| Part one To compare the detection efficiency of artificial intelligence-assisted diagnostic software and radiologists for pulmonary nodules with long diameter≥5mmObjective: To compare the detection performance of artificial intelligence-assisted pulmonary nodule diagnosis software(United Imaging Intelligent pulmonary nodule CT image auxiliary detection system,hereinafter referred to as AI)and radiologists for≥5mm nodules with different sizes,densities and locations,and to explore the influencing factors that affect the detection results of doctors.Methods:The 1mm thin-slice chest CT image data of 649 individuals who underwent physical examination in the Fourth Hospital of Hebei Medical University from October to December 2017 were retrospectively analyzed.According to the gold standard,441 positive nodules of 316 subjects were determined as the research object.The AI detection method(AI group)and the physician detection method(physician group)were used to analyze the detected pulmonary nodules,and the detection rate,positive predictive value and the number of misdiagnosed nodules per patient of the two methods were calculated.Results: There were 441 nodules confirmed by the gold standard,including 101 in the inner band,154 in the middle band,and 186 in the outer band.There were 378 nodules with 5mm≤long diameter<8mm,43 nodules with 8-10 mm and 20 nodules with 10mm<There were 280 solid nodules,146 pure ground glass nodules and 15 Part-solid nodules.A total of 316 nodules were detected in the radiologi-st group,including 301 true positive nodules and 15 false positive nodules.The overall detection rate was 68.3%,the positive predictive value was 95.3%,and the number of misdiagnosed nodules per person was 0.05.597 nodules were detected in the AI group,including 429 true positive nodules and 168 false positive nodules.The overall detection rate was 97.3%,the positive predictive value was 72.9%,and the number of misdiagnosed nodules per person was 0.53.The overall detection rate of AI group was 29% higher than that of the physician group,but the number of false positive nodules in the physician group was less than that in the AI group,and the positive predictive value of nodules was higher than that in the AI group.The detection rate of pulmonary nodules in the AI group was higher than that in the physician group in 5mm≤long diameter<8mm,solid and pure ground glass nodules,and in each location,and the differences were statistically signi-ficant(P<0.05).Solid nodule density,pure ground glass nodule densit-y,short diameter of nodule and nodule located in the outer zone were independent influencing factors for nodule detection in the radiologist group.Conclusions: The density,short diameter and location of nodules can all affect the performance of doctors in detecting nodules.AI can still play an important auxiliary role in improving the detection efficiency of doctors within the range of nodules with long diameter≥5mm.Meanwhile,in order to reduce the excessive detection of AI false nodules,doctors comprehensive judgment is indispensable.Part two To compare the detection ability of artificial intelligence-assistd diagnostic software for pulmonary nodules under 5mm and 1m m reconstruction thickness and reconstruction intervalObjective: To compare the differences in the detection rate of pulmonary nodules,the determination of the long diameter,short diameter,volume and density of the detected nodules,and the average number of misdiagnosed nodules under different reconstruction slice thickness and different reconstruction intervals by artificial intelligences assisted diagnosis software for pulmonary nodules(United Image intelligent pulmonary nodule CT image auxiliary detection system,AI).Methods:A total of 31 cases of physical examination in September 2022 were retrospectively selected.The chest CT image data of the subjects were analyzed under different reconstruction slice thickness and reconstruction interval of the thin and thick groups(thin group(group A): the reconstruction slice thickness and reconstruction interval were 1mm;Thick layer group(group B): The location,density,size(long diameter,short diameter)of the detected nodules in each group were recorded,and the number of false positive nodules was marked.The results of combined with the thin-layer group reading were used as the gold standard.The interval between the two groups was 2-3 weeks during the washout period.Results: There were 150 non-calcified nodules confirmed by gold standard,including 53 nodules with long diameter ≥5mm,32 nodules with long diameter > 5mm,68 solid nodules,79 pure ground glass nodules and 3 partially solid nodules.The overall detection rate of nodules in thin layer group,the detection rate of nodules with long diameter ≥5mm and long diameter > 5mm were 100%.In the thick layer group,the overall detection rate of nodules,the detection rates of nodules with length ≥5mm and diameter > 5mm were 56.0%,81.1% and 93.8%,respectively.The overall nodular detection rate and the nodular detection rate of long diameter ≥5mm in thin layer group were higher than those in thick layer group,and the differences were statistically significant(P < 0.05).In the range of long diameter > 5mm nodular detection rate,there was no significant difference between the two groups(P > 0.05).There was a good consistency in the determination of detected nodule density between the two groups(Kappa value 0.647),and there was no statistical difference in the determination results of detected nodule long diameter and short diameter(P > 0.05),The nodule volume of the thick layer group was higher than that of the thin layer group,and there was a significant difference between the two groups(P < 0.05).The number of misdiagnosed nodules per capita in the thin layer group was 0.7 / person(22/31),and the number of misdiagnosed nodules per capita in the thick layer group was 1.5 / person(46/31),and there was a statistical difference between the two groups(P < 0.05).Conclusions: Thick-layer images can be used to observe the size and density of nodules with diameter>5mm,but volume measurement is not recommended for follow-up nodules. |
PDF Full Text Request