Radiologist-Trained AI Model for Identifying Suboptimal Chest-Radiographs.

RATIONALE AND OBJECTIVES: Suboptimal chest radiographs (CXR) can limit interpretation of critical findings. Radiologist-trained AI models were evaluated for differentiating suboptimal(sCXR) and optimal(oCXR) chest radiographs. MATERIALS AND METHODS: Our IRB-approved study included 3278 CXRs from adult patients (mean age 55 ± 20 years) identified from a retrospective search of CXR in radiology reports from 5 sites. A chest radiologist reviewed all CXRs for the cause of suboptimality. The de-identified CXRs were uploaded into an AI server application for training and testing 5 AI models. The training set consisted of 2202 CXRs (n = 807 oCXR; n = 1395 sCXR) while 1076 CXRs (n = 729 sCXR; n = 347 oCXR) were used for testing. Data were analyzed with the Area under the curve (AUC) for the model's ability to classify oCXR and sCXR correctly. RESULTS: For the two-class classification into sCXR or oCXR from all sites, for CXR with missing anatomy, AI had sensitivity, specificity, accuracy, and AUC of 78%, 95%, 91%, 0.87(95% CI 0.82-0.92), respectively. AI identified obscured thoracic anatomy with 91% sensitivity, 97% specificity, 95% accuracy, and 0.94 AUC (95% CI 0.90-0.97). Inadequate exposure with 90% sensitivity, 93% specificity, 92% accuracy, and AUC of 0.91 (95% CI 0.88-0.95). The presence of low lung volume was identified with 96% sensitivity, 92% specificity, 93% accuracy, and 0.94 AUC (95% CI 0.92-0.96). The sensitivity, specificity, accuracy, and AUC of AI in identifying patient rotation were 92%, 96%, 95%, and 0.94 (95% CI 0.91-0.98), respectively. CONCLUSION: The radiologist-trained AI models can accurately classify optimal and suboptimal CXRs. Such AI models at the front end of radiographic equipment can enable radiographers to repeat sCXRs when necessary.

Curbing Inappropriate Usage of STAT Imaging at a Large Academic Medical Center.

PURPOSE: To evaluate a new system for processing and performing inpatient STAT diagnostic imaging with respect to utilization and time-based performance metrics. MATERIALS AND METHODS: This HIPAA-compliant study had institutional review board approval; informed consent was not required. The radiology information system of a large academic medical center was queried for all inpatient diagnostic imaging exams performed and interpreted from August 1, 2010, to October 31, 2012. Using customized software, data were evaluated based on order priority (non-STAT or STAT) and exam modality with respect to exam volume and time-based performance metrics (time-to-performance and preliminary interpretation time). Data were compared over 3 periods: August 1, 2010, to October 31, 2010 (preimplementation period); November 1, 2010, to October 31, 2011 (year 1 postimplementation); and November 1, 2011, to October 31, 2012 (year 2 postimplementation). RESULTS: In the first year after implementation of the new STAT policy, the percentage of inpatient exams ordered STAT significantly decreased from 22.1% to 5.4% (P < 0.001). This represented a proportional decrease of 26% (CT), 16% (MRI), 20% (US), and 24% (radiographs) relative to pre-STAT policy levels. The median time-to-performance and median preliminary interpretation time significantly decreased for all modalities after implementation of the policy (P < 0.05 for all modalities), decreasing by an average of 104 and 162 minutes, respectively. These changes persisted throughout year 2 postimplementation. CONCLUSION: A new institutional system for handling inpatient STAT diagnostic imaging results in a decreased number of STAT exams ordered and improved time-based performance metrics, thereby increasing workflow efficiency.

Repeat rates in digital chest radiography and strategies for improvement.

PURPOSE: To determine the repeat rate (RR) of chest radiographs acquired with portable computed radiography (CR) and installed direct radiography (DR) and to develop and assess strategies designed to decrease the RR. MATERIALS AND METHODS: The RR and reasons for repeated digital chest radiographs were documented over the course of 16 months while a task force of thoracic radiologists, technologist supervisors, technologists, and information technology specialists continued to examine the workflow for underlying causes. Interventions decreasing the RR were designed and implemented. RESULTS: The initial RR of digital chest radiographs was 3.6% (138/3818) for portable CR and 13.3% (476/3575) for installed DR systems. By combining RR measurement with workflow analysis, targets for technical and teaching interventions were identified. The interventions decreased the RR to 1.8% (81/4476) for portable CR and to 8.2% (306/3748) for installed DR. CONCLUSIONS: We found the RR of direct digital chest radiography to be significantly higher than that of computed chest radiography. We believe this is due to the ease with which repeat images can be obtained and discarded, and it suggests the need for ongoing surveillance of RR. We were able to demonstrate that strategies to lower the RR, which had been developed in the era of film-based imaging, can be adapted to the digital environment. On the basis of our findings, we encourage radiologists to assess their own departmental RRs for direct digital chest radiography and to consider similar interventions if necessary to achieve acceptable RRs for this modality.