Abstract 136TiP
Background
The CT values assigned to pulmonary nodules bear pivotal prognostic significance in distinguishing between benign and malignant lung nodules and gauging their invasiveness. Higher average CT values of nodules have been demonstrated to correlate with heightened levels of infiltration. However, the substantial influence of pulmonary vascularization within these nodules obfuscates accurate CT value detection. To enhance the diagnostic precision for differentiating benign from malignant pulmonary nodules and to refine assessments of their infiltrative characteristics, this research posits the application of AI-driven devascularization through Synapse 3D software. This innovative approach seeks to eliminate the confounding effect of pulmonary vascular shadows within CT images, thereby augmenting the predictive potential of CT values in the diagnosis of pulmonary nodules.
Trial design
The clinical records of patients who underwent surgical resection for lung cancer at the Second Affiliated Hospital of Nanchang University between March 2019 and December 2023 were subjected to retrospective analysis. The average CT values pertaining to pulmonary nodules, both in their original state and subsequent to the application of vascularization elimination, were computationally ascertained via the employment of Synapse 3D software. Consequent to the software's application, each nodule underwent a process of AIassisted de-vascularization, facilitated by the advanced capabilities of Synapse 3D. The appraisal of the predictive efficacy inherent to the devised models was undertaken by means of receiver operating characteristic (ROC) curves, while the areas beneath these curves (AUCs) furnished a comprehensive gauge of performance. Sensitivity, specificity, accuracy, positive predictive value, and negative predictive value were computed to assess the model's discerning capabilities. Two sets of data, mean CT values and mean CT values after vascular isolation of pulmonary nodules, were used to build separate diagnostic models, and the areas under the ROC curves were calculated and compared between the two sets of models.
Legal entity responsible for the study
The authors.
Funding
This research was funded by the National Natural Science Foundation of China (81860379, 82160410 and 81560345). Science and Technology Planning Project at the Department of Science and Technology of Jiangxi Province, China (20171BAB 205075 and 20162BCB 23058). Jiangxi Provincial Science and Technology Department Key Projects [grant number 20212ACB206018]. Jiangxi Provincial Science and Technology Department Key R&D Program [grant number 20223BBG71009]. Jiangxi Postgraduate Innovation Special Fund Project (YC2022—s226).
Disclosure
All authors have declared no conflicts of interest.