Background: This study aimed to compare capabilities of computer-aided volumetry (CADv) with convolutional neural network (CNN) for differentiating invasive from non-invasive adenocarcinoma groups and evaluating the outcomes of hybrid-type iterative reconstruction (IR) and deep learning reconstruction (DLR) on high-definition CT (HDCT)s at standard-dose (SDCT), reduced-dose (RDCT), and ultra-low-dose (ULDCT) levels.Methodology: 181 patients with 181 lung adenocarcinomas who underwent thin-section HDCT at SDCT, RDCT, and ULDCT levels and reconstructed using both methods were included in this study.Then, the consolidation-to-tumor ratio (CTR) on all HDCT data was measured using 3D volumetry (CTRvolume) and 2D maximum longest diameter measurement (CTRLAD) by CADv with CNN.To compare the diagnostic performance of all CTRLADs and CTRvolumes for differentiating between invasive and non-invasive cases, a receiver operating characteristic analysis was performed.Then, the sensitivity, specificity, and accuracy were compared among all HDCTs using McNemar's test.Diseasefree survival (DFS) and overall survival (OS) between two groups, divided by each CTR, were assessed using the Kaplan-Meier method followed by log-rank test.Results: Each area under the curve of CTRvolume was significantly larger than that of CTRLAD (p<0.05).Sensitivity and accuracy of CTRvolume were significantly higher than those of CTRLADs for all HDCTs (p<0.05).There were no significant differences in all DFSs or all OSs between two groups. Conclusion: CADv with CNN shows promise for accurately measuring nodule components, J o u r n a l P r e -p r o o f 5 differentiating invasive from non-invasive groups, and evaluating postoperative patient outcomes in lung adenocarcinoma on HDCT with SDCT, RDCT and ULDCT levels and reconstructed by applied two methods.
Ozawa et al. (Fri,) studied this question.