Effect of Irreversible Compression on the Pulmonary Nodule Detection Rate in Chest Radiographs Using AI Software

Objectives: Whereas irreversible compression of Digital Imaging and Communications in Medicine (DICOM) files can reduce data size, research on its impact on diagnostic ability when using artificial intelligence (AI) software is limited. The objective was to determine the effect that irreversible compression has on diagnostic ability when using AI software. In addition, the effect of nodal properties on computed tomography (CT) on detection rates was examined. Methods: A total of 335 patients with pulmonary nodules were included. Chest radiographs were subjected to irreversible compression at 10:1 and 50:1 ratios. The associations between the detection rate of the AI software and factors such as location on CT, morphology, and diameter, were determined. Results: The number of positive cases identified with the AI imaging software was as follows: 188 cases (56.1%) with no compression, 184 cases (54.9%) with 10:1 compression, and 175 cases (52.2%) with 50:1 compression. There was a significant difference between the uncompressed images and the 50:1 compressed images, as well as between the 10:1 compressed images and the 50:1 compressed images (all p < 0.05). With all compression ratios, there were significant differences in the associations between the AI software’s nodule detection rate and the target nodule’s maximum diameter, minimum diameter, morphology, and overlap with multiple organs on CT (all p < 0.0001). Conclusions: The detection rate by the AI software of lung tumors on chest radiographs showed no significant difference when images were subjected to 10:1 irreversible compression; however, there was a significant decrease when subjected to 50:1 irreversible compression.