Investigation Of Pulmonary Ground-glass Density Nodules By DL-CAD And MSCT

Part ? The Value of Deep-learning based Computeraided Diagnosis system in Differential Diagnosis of Pulmonary Ground-glass nodulesObjective : To investigate the value of deep-learning-based computer-aided diagnosis(DL-CAD)system in differentiating ground-glass nodules(GGNs)with different histopathology.Methods : From January 2017 to December 2019,128 patients with 167 pulmonary GGNs detected during screening test in our hospital were enrolled.Based on the histopathology result,these nodules were divided into malignant group(n =153),including invasive adenocarcinoma(n = 99),minimally invasive adenocarci-noma(n = 13),adenocarcinoma in situ(n = 22),and atypical adenomatous hyperplasia(n = 19)and benign group(n = 14),including foreign body granulomas(n= 1),alveolar epithelial hyperplasia with interstitial cell infiltration(n = 13).All patients underwent multi-slice spiral CT(MSCT)examination.DL-CAD automatically detected and rated each nodule with malignancy risk.Using pathological results as the gold standard for diagnosis,the diagnostic efficacy of the DL-CAD system in differentiating benign and malignant nodules was analyzed statistically.Results:Among the 167 targeted GGNs,the DL-CAD successfully detected 166 GGNs and missed 1 GGN.With pathology as reference,DL-CAD corrected 140 out of 167 GGNs with match rate of 83.83%.In addition,the DL-CAD showed significantly better performance in diagnosing mixed ground-glass density nodules(mGGN)than pure ground-glass density nodules(pGGN)(91.67% VS 80.19 %,P =0.027).Conclusions:The DL-CAD has certain value in the diagnosis of GGNs.Despite better performance in diagnosing mGGN rather than pGGN,the risk still exists for DL-CAD to misdiagnose and miss pulmonary nodules,so that when interpreting its result,clinicians need comprehensive consideration in clinical scenario.Part ? Diagnostic Value of common Multi-slice CT Imaging Findings in Classifying Different Subtypes of Lung adenocarcinoma with Ground-glass opacityObjective:To explore the Diagnostic value of common multi-slice CT(MSCT)imaging findings in classifying lung adenocarcinoma subtypes with ground-glass opacity.Methods: From January 2017 to December 2019,126 patients with 153 pulmonary ground-glass nodules(GGNs)in our hospital were collected,which includes invasive adenocarcinoma(n = 99),minimally invasive adenocarcinoma(n =13),adenocarcinoma in situ(n = 22),and atypical adenomatous hyperplasia(n = 19)based on its histopathology.Among them,adenocarcinoma in situ and atypical adenomatous hyperplasia were classified as pre-invasive lesions.All patients underwent MSCT examination and multiple imaging findings were obtained based on MSCT images,including location,size,shape,density,tumor-lung interface,lobulation sign,spiculation sign,vacuole sign,vascular convergence sign,pleural retraction sign,etc.The correlation between common imaging findings on MSCT and different lung ground-glass adenocarcinoma subtypes was analyzed.Results:In terms of average size,invasive adenocarcinoma(15.14 ± 6.26 mm),minimally invasive adenocarcinoma(8.71 ± 2.36 mm),and pre-invasive lesions(8.47± 3.05)mm differed significantly(P<0.001).Between invasive adenocarcinoma and minimally invasive adenocarcinoma,significant difference was found in vascular convergence signs(P <0.007),but not in nodule shape(P = 0.053),density(P=0.073),and tumor-lung interface(P= 0.681),lobulation sign(P= 0.819),spiculation sign(P = 0.417),vacuole sign(P= 0.462),and pleural retraction sign(P = 0.128).As for invasive adenocarcinoma and pre-invasive lesions,significant difference existed in nodule shape(P< 0.001),density(P< 0.001),spiculation sign(P= 0.008),vascular convergence sign(P < 0.001),pleural retraction sign(P< 0.001);but not in tumor-lung interface(P= 0.583)and lobulation sign(P = 0.220).In contrast,minimally invasive adenocarcinoma and pre-invasive lesions differed significantly in nodule density(P = 0.049),spiculation sign(P= 0.026),and vascular convergence sign(P = 0.001),but not in shape(P= 0.730),tumor-lung interface(P = 0.496),lobulation sign(P= 0.653),vacuole sign(P= 0.782),and pleural retraction sign(P=0.424).The receiver operative characteristic(ROC)analysis was used to evaluate the contribution of nodule size to differentiating invasive adenocarcinoma and minimally invasive adenocarcinoma,which showed that the optimal sensitivity and specificity of0.747 and 0.846,respectively,for diagnosing invasive adenocarcinoma was obtained when setting the threshold of 10.03 mm.Likewise,for differentiating invasive adenocarcinoma and pre-invasive lesions,the thresholds of 10.25 mm gave rise to the optimal sensitivity and specificity of 0.747 and 0.805,respectively.Conclusions: The common MSCT imaging findings have shown robust diagnostic value in differentiating ground-glass lung adenocarcinoma subtypes with satisfactory specificity,especially for imaging findings,such as vascular convergence sign and size,which played a significant role in their differential diagnosis.However,size showed no contribution to differentiating invasive adenocarcinoma and minimally invasive adenocarcinoma or pre-invasive lesions.