PORTLAND – AI software can identify gaps in care of people with vertebral compression fractures, according to a presentation on May 21 at the Society for Imaging Informatics in Medicine (SIIM) annual meeting.
Mohamed Ibrahim, MD, presented a study May 21 on how AI can identify gaps in care for patients with vertebral compression fractures.
The finding is from a study in which the model flagged 13% of chest x-ray exams from a large dataset, of which 79% were discordant with the radiology reports, noted Mohamed Ibrahim, MD, of the University of Alabama at Birmingham.
“Vertebral compression fracture is usually underreported, especially on chest x rays. Looking at the literature, more than 50% of these cases go underreported,” Ibrahim said.
Vertebral compression fractures (VCFs) are breaks in spine vertebra, which causes the bone to collapse. VCFs are common in patients with osteoporosis and over time can cause the spine to curve. Identifying VCFs is crucial for risk-stratifying patients for pharmacologic therapy in osteoporosis management, Ibrahim explained.
In this retrospective study, Ibrahim and colleagues aimed to estimate the gap in care by based on the incidence of missed VCFs using AI (Protect Her, Covera Health) as a subsequent reader.
The researchers included a dataset of 16,066 consecutive posteroanterior and lateral chest x-ray exams. AI-flagged exams were classified as positive or negative for VCF. In addition, radiology reports were analyzed using natural language processing to determine whether VCF was reported. Lastly, a radiologist reviewed the images to establish the reference standard in a subset of discordant results.
The AI model flagged 13% (2,134 out of 16,066) exams for VCF, of which 79% (1,695 out of 2,134) were discordant with the radiology reports. However, 1.5% (237 out of 16,066) of the x-rays were positive for VCF, but missed by AI, Ibrahim noted.
In the subset of exams reviewed by the radiologist, the incidence of VCF was 9.6% (199 out of 2,072). VCF was only reported in 30.1% (60 out of 199) of these exams and VCF was not reported in 69.8% (139 out of 199) of confirmed VCF cases, according to the findings.
“So the conclusion is that vertebral compression fractures are underreported, and there is a huge gap. AI could be a potential use to narrow this gap and affect patient management in the longer run,” Ibrahim said.
Limitations of the study included that this was a single-center study and that the analysis was based on the identification of moderate to severe cases of VCF. Mild fractures were considered false-positives in the analysis, but these cases may be clinically significant in patients with osteoporosis, Ibrahim noted. Also, the x-ray report text was not reviewed by a human, he said.
Next steps include using the “2+1” method in which two readers will independently review all images, followed by adjudication of discrepancies by a senior cardiothoracic radiologist, he added.
“All reports need to be manually reviewed and we will correlate our findings with patients’ clinical histories and outcomes,” Ibrahim concluded.
