Preliminary research revealed an area under the curve (AUC) of 85 percent for an artificial intelligence (AI) algorithm in diagnosing COVID-19 on initial chest X-rays in comparison to a consensus 71 percent AUC for five radiologists.
While reverse transcription-polymerase chain reaction (RT-PCR) testing is considered the standard for the diagnosis of COVID-19, an emerging artificial intelligence (AI) algorithm may enhance early diagnosis of the disease on initial chest radiography screening.
In a recent retrospective study published in Insights into Imaging, researchers compared radiologist assessment of initial chest X-rays for COVID-19 versus an AI algorithm model comprised from four deep convolutional neural networks. The algorithm was trained with frontal chest X-ray images from 3,647 patients who were positive for COVID-19 and 1,908 patients who did not have COVID-19, according to the study.
In the testing of the algorithm with a subset of initial X-ray images for 225 patients who tested positive for COVID-19 and 255 patients who tested negative for the disease, the study authors found that the algorithm had an area under the curve (AUC) of 85 percent versus a 71 percent AUC for a consensus of five radiologists with experience ranging between 1 to 14 years.
Acknowledging the algorithm’s lower sensitivity for diagnosing COVID-19 in comparison to similar algorithms from other studies, the study authors said the focus of their research was the initial X-ray screening “when pulmonary findings are less conspicuous” than more advanced stages of COVID-19. The researchers said this focus is more conducive to real-world applications involving early diagnosis to facilitate effective triage and timely treatment intervention.
“The advantage of AI may lie in its ability to identify changes in medical images before they become visible to the human eye. That would be the case with COVID-19 patients before abnormal lung densities were conspicuous in (chest X-rays),” wrote study co-author Juan I. Rayo Madrid, MD, who is affiliated with the Department of Nuclear Medicine at Complejo Hospitalario Universitario de Badajoz in Badajoz, Spain, and colleagues.
While recognizing that the predictive power of initial X-ray screening is not optimal for diagnosis of COVID-19, the study authors noted that the modality is inexpensive and widely available. These attributes could make the combination of the AI algorithm and initial X-rays beneficial in limited resource environments where RT-PCR testing results are delayed or unavailable, according to Madrid and colleagues.
“Applying this tool systematically to all (chest X-rays) performed in any health center would allow detection of the disease early without other tests,” noted the study authors.
Madrid and colleagues also suggested that the combination of chest X-rays with adjunctive AI may be more beneficial than RT-PCR testing for ongoing monitoring of lung involvement in patients diagnosed with emerging variants of COVID-19.
“We have experienced continuous outbreaks of new variants whose severity, infectious potential, and symptoms are initially uncertain and need to be intensively investigated for weeks or months. In this context, the RT-PCR test may be less sensitive,” said Madrid and colleagues.
In regard to study limitations, Madrid and coauthors noted the AI algorithm software was trained and tested in the same centers. They acknowledged a lack of information on symptoms, signs and additional tests ordered for patients suspected of having COVID-19. The study also did not examine the potential contribution of the AI algorithm software to radiologist assessment in clinical practice, according to Madrid and colleagues.
CT Study: AI Algorithm Comparable to Radiologists in Differentiating Small Renal Masses
May 14th 2024An emerging deep learning algorithm had a lower AUC and sensitivity than urological radiologists for differentiating between small renal masses on computed tomography (CT) scans but had a 21 percent higher sensitivity rate than non-urological radiologists, according to new research.
MRI-Based Deep Learning Algorithm Shows Comparable Detection of csPCa to Radiologists
May 8th 2024In a study involving over 1,000 visible prostate lesions on biparametric MRI, a deep learning algorithm detected 96 percent of clinically significant prostate cancer (csPCa) in comparison to a 98 percent detection rate for an expert genitourinary radiologist.
FDA Clears AI-Powered Qualitative Perfusion Mapping for Cone-Beam CT
May 6th 2024Reportedly validated in more than 10 clinical trials, the AngioFlow perfusion imaging software enables timely identification of brain regions with cerebral blood flow reduction and those with significant hypoperfusion.