of high-flow nasal cannula (HFNC) therapy in infants with severe pneumonia.

METHODS

This retrospective study analyzed the clinical data of 150 infants diagnosed with severe pneumonia and treated with HFNC therapy at our hospital from January 2021 to December 2021. These patients were divided into two groups based on their treatment outcomes: the HFNC success group (n = 112) and the HFNC failure group (n = 38). LUS was utilized to evaluate the patients' lung conditions, and blood gas results were recorded for both groups upon admission and after 12 h of HFNC therapy.

RESULTS

At admission, no significant differences were observed between the two groups in terms of age, gender, respiratory rate, partial pressure of oxygen, and partial pressure of carbon dioxide. However, the P/F ratios at admission and after 12 h of HFNC therapy were significantly lower in the HFNC failure group (193.08 ± 49.14, 228.63 ± 80.17, respectively) compared to the HFNC success group (248.51 ± 64.44, 288.93 ± 57.17, respectively) (p < 0.05). Likewise, LUS scores at admission and after 12 h were significantly higher in the failure group (18.42 ± 5.3, 18.03 ± 5.36, respectively) than in the success group (15.09 ± 4.66, 10.71 ± 3.78, respectively) (p < 0.05). Notably, in the success group, both P/F ratios and LUS scores showed significant improvement after 12 h of HFNC therapy, a trend not observed in the failure group. Multivariate regression analysis indicated that lower P/F ratios and higher LUS scores at admission and after 12 h were predictive of a greater risk of HFNC failure. ROC analysis demonstrated that an LUS score > 20.5 at admission predicted HFNC therapy failure with an AUC of 0.695, a sensitivity of 44.7%, and a specificity of 91.1%. A LUS score > 15.5 after 12 h of HFNC therapy had an AUC of 0.874, with 65.8% sensitivity and 89.3% specificity. An admission P/F ratio < 225.5 predicted HFNC therapy failure with an AUC of 0.739, 60.7% sensitivity, and 71.1% specificity, while a P/F ratio < 256.5 after 12 h of HFNC therapy had an AUC of 0.811, 74.1% sensitivity, and 73.7% specificity.

CONCLUSION

Decreased LUS scores and increased P/F ratio demonstrate a strong correlation with successful HFNC treatment outcomes in infants with severe pneumonia. These findings may provide valuable support for clinicians in managing such cases.

Pre-existing Interstitial Lung Abnormalities and Immune Checkpoint Inhibitor-Related Pneumonitis in Solid Tumors: A Retrospective Analysis

BACKGROUND

Immune checkpoint inhibitors (ICIs) have demonstrated efficacy over previous cytotoxic chemotherapies in clinical trials among various tumors. Despite their favorable outcomes, they are associated with a unique set of toxicities termed as immune-related adverse events (irAEs). Among the toxicities, ICI-related pneumonitis has poor outcomes with little understanding of its risk factors. This retrospective study aimed to investigate whether pre-existing interstitial lung abnormality (ILA) is a potential risk factor for ICI-related pneumonitis.

MATERIALS AND METHODS

Patients with non-small cell lung cancer, malignant melanoma, renal cell carcinoma, and gastric cancer, who was administered either nivolumab, pembrolizumab, or atezolizumab between September 2014 and January 2019 were retrospectively reviewed. Information on baseline characteristics, computed tomography findings before administration of ICIs, clinical outcomes, and irAEs were collected from their medical records. Pre-existing ILA was categorized based on previous studies.

RESULTS

Two-hundred-nine patients with a median age of 68 years were included and 23 (11.0%) developed ICI-related pneumonitis. While smoking history and ICI agents were associated with ICI-related pneumonitis (P = .005 and .044, respectively), the categories of ILA were not associated with ICI-related pneumonitis (P = .428). None of the features of lung abnormalities were also associated with ICI-related pneumonitis. Multivariate logistic analysis indicated that smoking history was the only significant predictor of ICI-related pneumonitis (P = .028).

CONCLUSION

This retrospective study did not demonstrate statistically significant association between pre-existing ILA and ICI-related pneumonitis, nor an association between radiologic features of ILA and ICI-related pneumonitis. Smoking history was independently associated with ICI-related pneumonitis. Further research is warranted for further understanding of the risk factors of ICI-related pneumonitis.

Attention-enhanced architecture for improved pneumonia detection in chest X-ray images

In this paper, we propose an attention-enhanced architecture for improved pneumonia detection in chest X-ray images. A unique attention mechanism is integrated with ResNet to highlight salient features crucial for pneumonia detection. Rigorous evaluation demonstrates that our attention mechanism significantly enhances pneumonia detection accuracy, achieving a satisfactory result of 96% accuracy. To address the issue of imbalanced training samples, we integrate an enhanced focal loss into our architecture. This approach assigns higher weights to minority classes during training, effectively mitigating data imbalance. Our model's performance significantly improves, surpassing that of traditional approaches such as the pretrained ResNet-50 model. Our attention-enhanced architecture thus presents a powerful solution for pneumonia detection in chest X-ray images, achieving an accuracy of 98%. By integrating enhanced focal loss, our approach effectively addresses imbalanced training sample. Comparative analysis underscores the positive impact of our model's spatial and channel attention modules. Overall, our study advances pneumonia detection in medical imaging and underscores the potential of attention-enhanced architectures for improved diagnostic accuracy and patient outcomes. Our findings offer valuable insights into image diagnosis and pneumonia prevention, contributing to future research in medical imaging and machine learning.

The role of lung ultrasound for detecting atelectasis, consolidation, and/or pneumonia in the adult cardiac surgery population: A scoping review of the literature

OBJECTIVES

Postoperative pulmonary complications (PPCs) frequently occur after cardiac surgery and may lead to adverse patient outcomes. Traditional diagnostic tools such as auscultation or chest x-ray have inferior diagnostic accuracy compared to the gold standard (chest computed tomography). Lung ultrasound (LUS) is an emerging area of research combating these issues. However, no review has employed a formal search strategy to examine the role of LUS in identifying the specific PPCs of atelectasis, consolidation, and/or pneumonia or investigated the ability of LUS to predict these complications in this cohort. The objective of this study was to collate and present evidence for the use of LUS in the adult cardiac surgery population to specifically identify atelectasis, consolidation, and/or pneumonia.

REVIEW METHOD USED

A scoping review of the literature was completed using predefined search terms across six databases which identified 1432 articles. One additional article was included from reviewing reference lists. Six articles met the inclusion criteria, providing sufficient data for the final analysis.

DATA SOURCES

Six databases were searched: MEDLINE, Embase, CINAHL, Scopus, CENTRAL, and PEDro. This review was not registered.

REVIEW METHODS

The review followed the PRISMA Extension for Scoping Reviews.

RESULTS

Several LUS methodologies were reported across studies. Overall, LUS outperformed all other included bedside diagnostic tools, with superior diagnostic accuracy in identifying atelectasis, consolidation, and/or pneumonia. Incidences of PPCs tended to increase with each subsequent timepoint after surgery and were better identified with LUS than all other assessments. A change in diagnosis occurred at a rate of 67% with the inclusion of LUS and transthoracic echocardiography in one study. Pre-established assessment scores were improved by substituting chest x-rays with LUS scans.

CONCLUSION

The results of this scoping review support the use of LUS as a diagnostic tool after cardiac surgery; however, they also highlighted a lack of consistent methodologies used. Future research is required to determine the optimal methodology for LUS in diagnosing PPCs in this cohort and to determine whether LUS possesses the ability to predict these complications and guide proactive respiratory supports after extubation.

A Novel Approach to the Technique of Lung Region Segmentation Based on a Deep Learning Model to Diagnose COVID-19 X-ray Images

BACKGROUND

The novel coronavirus pandemic has caused a global health crisis, placing immense strain on healthcare systems worldwide. Chest X-ray technology has emerged as a critical tool for the diagnosis and treatment of COVID-19. However, the manual interpretation of chest X-ray films has proven to be inefficient and time-consuming, necessitating the development of an automated classification system.

OBJECTIVE

In response to the challenges posed by the COVID-19 pandemic, we aimed to develop a deep learning model that accurately classifies chest X-ray images, specifically focusing on lung regions, to enhance the efficiency and accuracy of COVID-19 and pneumonia diagnosis.

METHODS

We have proposed a novel deep network called "FocusNet" for precise segmentation of lung regions in chest radiographs. This segmentation allows for the accurate extraction of lung contours from chest X-ray images, which are then input into the classification network, ResNet18. By training the model on these segmented lung datasets, we sought to improve the accuracy of classification.

RESULTS

The performance of our proposed system was evaluated on three types of lung regions in normal individuals, COVID-19 patients, and those with pneumonia. The average accuracy of the segmentation model (FocusNet) in segmenting lung regions was found to be above 90%. After reclassification of the segmented lung images, the specificities and sensitivities for normal, COVID-19, and pneumonia were excellent, with values of 98.00%, 99.00%, 99.50%, and 98.50%, 100.00%, and 99.00%, respectively. ResNet18 achieved impressive area under the curve (AUC) values of 0.99, 1.00, and 0.99 for classifying normal, COVID-19, and pneumonia, respectively, on the segmented lung datasets. Moreover, the AUC values of the three groups increased by 0.02, 0.02, and 0.06, respectively, when compared to the direct classification of unsegmented original images. Overall, the accuracy of lung region classification after processing the datasets was 99.3%.

CONCLUSION

Our deep learning-based automated chest X-ray classification system, incorporating lung region segmentation using FocusNet and subsequent classification with ResNet18, has significantly improved the accuracy of diagnosing respiratory lung diseases, including COVID-19. The proposed approach has great potential to revolutionize the diagnosis of COVID-19 and other respiratory lung diseases, offering a valuable tool to support healthcare professionals during health crises.

Bilateral pulmonary sequestrations: Distinct types sharing a common celiac artery origin, presented with pneumonia and associated findings in a pediatric case

A 14-year-old boy presented with shortness of breath, cough, and mild chest pain, with a history of intermittent milder symptoms. Physical examination and initial tests showed mild iron deficiency anemia, elevated C-reactive protein, and normal vital signs. Chest radiograph revealed abnormalities in the left lower zone, leading to contrast-enhanced chest CT. The CT scan revealed bilateral intrapulmonary sequestrations, both deriving blood supply from a common trunk originating from the celiac artery. The patient's symptoms initially attributed to a pulmonary infection improved with antibiotic therapy. Pulmonary sequestration is a congenital anomaly characterized by aberrant lung tissue lacking connections to bronchial tree or pulmonary arteries. It can lead to recurrent pulmonary infections and postinfectious sequelae. This case presented a unique bilateral sequestration, both originating from the celiac artery. Radiologists should be aware of sequestration types and associated anomalies, even in atypical locations. Blood supply can originate from various arteries, not just the aorta.

Automated Cough Sound Analysis for Detecting Childhood Pneumonia

Pneumonia is one of the leading causes of death in children. Prompt diagnosis and treatment can help prevent these deaths, particularly in resource poor regions where deaths due to pneumonia are highest. Clinical symptom-based screening of childhood pneumonia yields excessive false positives, highlighting the necessity for additional rapid diagnostic tests. Cough is a prevalent symptom of acute respiratory illnesses and the sound of a cough can indicate the underlying pathological changes resulting from respiratory infections. In this study, we propose a fully automated approach to evaluate cough sounds to distinguish pneumonia from other acute respiratory diseases in children. The proposed method involves cough sound denoising, cough sound segmentation, and cough sound classification. The denoising algorithm utilizes multi-conditional spectral mapping with a multilayer perceptron network while the segmentation algorithm detects cough sounds directly from the denoised audio waveform. From the segmented cough signal, we extract various handcrafted features and feature embeddings from a pretrained deep learning network. A multilayer perceptron is trained on the combined feature set for detecting pneumonia. The method we propose is evaluated using a dataset comprising cough sounds from 173 children diagnosed with either pneumonia or other acute respiratory diseases. On average, the denoising algorithm improved the signal-to-noise ratio by 44%. Furthermore, a sensitivity and specificity of 91% and 86%, respectively, is achieved in cough segmentation and 82% and 71%, respectively, in detecting childhood pneumonia using cough sounds alone. This demonstrates its potential as a rapid diagnostic tool, such as using smartphone technology.

Radiographic pneumonia in young febrile infants presenting to the emergency department: secondary analysis of a prospective cohort study

OBJECTIVE

The lack of evidence-based criteria to guide chest radiograph (CXR) use in young febrile infants results in variation in its use with resultant suboptimal quality of care. We sought to describe the features associated with radiographic pneumonias in young febrile infants.

STUDY DESIGN

Secondary analysis of a prospective cohort study in 18 emergency departments (EDs) in the Pediatric Emergency Care Applied Research Network from 2016 to 2019. Febrile (≥38°C) infants aged ≤60 days who received CXRs were included. CXR reports were categorised as 'no', 'possible' or 'definite' pneumonia. We compared demographics, clinical signs and laboratory tests among infants with and without pneumonias.

RESULTS

Of 2612 infants, 568 (21.7%) had CXRs performed; 19 (3.3%) had definite and 34 (6%) had possible pneumonias. Patients with definite (4/19, 21.1%) or possible (11/34, 32.4%) pneumonias more frequently presented with respiratory distress compared with those without (77/515, 15.0%) pneumonias (adjusted OR 2.17; 95% CI 1.04 to 4.51). There were no differences in temperature or HR in infants with and without radiographic pneumonias. The median serum procalcitonin (PCT) level was higher in the definite (0.7 ng/mL (IQR 0.1, 1.5)) vs no pneumonia (0.1 ng/mL (IQR 0.1, 0.3)) groups, as was the median absolute neutrophil count (ANC) (definite, 5.8 K/mcL (IQR 3.9, 6.9) vs no pneumonia, 3.1 K/mcL (IQR 1.9, 5.3)). No infants with pneumonia had bacteraemia. Viral detection was frequent (no pneumonia (309/422, 73.2%), definite pneumonia (11/16, 68.8%), possible pneumonia (25/29, 86.2%)). Respiratory syncytial virus was the predominant pathogen in the pneumonia groups and rhinovirus in infants without pneumonias.

CONCLUSIONS

Radiographic pneumonias were uncommon in febrile infants. Viral detection was common. Pneumonia was associated with respiratory distress, but few other factors. Although ANC and PCT levels were elevated in infants with definite pneumonias, further work is necessary to evaluate the role of blood biomarkers in infant pneumonias.

Effective processing pipeline PACE 2.0 for enhancing chest x-ray contrast and diagnostic interpretability

Preprocessing is an essential task for the correct analysis of digital medical images. In particular, X-ray imaging might contain artifacts, low contrast, diffractions or intensity inhomogeneities. Recently, we have developed a procedure named PACE that is able to improve chest X-ray (CXR) images including the enforcement of clinical evaluation of pneumonia originated by COVID-19. At the clinical benchmark state of this tool, there have been found some peculiar conditions causing a reduction of details over large bright regions (as in ground-glass opacities and in pleural effusions in bedridden patients) and resulting in oversaturated areas. Here, we have significantly improved the overall performance of the original approach including the results in those specific cases by developing PACE2.0. It combines 2D image decomposition, non-local means denoising, gamma correction, and recursive algorithms to improve image quality. The tool has been evaluated using three metrics: contrast improvement index, information entropy, and effective measure of enhancement, resulting in an average increase of 35% in CII, 7.5% in ENT, 95.6% in EME and 13% in BRISQUE against original radiographies. Additionally, the enhanced images were fed to a pre-trained DenseNet-121 model for transfer learning, resulting in an increase in classification accuracy from 80 to 94% and recall from 89 to 97%, respectively. These improvements led to a potential enhancement of the interpretability of lesion detection in CXRs. PACE2.0 has the potential to become a valuable tool for clinical decision support and could help healthcare professionals detect pneumonia more accurately.

Attention-Guided Partial Domain Adaptation for Automated Pneumonia Diagnosis From Chest X-Ray Images

Deep neural networks (DNN) supported by multicenter large-scale Chest X-Ray (CXR) datasets can efficiently perform tasks such as disease identification, lesion segmentation, and report generation. However, the non-ignorable inter-domain heterogeneity caused by different equipment, ethnic groups, and scanning protocols may lead to dramatic degradation in model performance. Unsupervised domain adaptation (UDA) methods help alleviate the cross-domain discrepancy for subsequent analysis. Nevertheless, they may be prone to: 1) spatial negative transfer: misaligning non-transferable regions which have inadequate knowledge, and 2) semantic negative transfer: failing to extend to scenarios where the label spaces of the source and target domain are partially shared. In this work, we propose a classification-based framework named attention-guided partial domain adaptation (AGPDA) network for overcoming these two negative transfer challenges. AGPDA is composed of two key modules: 1) a region attention discrimination block (RADB) to generate fine-grained attention value via lightweight region-wise multi-adversarial networks. 2) a residual feature recalibration block (RFRB) trained with class-weighted maximum mean discrepancy (MMD) loss for down-weighing the irrelevant source samples. Extensive experiments on two publicly available CXR datasets containing a total of 8598 pneumonia (viral, bacterial, and COVID-19) cases, 7163 non-pneumonia or healthy cases, demonstrate the superior performance of our AGPDA. Especially on three partial transfer tasks, AGPDA significantly increases the accuracy, sensitivity, and F1 score by 4.35%, 4.05%, and 1.78% compared to recently strong baselines.

Investigation of Childhood Pneumonia With Thoracic Ultrasound: A Comparison Between X-ray and Ultrasound

ABSTRACT

Childhood pneumonia is a common entity, and chest x-rays are widely used as an initial diagnostic step. To avoid radiation exposure in the pediatric age group, we assessed whether the diagnostic accuracy of ultrasound (US) imaging is sufficient in the diagnosis. One hundred thirty-three children with pneumonia (72 girls/61 boys) were participated to study between 2019 and 2021. All participants had a chest x-ray. Radiologists who perform the US scans and interpret the x-rays were blinded to each other. A comparative analysis was also done to assess US findings on pneumonia for different age groups. We compare the diagnostic accuracy of US and x-rays by McNemar test and receiver operating characteristic curves. Intraclass correlation coefficient values were calculated for the assessment of interobserver agreement of x-ray evaluation. The participants' ages ranged from 1 month to 17 years and 8 months with a median age of 24 months (Q 1 : 8 and Q 3 : 66 months). Hospital stay lengths were longer, consolidation depths were greater, and presence of air bronchogram or pleural effusion was more frequent in school-age children. The proportion of consolidation seen on chest x-ray and transthoracic US scan was significantly different ( P < 0.001). The area under the curve was greater in the US than in the chest x-ray (area under the curve, 0.94 and 0.76 respectively). There was a good agreement between the 2 interpreters on chest x-ray assessment ( κ = 0.661). The thoracic US can be used as a safe and efficient imaging tool in the diagnosis of pediatric pneumonia.

Testing the performance, adequacy, and applicability of an artificial intelligence model for pediatric pneumonia diagnosis

BACKGROUND

Community-acquired Pneumonia (CAP) is a common childhood infectious disease. Deep learning models show promise in X-ray interpretation and diagnosis, but their validation should be extended due to limitations in the current validation workflow. To extend the standard validation workflow we propose doing a pilot test with the next characteristics. First, the assumption of perfect ground truth (100% sensitive and specific) is unrealistic, as high intra and inter-observer variability have been reported. To address this, we propose using Bayesian latent class models (BLCA) to estimate accuracy during the pilot. Additionally, assessing only the performance of a model without considering its applicability and acceptance by physicians is insufficient if we hope to integrate AI systems into day-to-day clinical practice. Therefore, we propose employing explainable artificial intelligence (XAI) methods during the pilot test to involve physicians and evaluate how well a Deep Learning model is accepted and how helpful it is for routine decisions as well as analyze its limitations by assessing the etiology. This study aims to apply the proposed pilot to test a deep Convolutional Neural Network (CNN)-based model for identifying consolidation in pediatric chest-X-ray (CXR) images already validated using the standard workflow.

METHODS

For the standard validation workflow, a total of 5856 public CXRs and 950 private CXRs were used to train and validate the performance of the CNN model. The performance of the model was estimated assuming a perfect ground truth. For the pilot test proposed in this article, a total of 190 pediatric chest-X-ray (CXRs) images were used to test the CNN model support decision tool (SDT). The performance of the model on the pilot test was estimated using extensions of the two-test Bayesian Latent-Class model (BLCA). The sensitivity, specificity, and accuracy of the model were also assessed. The clinical characteristics of the patients were compared according to the model performance. The adequacy and applicability of the SDT was tested using XAI techniques. The adequacy of the SDT was assessed by asking two senior physicians the agreement rate with the SDT. The applicability was tested by asking three medical residents before and after using the SDT and the agreement between experts was calculated using the kappa index.

RESULTS

The CRXs of the pilot test were labeled by the panel of experts into consolidation (124/176, 70.4%) and no-consolidation/other infiltrates (52/176, 29.5%). A total of 31/176 (17.6%) discrepancies were found between the model and the panel of experts with a kappa index of 0.6. The sensitivity and specificity reached a median of 90.9 (95% Credible Interval (CrI), 81.2-99.9) and 77.7 (95% CrI, 63.3-98.1), respectively. The senior physicians reported a high agreement rate (70%) with the system in identifying logical consolidation patterns. The three medical residents reached a higher agreement using SDT than alone with experts (0.66±0.1 vs. 0.75±0.2).

CONCLUSIONS

Through the pilot test, we have successfully verified that the deep learning model was underestimated when a perfect ground truth was considered. Furthermore, by conducting adequacy and applicability tests, we can ensure that the model is able to identify logical patterns within the CXRs and that augmenting clinicians with automated preliminary read assistants could accelerate their workflows and enhance accuracy in identifying consolidation in pediatric CXR images.

Deep Learning-Based Computer-Aided Detection System for Preoperative Chest Radiographs to Predict Postoperative Pneumonia

RATIONALE AND OBJECTIVES

The role of preoperative chest radiography (CR) for prediction of postoperative pneumonia remains uncertain. We aimed to develop and validate a prediction model for postoperative pneumonia incorporating findings of preoperative CRs evaluated by a deep learning-based computer-aided detection (DL-CAD) system MATERIALS AND METHODS: This retrospective study included consecutive patients who underwent surgery between January 2019 and March 2020 and divided into development (surgery in 2019) and validation (surgery between January and March 2020) cohorts. Preoperative CRs obtained within 1-month before surgery were analyzed with a commercialized DL-CAD that provided probability values for the presence of 10 different abnormalities in CRs. Logistic regression models to predict postoperative pneumonia were built using clinical variables (clinical model), and both clinical variables and DL-CAD results for preoperative CRs (DL-CAD model). The discriminative performances of the models were evaluated by area under the receiver operating characteristic curves.

RESULTS

In development cohort (n = 19,349; mean age, 57 years; 11,392 men), DL-CAD results for pulmonary nodules (odds ratio [OR, for 1% increase in probability value], 1.007; p = 0.021), consolidation (OR, 1.019; p < 0.001), and cardiomegaly (OR, 1.013; p < 0.001) were independent predictors of postoperative pneumonia and were included in the DL-CAD model. In validation cohort (n = 4957; mean age, 56 years; 2848 men), the DL-CAD model exhibited a higher AUROC than the clinical model (0.843 vs. 0.815; p = 0.012).

CONCLUSION

Abnormalities in preoperative CRs evaluated by a DL-CAD were independent risk factors for postoperative pneumonia. Using DL-CAD results for preoperative CRs led to an improved prediction of postoperative pneumonia.

Investigation of the relationship between pulmonary lesions based on lung ultrasound and respiratory clinical signs in foals with suspected pulmonary rhodococcosis

Rhodococcus equi is a widely recognized bacterium responsible for pneumonia in preweaned foals. On endemic farms, foals with a subclinical course of the disease usually outnumber those presenting clinical signs. The disease is typically chronic and mainly manifests as fever and dyspnoea. Currently, field diagnosis is often based on lung ultrasound (LUS); however, both diagnostic and therapeutic approaches vary among practitioners and considerably change over time. This longitudinal, prospective study was designed to describe the appearance and progression of rhodococcal pulmonary lesions during the first months of life based on LUS and to evaluate the relationship between the presence and severity of rhodococcal pulmonary lesions and the occurrence of respiratory clinical signs in foals from farms with endemic R. equi infections. Nearly 26% of foals demonstrated respiratory signs highly suggestive of pulmonary rhodococcosis, and approximately 70% of the foals had abnormalities detected on LUS without concurrent clinical signs. The appearance and development of LUS abnormalities were age-related. An abscess diameter exceeding 15 mm in LUS and other pleural lesions were significantly linked with the occurrence of clinical signs suggestive of pulmonary rhodococcosis (P < 0.001) and may be considered predictive factors of rhodococcal pneumonia in foals.

Three-dimensional ultrashort echo time magnetic resonance imaging in pediatric patients with pneumonia: a comparative study

BACKGROUND

UTE has been used to depict lung parenchyma. However, the insufficient discussion of its performance in pediatric pneumonia compared with conventional sequences is a gap in the existing literature. The objective of this study was to compare the diagnostic value of 3D-UTE with that of 3D T1-GRE and T2-FSE sequences in young children diagnosed with pneumonia.

METHODS

Seventy-seven eligible pediatric patients diagnosed with pneumonia at our hospital, ranging in age from one day to thirty-five months, were enrolled in this study from March 2021 to August 2021. All patients underwent imaging using a 3 T pediatric MR scanner, which included three sequences: 3D-UTE, 3D-T1 GRE, and T2-FSE. Subjective analyses were performed by two experienced pediatric radiologists based on a 5-point scale according to six pathological findings (patchy shadows/ground-glass opacity (GGO), consolidation, nodule, bulla/cyst, linear opacity, and pleural effusion/thickening). Additionally, they assessed image quality, including the presence of artifacts, and evaluated the lung parenchyma. Interrater agreement was assessed using intraclass correlation coefficients (ICCs). Differences among the three sequences were evaluated using the Wilcoxon signed-rank test.

RESULTS

The visualization of pathologies in most parameters (patchy shadows/GGO, consolidation, nodule, and bulla/cyst) was superior with UTE compared to T2-FSE and T1 GRE. The visualization scores for linear opacity were similar between UTE and T2-FSE, and both were better than T1-GRE. In the case of pleural effusion/thickening, T2-FSE outperformed the other sequences. However, statistically significant differences between UTE and other sequences were only observed for patchy shadows/GGO and consolidation. The overall image quality was superior or at least comparable with UTE compared to T2-FSE and T1-GRE. Interobserver agreements for all visual assessments were significant and rated "substantial" or "excellent."

CONCLUSIONS

In conclusion, UTE MRI is a useful and promising method for evaluating pediatric pneumonia, as it provided better or similar visualization of most imaging findings compared with T2-FSE and T1-GRE. We suggest that the UTE MRI is well-suited for pediatric population, especially in younger children with pneumonia who require longitudinal and repeated imaging for clinical care or research and are susceptible to ionizing radiation.

Comparison of Ultrasound and CT Imaging for the Diagnosis of Coronavirus Disease and Influenza A Pneumonia

OBJECTIVE

The outbreak of coronavirus disease (COVID-19) coincided with the season of influenza A pneumonia, a common respiratory infectious disease. Therefore, this study compared ultrasonography and computed tomography (CT) for the diagnosis of the two diseases.

METHODS

Patients with COVID-19 or influenza A infection hospitalized at our hospital were included. The patients were examined by ultrasonography every day. The CT examination results within 1 day before and after the day of the highest ultrasonography score were selected as the controls. The similarities and differences between the ultrasonography and CT results in the two groups were compared.

RESULTS

There was no difference between the ultrasonography and CT scores (P = .307) for COVID-19, while there was a difference between ultrasonography and CT scores for influenza A pneumonia (P = .024). The ultrasonography score for COVID-19 was higher than that for influenza A pneumonia (P = .000), but there was no difference between the CT scores (P = .830). For both diseases, there was no difference in ultrasonography and CT scores between the left and right lungs; there were differences between the CT scores of the upper and middle lobes, as well as between the upper and lower lobes of the lungs; however, there was no difference between the lower and middle lobes of the lungs.

CONCLUSION

Ultrasonography is equivalent to the gold standard CT for diagnosing and monitoring the progression of COVID-19. Because of its convenience, ultrasonography has important application value. Furthermore, the diagnostic value of ultrasonography for COVID-19 is higher than that for influenza A pneumonia.

Comparison between original SARS-CoV-2 strain and omicron variant on thin-section chest CT imaging of COVID-19 pneumonia

OBJECTIVES

We investigated different computed tomography (CT) features between Omicron-variant and original-strain SARS-CoV‑2 pneumonia to facilitate the clinical management.

MATERIALS AND METHODS

Medical records were retrospectively reviewed to select patients with original-strain SARS-CoV‑2 pneumonia from February 22 to April 22, 2020, or Omicron-variant SARS-CoV‑2 pneumonia from March 26 to May 31, 2022. Data on the demographics, comorbidities, symptoms, clinical types, and CT features were compared between the two groups.

RESULTS

There were 62 and 78 patients with original-strain or Omicron-variant SARS-CoV‑2 pneumonia, respectively. There were no differences between the two groups in terms of age, sex, clinical types, symptoms, and comorbidities. The main CT features differed between the two groups (p = 0.003). There were 37 (59.7%) and 20 (25.6%) patients with ground-glass opacities (GGO) in the original-strain and Omicron-variant pneumonia, respectively. A consolidation pattern was more frequently observed in the Omicron-variant than original-strain pneumonia (62.8% vs. 24.2%). There was no difference in crazy-paving pattern between the original-strain and Omicron-variant pneumonia (16.1% vs. 11.6%). Pleural effusion was observed more often in Omicron-variant pneumonia, while subpleural lesions were more common in the original-strain pneumonia. The CT score in the Omicron-variant group was higher than that in the original-strain group for critical-type (17.00, 16.00-18.00 vs. 16.00, 14.00-17.00, p = 0.031) and for severe-type (13.00, 12.00-14.00 vs 12.00, 10.75-13.00, p = 0.027) pneumonia.

CONCLUSION

The main CT finding of the Omicron-variant SARS-CoV‑2 pneumonia included consolidations and pleural effusion. By contrast, CT findings of original-strain SARS-CoV‑2 pneumonia showed frequent GGO and subpleural lesions, but without pleural effusion. The CT scores were also higher in the critical and severe types of Omicron-variant than original-strain pneumonia.

Differential clinical and CT imaging features of pneumonic-type primary pulmonary lymphoma and pneumonia: a retrospective multicentre observational study

INTRODUCTION

Pneumonic-type primary pulmonary lymphoma (PPL) is often misdiagnosed as pneumonia in clinical practice. However, this disease requires different treatments, which calls for a correct diagnosis.

MATERIALS AND METHODS

A total of 227 patients with pneumonic-type PPL (n=72) and pneumonia (n=155) from 7 institutions were retrospectively enrolled between January 2017 and January 2022. Clinical features (age, sex, cough, sputum, fever, haemoptysis, chest pain, smoking, weight loss and laboratory results (haemoglobin, white blood cell count, C reactive protein level and erythrocyte sedimentation rate)) and CT imaging characteristics (air bronchogram, bronchiectasis, halo sign, pleural traction, pleural effusion, lymphadenopathy, lesion maximum diameter and CT attenuation value) were analysed. Receiver operating characteristic curve analysis was performed for model construction based on independent predictors in identifying pneumonic-type PPL. In addition, we used a calibration curve and decision curve analysis to estimate the diagnostic efficiency of the model.

RESULTS

The patients with pneumonia showed a higher prevalence of sputum, fever, leucocytosis and elevation of C reactive protein level than those with pneumonic-type PPL (p=0.002, p<0.001, p=0.011 and p<0.001, respectively). Bronchiectasis, halo sign and higher CT attenuation value were more frequently present in pneumonic-type PPL than in pneumonia (all p<0.001). Pleural effusion was more commonly observed in patients with pneumonia than those with pneumonic-type PPL (p<0.001). Also, sputum, fever, elevation of C reactive protein level, halo sign, bronchiectasis, pleural effusion and CT attenuation value were the independent predictors of the presence of pneumonic-type PPL with an area under the curve value of 0.908 (95% CI, 0.863 to 0.942).

CONCLUSION

Pneumonic-type PPL and pneumonia have different clinical and imaging features. These differential features could be beneficial in guiding early diagnosis and subsequent initiation of therapy.

[POCUS : diagnosis of pneumonia by lung ultrasonography]

Point-Of-Care Ultrasonography (POCUS) has emerged to support the diagnosis process and management strategies. Its use for the diagnosis of pneumonia has been shown to be reliable and effective over the past decade. Various ultrasonography patterns exist, none of which are pathognomonic for pneumonia. Therefore, POCUS findings must be interpreted in association with the clinical setting. POCUS enables early identification of complications such as parapneumonic effusion and pulmonary abscess. It also provides guidance for invasive procedure such as thoracocentesis and pleural drainage. The forthcoming results of the Swiss OCTOPLUS study will provide data on the clinical and economic impact of a diagnostic strategy based on targeted lung ultrasonography.

Computer-aided diagnosis of chest X-ray for COVID-19 diagnosis in external validation study by radiologists with and without deep learning system

To evaluate the diagnostic performance of our deep learning (DL) model of COVID-19 and investigate whether the diagnostic performance of radiologists was improved by referring to our model. Our datasets contained chest X-rays (CXRs) for the following three categories: normal (NORMAL), non-COVID-19 pneumonia (PNEUMONIA), and COVID-19 pneumonia (COVID). We used two public datasets and private dataset collected from eight hospitals for the development and external validation of our DL model (26,393 CXRs). Eight radiologists performed two reading sessions: one session was performed with reference to CXRs only, and the other was performed with reference to both CXRs and the results of the DL model. The evaluation metrics for the reading session were accuracy, sensitivity, specificity, and area under the curve (AUC). The accuracy of our DL model was 0.733, and that of the eight radiologists without DL was 0.696 ± 0.031. There was a significant difference in AUC between the radiologists with and without DL for COVID versus NORMAL or PNEUMONIA (p = 0.0038). Our DL model alone showed better diagnostic performance than that of most radiologists. In addition, our model significantly improved the diagnostic performance of radiologists for COVID versus NORMAL or PNEUMONIA.