| Objective In recent years,Artificial Intelligence(AI)has been widely used in the imaging diagnosis of lung lesions(including pulmonary nodules and masses,etc.),and has obtained obvious application value in many aspects.However,currently,AI system has a high false negative rate in the diagnosis of multiple types of nodules,and cannot replace manual work in terms of accuracy.At the same time,some research results show that the type and thickness of Computed Tomography films have a certain influence on the diagnostic accuracy of AI,and the selection of the correct examination method for AI analysis is also of great value in diagnosing the nature of lesions as early as possible in clinic."Deepwise AI Medical Image Assisted Diagnosis System" is a recently developed AI-assisted image diagnosis system.This study aims to explore its diagnostic effect on lung space-occupying lesions with different scanning layer spacing and the factors affecting its diagnostic accuracy.Methods A total of 821 patients with definite postoperative pathological results who underwent surgery in the Department of Thoracic Surgery,Affiliated Hospital of Qingdao University from June 2021 to March 2022 were retrospectively collected.According to the scanning layer spacing of CT slices,they were divided into thick layer CT(≥3 mm)and thin layer CT(<3 mm).CT examination images obtained from 821 patients under two different scanning layer spacing were imported into the " Deepwise Medical AI Medical image Auxiliary diagnosis System",and the risk of nodules was judged as "low risk" or "high risk" by AI.Multi-factor analysis was performed by combining various imaging indicators and clinical characteristic values of patients.The final pathological diagnosis result of the patient’s intraoperative specimens was taken as the gold standard and compared with the corresponding diagnosis result of AI system.Results(1)The sensitivity,specificity,positive predictive value and coincidence rate of AI in evaluating lung space-occupying lesion image data with different scanning layer spacing were all high,and the Area Under Curve(AUC)of Receiver Operating Characteristic(ROC)were all over 0.90.(2)When scanning thick slice CT with layer spacing≥3 mm,the relationship between the age of patients,systolic blood pressure,tumor size and AI diagnostic accuracy was analyzed,and the results were statistically significant(P<0.05).When diagnosing thin-slice CT images <3 mm,there were statistically significant differences between the preoperative smoking history,blood transfusion history and chemotherapy history and the final diagnostic accuracy(P<0.05).(3)The diagnostic accuracy of AI and the analysis results of the basic imaging characteristics showed that the presence or absence of pleural pull in the two thickness CT slices would affect the diagnostic accuracy of AI;The length and diameter of lung space occupying lesions,edge definition and lobular presence affected the diagnostic accuracy of AI in thick slice CT(P<0.05).(4)Features such as long diameter and short diameter,maximum surface area,surface area,3D long diameter,average short diameter and long diameter,and minimum CT value of lesions on CT film measured by AI system all have significant differences in the diagnostic accuracy of AI analysis of CT film of various thicknesses(P<0.05).(5)There was significant difference in the diagnostic accuracy of AI for neuroendocrine carcinoma and Non-small-cell Lung Cancer(NSCLC)in the diagnosis of thin slice CT(P<0.05).Conclusion: "Deepwise AI Medical Image Assisted Diagnosis System" showed high sensitivity,specificity,positive predictive value,coincidence rate and AUC in evaluating the image data of lung space-occupying lesions with two different scanning layers spacing,showing a good diagnostic effect.This system can better assist the work of imaging department and thoracic surgeons to make a rapid and accurate diagnosis of lung space occupying lesions,and then develop a reasonable treatment plan for patients,but manual radiography is still an indispensable part.Overall,the system shows better diagnostic results for thin-slice CT scans less than 3 mm thick.In terms of clinical and imaging basic features and imaging texture features,the long diameter and short diameter of patients’ lesions,the presence or absence of pleural stretch,the short diameter and its average value measured by AI,the 3D long diameter,the maximum surface area,surface area,and the minimum CT value were significantly correlated with the diagnostic accuracy when the AI system analyzed CT images of various scanning thickens(P<0.05). |
PDF Full Text Request