Analysis Of The Efficacy Of Artificial Intelligence-assisted CT Examination For The Detection Of Fresh Rib Fractures

ObjectivesIn this study,Artificial Intelligence(AI)software was used to assist chest electronic meter.Computed Tomograph(CT)images automatically identify rib fractures to select the most to evaluate the detection efficiency of artificial intelligence automatic detection technology for fresh rib fracture.MethodsIn this study,we collected 352 patients with emergency chest trauma who received 16-row spiral CT under the chest scan modality(including 5mm thick layer group and 1mm thin build group after 5mm thick layer scan)and 93 patients with emergency chest trauma under the rib fracture scan modality(1mm thin scan group)in our hospital from September 2019 to September 2021,respectively,and their CT image data were used as three AI experimental group data sets.The images were read directly by the rib fracture detection aid(uAI BoneCare)and analyzed for diagnosis.The reference standards for rib fractures were developed by a panel of three senior experts at the associate director level or higher,and their common diagnostic results were used as the basis for the reference standards detected by the AI software.The first part measured and compared separately the sensitivity,specificity,accuracy,false-positive rate,false-negative rate,positive predictive value,and variability of the corresponding indexes between different levels(patient level,rib level,and lesion level)detected by the AI software for different layer thicknesses.The second part evaluates the rib fractures detected by the AI software at different levels with the same layer thickness(1 mm reconstruction)and compares the sensitivity,specificity,accuracy,false-positive rate,false-negative rate,positive predictive value and negative predictive value between different types of rib fractures(dislocation fracture,minor fracture,cortical distortion)at the focal level.The effectiveness of the AI software in automatically detecting rib fractures was evaluated.Results1.When the CT scan layer thickness was analyzed for the study:(1)The sensitivity of the AI software for rib fracture detection on direct scans with 5-mm thick layers and on later images with 1-mm thin layer reconstruction was 66.96%and 76.45%and 64.84%and 79.52%at the rib and lesion levels,respectively,with significant differences between the two scanning modalities(p=0.000),and the AI software had a 1mm thin-layer reconstructed images were detected with higher sensitivity,and there was no significant difference between the two scanning modalities at the remaining levels(p>0.05).(2)The sensitivity,specificity,accuracy,false-positive rate,false-negative rate,positive predictive value and negative predictive value of the AI software for rib fracture detection on the images of 1 mm thin-layer reconstruction and 1 mm thin-layer scan were not significantly different(p>0.05)at all three levels(patient level,rib level and lesion level).2.When analyzed with the rib fracture types studied:the AI software had different efficacy in detecting total fracture changes,misaligned fractures and non-misaligned fractures(minor fractures and cortical distortions)at the focal level,where the sensitivity of the AI software in detecting different types was 81.77%,94.76%,77.13%and 75.59%,respectively(χ2=63.77,p=0.000),positive predictive values of 75.45%,84.78%,77.48%and 69.06%(χ2=35.84,p=0.000),respectively,and false negative rates of 18.23%,5.24%,22.87%and 24.14%(χ2=62.77,p=0.000).The AI software had the highest sensitivity and positive predictive value for misaligned fracture detection among all fracture types,while the false negative rate for misaligned fracture detection was the lowest among all fracture types.Conclusion1.The rib fracture-assisted detection software has a good ability to detect rib fractures,so the imaging department can use the AI software as a primary screening detection method for acute chest trauma.2.In the detection of rib fractures by different scanning methods,1mm thin-layer reconstruction has higher sensitivity and accuracy in the detection of rib and lesion level relative to 5mm thick-layer scanning with AI software,while 1mm thin-layer reconstruction is consistent with the detection ability of 1mm thin-layer scanning with AI software,so 1mm thin-build with 5mm scanning can be the best way to detect rib fractures by AI.3.In terms of fresh rib fracture detection type,AI is most sensitive to detect misaligned rib fractures and less sensitive to minor fractures and cortical distortions.Therefore,AI software can be used as a primary screening and validation tool for rib fractures to make risk prediction of patient condition according to the specific type of fracture and thus prioritize the stratification of critically ill patients.