Objective:To investigate the performance,sensitivity,and causes of misdiagnosis and missed diagnosis of AI in the detection of rib fractures.Methods:This study retrospectively analyzed 100 patients diagnosed with rib fractures by CT in our hospital.The uAI Discover artificial intelligence software(uAI)was used for automatic identification of rib fractures.Two radiologists(doctor A,doctor B)conducted diagnoses separately,and then re-diagnosed using the uAI software.The joint reading results of two senior radiologists were used as the"gold standard"for diagnosis.Diagnostic errors included:AI or doctors misdiagnosing a site as a fracture or misjudging the fracture type compared to the gold standard.If the actual fracture site and type were consistent with the gold standard,the diagnosis was considered correct.Results:A total of 424 rib fractures were detected in 100 patients,involving 563 sites,including 131 dislocated fractures,218 non-dislocated fractures,and 214 old fractures.Doctor A and doctor B had 39 and 31 diagnostic errors,respectively,when diagnosing independently.The accuracy of uAI for the three types of fractures was higher than that of the two doctors.Doctors A and B had higher diagnostic accuracy with the assistance of uAI compared to using uAI alone.uAI had 49 diagnostic deviations,including 14 missed diagnoses,12 false positives(mainly non-dislocated fractures),and 23 classification errors(mainly old fractures).The missed diagnosis rates of uAI were 4.27%for dislocated fractures,3.67%for non-dislocated fractures,and 1.52%for old fractures.Conclusion:AI has potential in the diagnosis of rib fractures.By optimizing algorithms,strengthening machine learning,verification,and collaboration with doctors,AI is expected to play a greater role in clinical practice in the future.
artificial intelligencerib fracturesdiagnostic accuracymisdiagnosis and missed diagnosisgold standard
维普
万方数据
