| ObjectiveThe accurate diagnosis of rib fractures is of great significance for clinical treatment and judicial practice,and previous studies have established the gold standard based on early CT results,and there is a personal subjective bias,which may cause certain errors.This study aims to establish the gold standard for comparing the diagnostic results of early CT and follow-up CT to obtain better diagnostic accuracy,and to explore the clinical application value of artificial intelligence(AI)-assisted detection technology in the diagnosis of rib fractures by early CT.Materials and MethodsThe imaging data of adult patients who underwent early CT and follow-up CT scans by multi-slice helical CT from January 2018 to May 2022 were collected,and after screening by inclusion criteria and exclusion criteria,a total of 203 patient image data were included as test set data,and all enrolled case data were read by four reading methods.The patients were divided into three age groups:young group(≤39 years old),middle-aged group(40-59 years old),and elderly group(≥60 years old),and the ribs were divided into four parts:rib-cartilage junction,anterior rib segment,lateral rib segment,and posterior rib segment,and rib fracture was divided into two fracture types:fresh fracture and old fracture.Two senior radiologists compared early CT and follow-up CT and confirmed 781 fractures as the gold standard for rib fractures.Using Python programming language as the main data analysis tool,the diagnostic accuracy of four reading methods for rib fracture was calculated,and the detection rate differences between different reading methods in different age groups,different rib parts and different rib fracture types were compared,and the differences in reading time and diagnostic confidence score between independent reading and AI+physician were analyzed.Results(1)The AUC of AI independent reading for rib fracture diagnosis was 0.833(95%CI:0.798～0.885),sensitivity was 52.50%,specificity was 97.85%,missed diagnosis rate was47.50%,accuracy was 95.92%,positive likelihood ratio(LR~+)was 24.41,and negative likelihood ratio(LR~-)was 0.485.The AUC of the physician independent reading for rib fracture diagnosis was 0.827(95%CI:0.755～0.899),sensitivity was 57.49%,specificity was 98.81%,missed diagnosis rate was 42.51%,accuracy was 97.07%,LR~+was 48.22,and LR~-was 0.43.The AUC of AI+physicians for rib fracture diagnosis was 0.904(95%CI:0.870～0.939),with a sensitivity of 77.59%,specificity of 99.31%,missed diagnosis rate of22.41%,accuracy of 98.47%,LR~+was 112.45,and LR~-was 0.226.The AUC of physician+reviewer for rib fracture diagnosis was 0.842(95%CI:0.775～0.891),sensitivity was62.61%,specificity was 98.35%,missed diagnosis rate was 37.39%,accuracy was 96.87%,LR~+was 38.01,and LR~-was 0.38.(2)The detection rate of rib fractures in different age groups,AI independent reading had the highest detection rate in the middle-aged group(99.4%),AI+physician had the highest detection rate in the younger group(100%),and physician+reviewer had the highest detection rate in the elderly group(95.4%).The highest detection rate of physician independent reading in each age group was 87.7%,and there was no significant difference in detection rate between age groups(P=0.212>0.05).(3)The detection rate of fractures in different rib parts was the highest in AI independent reading and physician independent reading,with 94.8%and 96.0%,respectively;AI+physician had the highest detection rate of 95.9%in the posterior rib segment,AI+physician had no significant difference in the detection rate of fractures between the rib-cartilage junction and the anterior rib segment,and between the lateral rib segment and the posterior rib segment(P>0.05),and the physician+reviewer had the highest detection rate of 94.9%in the anterior rib segment.(4)The detection rate of different fracture types,AI independent reading detected 57.7%of fresh fractures and 56.0%of old fractures,and the difference in detection rate between groups was not statistically significant(P>0.05).The detection rate of fresh fractures and41.6%of old fractures was detected by physician independent reading,and the difference in detection rate between groups was statistically significant(P<0.05);AI+physician detected 89.2%of fresh fractures and 83.5%of old fractures,and there was no significant difference in detection rate between groups(P<0.05);physician+reviewer the detection rate of fresh fractures in 98.2%,and detected 97.4%of old fractures,and the difference in detection rate between groups was not statistically significant(P>0.05).(5)The average reading time of physician was 448±80.3(s/case),and the average reading time of AI+physician was 206.5±59.0(s/case),and the average reading time of AI+physician was shortened by about 53.8%.The average reading time of the two reading methods was significantly different(P<0.001).(6)The average diagnostic confidence score of physician independent reading was3.99±0.75(points/fracture),and the average diagnostic confidence score of AI+physician was 4.47±0.91(points/fracture),and the diagnostic confidence scores of the two reading methods were significantly different(P<0.05).(7)Most(71.0%)missed rib fractures were accompanied by visible fractures on the same or adjacent first to second rib.Conclusion(1)AI+physicians significantly improve the accuracy,sensitivity,specificity and positive predictive value of rib fracture diagnosis,reduce the missed diagnosis rate and misdiagnosis rate of fracture,and the ROC curve analysis shows that AI+physicians have good diagnostic value in rib fracture detection,and perform best in all judgment indicators,and the study shows that AI as an auxiliary tool has important clinical significance in rib fracture diagnosis.(2)As an auxiliary tool for rib fracture diagnosis,AI can significantly improve the detection rate of fractures by attending radiologists at all ages and at different rib locations.(3)Compared with independent reading by physicians,AI+physicians can shorten the reading time of rib fractures by an average of 53.8%,and AI as an auxiliary detection tool can improve the diagnostic efficiency of rib fractures,improve the confidence of attending physicians in the diagnosis of rib fractures,and help provide clear diagnostic results.(4)Most(71.0%)missed rib fractures with definite fractures on the same or adjacent ribs 1 to 2 and were occult fractures on early CT,possibly due to limitations of CT techniques. |
PDF Full Text Request