A team of researchers from Taiwan performed a first of its kind external validation of four AI algorithms used to detect pulmonary nodules in chest x-rays, sharing their results in Clinical Radiology. The classifiers could help radiologists improve medical imaging care as a whole.
Chest radiographs are typically the initial exam for patients with suspected pulmonary nodules, but the ever-growing workload for radiologists can cause misses that may result in delayed diagnoses or management in these patients, wrote C. H. Liang, National Yang-Ming University’s Department of Biomedical Imaging and Radiological Sciences in Taipei, Taiwan, and colleagues.
“The present study evaluates the diagnostic performance and efficacy of the QUIBIM Chest X-ray Classifier commercially available software for automatic detection of pulmonary nodules or masses on chest radiographs using four different deep-learning algorithms (heat map, nodule, mass, and abnormal probability algorithms),” Liang and colleagues wrote.
In the study, two radiologists verified the presence of pulmonary nodules/ masses in 100 patients (47 with nodules and 53 controls) using chest CT findings.
The mass algorithm was the top performer. It achieved an AUC of 0.916, 76.6% sensitivity and 88.68% specificity for detecting nodules and masses at the optimal probability score cut-off of 0.2884. For comparison, the heat map, abnormal probability, and nodule algorithms scored AUCs of 0.682, 0.810, 0.813, respectively.
“To the authors' knowledge, this is the first study to externally validate the diagnostic performance of AI deep-learning algorithms for the detection of clinically significant pulmonary nodules/masses, which were validated by chest CT,” the researchers wrote.
The classifier also processed images much faster than radiologists could, evaluating each at 94.07±16.54 seconds per case.
“The QUIBIM Chest X-ray Classifier could be used to automatically evaluate all chest radiographs efficiently despite the large volume of chest radiographs prescribed in the outpatient setting,” Liang et al. noted.
What might this look like in a clinical setting? Liang and colleagues imagined an instant medical alert system in which the heat map, integrated with the PACS, algorithm could help radiologists flag suspicious cases. An “early warning score-based alert” could also focus radiologists’ attention on cases with a higher probability score for nodules.
“In conclusion, deep-learning based CAD systems will likely play a vital role in the early detection and diagnosis of pulmonary nodules/masses on chest radiographs,” the researchers wrote. “In future applications, these algorithms could support triage workflow with double reading to improve sensitivity and specificity during the diagnostic process.”
Matt joined Chicago’s TriMed team in 2018 covering all areas of health imaging after two years reporting on the hospital field. He holds a bachelor’s in English from UIC, and enjoys a good cup of coffee and an interesting documentary.