| Part Ⅰ The value of "AI+Radiologists" in differentiating benign and malignant of subsolid nodules within 2cm Objective: Artificial intelligence(AI)software combined with radiologists were used to extract the qualitative and quantitative CT image features of ≤ 2cm Sub Solid Nodules(SSNs),and to evaluate the value of "AI+Radiologists" combined diagnosis in differentiating benign and malignant subsolid nodules within 2cm.Materials and Methods: From January 2017 and October 2020,retrospective analysis clinical data and the qualitative and quantitative imaging characteristics of 118 SSNs included in the study in the Second Clinical College of Yangtze University.Among them,25 cases showed definite improvement in absorption during the follow-up,and 93 cases underwent surgical resection and disease examination.If postoperative pathology is lung Adenocarcinoma,lung Adenocarcinoma should be classified according to the multidisciplinary classification of lung Adenocarcinoma in 2011,that is,Atypicaladenomatous Hyperplasia(AAH)and Adenocarcinoma in Situ(AIS),Minimally Invasive Adenocarcinoma(MIA)and Invasive Adenocarcinoma(IAC).According to postoperative pathology and follow-up results,all SSNs were divided into benign group(28 cases)and malignant group(90 cases).Benign group included 25 cases of infectious lesions and 3 cases of inflammatory pseudotumor.Malignant group included 11 cases of AAH 24 cases of AIS 3 cases of MIA 52 cases of IAC.The clinical data,qualitative and quantitative variables of SSNs in the benign group and malignant group were compared and analyzed.The qualitative and quantitative variables of SSNs were extracted by AI software combined with radiologists.For statistically significant variables,multiple logistic regression analysis and receiver operating characteristic(ROC)curve were used to evaluate the diagnostic value.Results:(1)In clinical data,the difference of age between the patients was statistically significant between the benign group and the malignant group,and the average age of the malignant group was higher than that of the benign group.Among the qualitative variables,the spiculation sign,air bronchogram,vascular convergence signs and pleura retraction of SSNs differed statistically between the two groups,and the probability of these characteristics appearing in the malignant group is greater than that of the benign group.Among quantitative variables,the differences between the 3DCT value-median,the 3DCT value-bias,the 3DCT value-energy in the benign group and the malignant group are statistically significant,and the 3DCT value-median,3DCT value-bias,3DCT value-energy of the malignant group SSNs are less than the benign group.(2)Multiple logistic regression analysis showed that the vascular convergence sign and patients’ age were an independent risk factor for the differentiation of malignant SSNs.Risk ratios were 1.10(95%CI,1.01～1.20;P = 0.036),30.72(95%CI,3.12～302.92;P = 0.003).3DCT value-median was a protective factor,with a hazard ratio of 0.99(95%CI,0.98～1.00;P = 0.004).(3)The diagnostic model of "AI+Radiologists" was established by combining patient age and vascular convergence signs,and the area under the curve(AUC)of diagnostic model was 0.868,and the diagnostic sensitivity and specificity were 75.6% and 89.3%,respectively.And the cut-off value of age and 3DCT value-median for identifying benign and malignant SSNs was 47.5 years and-610 HU,respectively.Conclusion: The extraction of CT image characteristics of ≤ 2cm SSNs by "AI+Radiologists" helps to identify benign and malignant.The combined diagnostic model of "AI-physician" showed that the patient’s age of ≥ 47.5 years old,SSNs pathology with vascular convergence signs was more likely to be malignant SSNs,while the 3DCT value-median ≥-610 HU SSNs were less likely to be malignant,which contributed to the clinical management and prognosis assessment of ≤ 2cm SSNs.Part Ⅱ The value of "AI+Radiologists" in predicting the invasive degree in pulmonary denocarcinoma of subsolid nodules within 2cm Objective: Artificial intelligence(AI)software combined with radiologists were used to extract the qualitative and quantitative CT image features of ≤2cm Sub Solid Nodules(SSNs),and to evaluate the value of "AI+Radiologists" combined diagnosis in predicting the invasive degree in pulmonary denocarcinoma of subsolid nodules within2 cm.Materials and Methods: From January 2017 and October 2020,retrospective analysis clinical data and the qualitative and quantitative imaging characteristics of 90 SSNs included in the study in the Second Clinical College of Yangtze University.According to postoperative pathology,the nodules were divided into AAH-MIA group(11 cases of AAH,24 cases of AIS,3 cases of MIA)and IAC group(52 cases of IAC).The qualitative and quantitative variables of SSNs were extracted by AI software and reviewed by radiologists.The clinical data,the qualitative and quantitative variables of the AAH-MIA group and the IAC group SSNs were compared and analyzed.The qualitative and quantitative variables of SSNs were extracted by AI software combined with radiologists.For statistically significant variables,multiple logistic regression analysis and ROC curve were used to evaluate the diagnostic value.Results:(1)In clinical data,the age of the patients and the type of nodules were statistically significant between the AAH-MIA group and the IAC group,and the average age of the IAC group was higher than that of the AH-MIA group and SSNs in IAC group were more likely to show PSNs.Among the qualitative variables,the spiculation sign,vacuole sign,air bronchogram signs,vascular convergence signs and pleura retraction of SSNs differed statistically between the two groups,and these features were more likely to occur in the IAC group.Among the quantitative variables,1D-longest diameter,1D-short diameter,2D-maximum cross-sectional area,2D-ROI mean CT value,3D-volume,3DCT value-maximum,3DCT value-mean,3DCT value-median,3DCT value-standard deviation,3DCT value-entropy showed statistically significant differences between the AAH-MIA group and the IAC group,and these two fixed characteristics were greater in the IAC group than that in the AAH-MIA group.(2)Multiple logistic regression analysis showed that the type of nodules,the vascular convergence sign and 2D-ROI mean CT value were an independent risk factor for the differentiation of the IAC group SSNs.And risk ratios were 8.27(95%CI,1.20～57.00;P = 0.032),1.71(95%CI,1.15～2.54;P = 0.008),1.01(95%CI,1.00～1.02;P = 0.028).(3)The diagnostic model of "AI+Radiologists" was established by combining the type of nodules,1D-longest diameter and 2D-ROI mean CT value.And the AUC of diagnostic model was 0.853,and the diagnostic sensitivity and specificity were 82.2% and 89.0%,respectively.And the ROC curve analysis showed the cut-off value of 1D-longest diameter and 2D-ROI mean CT value for identifying the AAH-MIA group and the IAC group was 13.5mm and-541.5HU,respectively.Conclusion: The extraction of CT image characteristics less than 2cm SSNs by "AI+Radiologists" helps to predict the degree of invasion of lung adenoma.The combined diagnostic model of "AI+Radiologists" showed that the longest diameter of nodules ≥ 13.5mm,average 2D-ROI CT value ≥-541.5HU and PSN nodules are more likely to be IAC pathological subtype,which is of great value in guiding clinical treatment and judging prognosis. |
PDF Full Text Request