Use of Artificial Intelligence to Triage Patients with Flu-Like Symptoms Using Imaging in Non-COVID-19 Hospitals during COVID-19 Pandemic: An Ongoing 8-Month Experience

Background Evaluation of suspected coronavirus disease-2019 (COVID-19) patient is a diagnostic dilemma as it commonly presents like influenza in early stages. Studies and guidelines have emerged both for and against the use of imaging as a frontline tool to investigate such patients. Reverse transcriptase-polymerase chain reaction (RT-PCR) is suggested as the backbone of diagnosis. We designed and tested a diagnostic algorithm using artificial intelligence (AI) to determine the role of imaging in the evaluation of patients with acute flu-like presentation. Materials and Methods Overall, 3,235 consecutive patients with flu-like presentation were evaluated over a period of 240 days. All patients underwent plain radiographs of chest with computer-aided detection for COVID-19 (CAD4COVID) AI analysis. Based on the threshold scores, they were divided into two groups: group A (score < 50) and group B (score > 50). Group A patients were discharged and put on routine symptomatic treatment and follow-up with RT-PCR, while group B patients underwent high-resolution computed tomography (HRCT) followed by COVID-19 AI analysis and RT-PCR test. These were then triaged into COVID-19 and non-COVID-19 subgroups based on COVID-19 similarity scores by AI, and lung severity scores were also determined. Results Group A had 2,209 (68.3%) patients with CAD4COVID score of <50 while 1,026 (31.7%) patients comprised group B. Also, 825 (25.5%) patients were COVID-19 positive with COVID-19 similarity threshold of >0.85 on AI. RT-PCR was positive in 415 and false-negative in 115 patients while 12 patients died before the test could be done. The sensitivity and specificity of CAD4COVID AI analysis on plain radiographs for detection of any lung abnormality combined with HRCT AI analysis was 97.9% and 99% using the above algorithm. Conclusion Combined use of chest radiographs and plain HRCT with AI-based analysis is useful and an accurate frontline tool to triage patients with acute flu-like symptoms in non-COVID-19 health care facilities.

Projection radiography of the ankle; the views used for different indications vary between medical imaging departments

Plain radiographs are used for initial evaluation of many conditions of the ankle. Many different radiographic views are described in positioning textbooks but evidence on which views to use, in which case, is scarce. The aim of this study was to map imaging procedures related to four indications for ankle projection radiography. A questionnaire was sent to all medical imaging departments in Iceland with questions about acquisition technique for ankle radiography views and which views were used for selected indications. Answer was received from 14 of the 28 departments. All departments gave very similar descriptions of the four most common views. In the case of trauma, all but one department used four views but for control of trauma or operation, four different combinations of views were found using from two to four images. For detrition and osteomyelitis, four views were more common in the larger departments but there was not a statistically significant difference. Eight different combinations of the number of views for the four indications were found. The study indicates that there is a need for standardization in image acquisition protocols. More studies are needed to support decisions about how many views are necessary for the most common ankle radiography indications.