| Objectives:1.To explore the qualitative diagnostic value of radiomics based on Plain Computed Tomography(CTN),enhanced CT(CT artery(CTA)combined CT venous(CTV)),and plain CT combined with enhanced CT(CTN+CTA+CTV)models for laryngeal squamous cell carcinoma(SCC)and squamous cell hyperplasia(SCH).2.To explore the qualitative diagnostic value of radiomic based on the enhanced T1 weighted image(T1WI)model,T1 WI +T2 weighted image(T2WI)model,and enhanced T1WI+ T1 WI +T2WI model for laryngeal squamous cell carcinoma(SCC)and squamous cell hyperplasia(SCH).3.Based on the above studies,the diagnostic efficacy of the CT radiomic models and the magnetic resonance imaging(MRI)radiomic models were evaluated to provide a reference for clinical diagnosis.4.To explore the predictive ability of radiomics models based on the diffusionweighted image(DWI)model,T1WI+T2WI model,and DWI+T1WI+T2WI model to diagnose local tumor recurrence from radiation-induced changes.Methods:1.Retrospective analysis was performed on 254 patients who underwent plain CT,enhanced laryngeal examination,and postoperative pathological examination in our hospital from June 2017 to December 2022,including 154 cases of SCC and 100 cases of SCH.CT plain scan sequence,arterial phase sequence,and venous phase sequence images of 254 patients were imported into the Insight Toolkit(ITK-SNAP)software,and area of interest lesions(ROI)were delineated layer by layer manually.Based on the first-order histogram feature,texture feature,morphological feature,and wavelet feature,the related radiomics parameters of lesions were extracted.R-studio software was used to randomly divide the training and testing group according to the ratio of 7:3.The heatmaps and least absolute shrinkage and selection operator(LASSO)methods were used for dimensionality reduction to obtain the optimal characteristic parameters,establish the prediction model,and draw the area-under-the-receiver operating characteristics-curve(AUC),and decision curve analysis(DCA)to evaluate the performance of the diagnostic radiomic.2.Retrospective analysis was performed on the patients who received plain MRI and enhancement and preoperative pathological examination in our hospital from June2017 to December 2022.A total of 247 patients were included in this study,including154 patients with SCC and 93 patients with SCH.T1,T2,and enhanced T1 sequence images of 247 patients were imported into ITK-SNAP software,and the ROI of lesions was segmented on axial,coronal,and sagittal planes at each level.The Artificial Intelligence Kit(A.K.)software extracts radiomics feature parameters from 3D ROI images obtained after segmentation.R-studio software was used to randomly divide the training group and the test group according to the ratio of 7:3.The radiomic heatmap and LASSO methods were used to reduce the dimension,obtain the optimal parameter characteristics,establish the radomics diagnosis model,and verify the differential diagnosis efficiency of radiomic by statistical methods.3.The AUC statistical method used R-studio software to verify the differential diagnosis efficiency of radiomic based on the above CT and MRI radiomics models.4.A retrospective analysis was performed on the patients who received radiotherapy and postoperative pathological examination from August 2015 to December 2022.A total of 133 patients were included in this study,including 51 patients with local recurrence and 82 patients with changes post-radiotherapy.T1,T2,and DWI sequences of 133 patients were imported into ITK-SNAP software,and the ROI of lesions was segmented based on the axial plane.A.K.software extracts radiomics feature parameters from ROI images obtained after segmentation.R-studio software was used to randomly divide the training group and the test group according to the ratio of 7:3.The radiomic heatmap and LASSO methods were used to reduce the dimension and obtain the optimal parameter characteristics,then establish the radiomics diagnosis model based on the extracted features,and lastly verify the differential diagnosis efficiency of radiomic by statistical methods.Results:1.The data of all patients in the training and testing cohorts were consistent at baseline(P>0.05).After analyzing the clinical imaging data,we noticed no differences in the clinical data(age,gender,smoker status,tumor location,and alcohol consumption)for training and testing groups statistically.The P-values for the clinical data were not significant.The predictive performance for the three CT radiomics models was good in discriminating between SCC and SCH of laryngeal carcinoma.Radiomics model established based on plain CT combined with enhanced CT,AUC value,accuracy,sensitivity,specificity,positive predictive value,and negative predictive value were0.985(95% Confidence interval(CI): 0.971～0.999),0.965(95% CI:0.933～0.997),0.944,0.857,0.953,0.894,0.929,0.8,0.953,0.875,0.929,0.828;in the training group and validation group,respectively.To establish a radiomics model based on enhanced CT(CTA+CTV),AUC value,accuracy,sensitivity,specificity,positive predictive value,and negative predictive value are 0.965(95% CI:0.943～0.983),0.902(95% CI:0.846～0.95),0.91,0.805,0.916,0.851,0.9,0.733,0.933,0.833,0.875,0.759,in the training and validation groups,respectively.Based on the radiomics model established during the CTN,AUC value,accuracy,sensitivity,specificity,positive predictive value,and negative predictive value were 0.88(95% CI:0.839～0.919),0.852(95% CI:0.779～0.914),0.785,0.792,0.645,0.66,1.000,1.000,1.000,1.000,0.648,0.652 in the training group and the validation group,respectively.2.The data of all patients in the training and testing cohorts were consistent at baseline(P>0.05).After analyzing the clinical imaging data,we noticed no differences in the clinical data(age,gender,smoker status,tumor location,and alcohol consumption)for training and testing groups statistically.In contrast,we noticed significant differences in the clinical data(smoker status and tumor size)for training and testing groups.The predictive performance of the three MRI radiomics models had been promising in discriminating between SCC and SCH of laryngeal carcinoma.Radiomics model established based on enhanced T1WI+T1WI+T2WI,AUC value,accuracy,sensitivity,specificity,positive predictive value,and negative predictive value were0.958(95%CI: 0.933-0.983),0.878(95%CI: 0.803-0.952),0.884,0.792,0.907,0.766,0.846,0.821,0.907,0.878,0.846,0.676;in the training group and validation group,respectively.To establish a radiomics model based on enhanced T1 WI,AUC value,accuracy,sensitivity,specificity,positive predictive value,and negative predictive value are 0.912(95%CI: 0.872-0.953),0.863(95%CI: 0.781-0.944),0.791,0.787,0.804,0.723,0.769,0.893,0.851,0.919,0.704,0.658,in the training and validation groups,respectively.Based on the radiomics model established during the T1WI+T2WI,AUC value,accuracy,sensitivity,specificity,positive predictive value,and negative predictive value were 0.916(95%CI: 0.877-0.955),0.841(95%CI: 0.752-0.930),0.831,0.76,0.804,0.745,0.877,0.786,0.915,0.854,0.731,0.647 in the training group and the validation group,respectively.3.The AUC value of the radiomic model based on enhanced T1WI+T1WI+T2WI sequence was significantly higher than other radiomics models based on CTN,CTA+CTV,and CTN+CTA+CTV,as well as higher from those other radiomics models based on enhanced T1 WI and T1WI+T2WI.4.The data of all patients in the training and testing cohorts were consistent at baseline(P>0.05).After analyzing the clinical imaging data,we noticed no differences in the clinical data(age,gender,smoker status,tumor location,and alcohol consumption)for training and testing groups statistically.The P-values for the clinical data were not significant.The radiomics model based on the DWI sequence combined with T1WI+T2WI sequences obtained the highest efficiency in the diagnosis of local tumor recurrence from radiation-induced changes compared to other models,where the values of AUC,accuracy,sensitivity,specificity,positive predictive,and negative predictive were 0.965(95%CI: 0.924-1.000),0.940(95%CI: 0.855-1.000),0.892,0.90,0.877,0.917,0.917,0.875,0.943,0.917,0.875,and 0.825,in the training group and testing group,respectively.Additionally,the radiomics model established based on the DWI sequence achieved good performance,where the values of AUC,accuracy,sensitivity,specificity,positive predictive,and negative predictive were 0.925(95%CI: 0.867-0.982),0.888(95%CI: 0.783-0.994),0.871,0.825,0.86,0.792,0.889,0.875,0.925,0.905,0.737,and0.80;in the training group and validation group,respectively.While a radiomic model based on the T1WI+T2WI model achieved satisfactory performance but lower than other models,where the values of AUC,accuracy,sensitivity,specificity,positive predictive,and negative predictive were 0.878(95%CI: 0.810-0.946),0.760(95%CI:0.611-0.910),0.806,0.70,0.789,0.625,0.833,0.812,0.882,0.833,0.714,0.591,in the training and validation groups,respectively.Conclusions:1.The radiomics models based on CT sequences have a good differential diagnosis ability between SCC and SCH of laryngeal carcinoma,but the diagnostic efficiency of the three models is different.The radiomics model based on plain CT combined with enhanced CT is superior to other imaging radiomics models in the differential diagnosis of laryngeal carcinoma.The radiomics model based on CT has realized the accurate non-invasive diagnosis of SCC and SCH,which is helpful in developing personalized diagnosis and treatment plans for patients,improving the survival rate of patients and prolonging the survival time of patients.2.The establishment of radiomic models based on enhanced T1 WI,T1WI+T2WI,and enhanced T1WI+T1WI+T2WI have good differential diagnosis ability for SCC and SCH.In comparison,the Radscore model based on the enhanced T1WI+T1WI+T2WI model has relatively high performance in differential diagnosis,which is superior to the radiomics models based on enhanced T1 WI,T1WI+T2WI.The MRI-based radiomics model can improve the diagnostic accuracy of patients with laryngeal lesions,provide valuable preoperative information for clinical diagnosis and treatment,and guide clinicians to develop diagnosis and treatment plans in line with patients.3.The effectiveness of the above radiomics models based on CT and MRI was evaluated for differential diagnosis.Statistical analysis showed that the AUC value of the radiomic model based on enhanced T1WI+T1WI+T2WI was significantly higher than that of the radiomics based on CT sequences and radiomics based on enhanced T1 WI and T1WI+T2WI.It has good application prospects in the differential diagnosis of laryngeal squamous cell carcinoma and squamous cell hyperplasia.4.The construction of radiomics model based on T1 WI,T2WI,and DWI sequences has a good differential diagnosis ability for local tumor recurrence and radiation-induced changes.The radiomics model based on the DWI+T1WI+T2WI sequences is superior to other imaging radiomics models in discriminating local recurrence from radiation-induced changes.These models have been developed to help physicians detect local recurrence in the least possible time to save time and protect patients from disease complications.Finally,these experiments proved their efficiency in detecting local recurrence of laryngeal carcinoma through employed the machine learning method with MRI images.Hence,we recommend using the computer-aided diagnosis(CAD)technique because it has proven its efficiency and accuracy in diagnosing local recurrence and is an easy-to-use method. |
PDF Full Text Request