The Future of Respiratory Screening: AI, ECGs, and Ultrasound

Featured Buzz – January 11, 2026

ECGs Show Promise for Early Diagnosis of COPD
Electrocardiograms (ECGs) are widely used during routine clinical visits to screen patients for possible heart disease. Could they also be used to screen for COPD?

According to researchers from the Mount Sinai Health System who harnessed the power of artificial intelligence (AI) to analyze ECG results for COPD, the answer may be yes. The investigators began by collecting 208,231 ECGs from 18,225 patients with COPD at five Mount Sinai hospitals in the New York area, along with 552,771 ECGs from 59,356 controls matched by age, sex, and race.

Data from these patients were then used to train an AI-driven Convolutional Neural Network model designed to uncover cases of COPD based on ECG findings. The primary outcome measure was the accuracy of a new clinical COPD diagnosis based on ICD codes.

The model was tested on patients in three cohorts — an internal testing cohort, an external validation cohort, and an additional external validation cohort consisting of 258 COPD cases — along with 1,290 age- and sex-matched controls.

Results showed:

• 18,225 patients were identified with COPD, 64.6% of whom had COPD alone.
• The remaining patients had COPD along with comorbidities such as asthma (15.1%), heart failure (12%), obstructive sleep apnea (5.7%), lung cancer (3.1%), and acute bronchitis (1.5%).
• The model exhibited robust performance across diverse populations, with AUCs of 0.80 in internal testing, 0.82 in external validation, and 0.75 in the controls.
• ECG-derived model predictions were linked with spirometry data in a subsequent analysis, and P-wave changes were indicative of COPD.

The authors believe that while ECGs will not replace spirometry in the diagnosis of COPD, AI-enhanced ECG interpretation offers a pragmatic approach to screening that can lead to earlier treatment, ultimately improving quality of life for people with the condition.

They write, “Earlier recognition may facilitate timely smoking cessation, targeted therapies, and pulmonary rehabilitation, potentially slowing disease progression and reducing health care burden.” 

The study was published by eBioMedicine.

Read Full Paper

Lung Ultrasound Predicts Extubation Success in VLBW Infants
Researchers from the University of Chicago have found that lung ultrasound (LUS) scores can predict extubation success in preterm infants on mechanical ventilation.

The study compared outcomes among 45 very low birth weight (VLBW) infants who were intubated for the respiratory distress syndrome. Fifty-three extubation attempts were made in the group as a whole. A lung ultrasound was performed three to six hours prior to each extubation attempt, and a failed extubation was defined as requiring reintubation within seven days of the extubation attempt.

Key findings of the study include:

• The median LUS score on the day of the extubation attempt was 6. 23 of 45 infants (51.1%) were successfully extubated.
• The median mechanical ventilation duration for infants who were successfully intubated was 4 days, and 20 were extubated without corticosteroids.
• In the 22 infants who experienced extubation failure, the median duration of mechanical ventilation was 38 days.
• The median LUS score was significantly higher in infants who failed extubation than in those who were extubated successfully, 12 vs. 5.
• While LUS scores were similar between infants who did and did not receive dexamethasone, infants treated with dexamethasone who were successfully extubated had lower LUS scores than those who were not.

The authors believe the key take-home messages from their study are that the neonatal-adapted LUS score is an excellent predictor of extubation success when performed on the day of extubation and that dexamethasone treatment is not associated with lower LUS scores.

The study was published in the Journal of Perinatology.

Read Full Paper

Using AI to Diagnose LRTIs Could Cut Down on Inappropriate Antibiotic Use
Researchers from the University of California, San Francisco, have developed a new way to diagnose lower respiratory tract infections (LRTIs) in critically ill patients using the power of AI that they believe could lead to a more judicious use of antibiotics for conditions like pneumonia.

A key component of the LRTI diagnostic strategy is a gene called FABP4, which is found in lung fluid samples and tempers inflammation. FABP4 is expressed less in infected lung cells than in healthy lung cells. This biomarker was combined with a generative AI analysis of information from the electronic medical record, including the chest x-ray radiology report and the medical team’s clinical notes, to develop the final model.

The model — which the authors believe can be readily used by any clinician with access to a HIPAA-compliant GPT-4 interface — was then tested in an observational study of critically ill adults. Results showed it made a correct diagnosis 96% of the time, outperforming ICU clinicians. Most clinicians in the study chose to prescribe antibiotics for pneumonia treatment in these patients.

The authors believe using the model could have cut the use of inappropriate antibiotics by more than 80%. “This study suggests that integrating a host biomarker with large language model analysis can improve LRTI diagnosis in critically ill adults,” they write.

The team is now validating the model for use as a clinical test and plans to apply the lessons learned in this study to sepsis diagnosis. The study was published by Nature Communications.

Read Article
Read Full Paper