Can small-field-of-view reconstruction CT better evaluate lung nodule than conventional CT analyzed by multi-instance learning lung nodule system ?

Background: To subjectively and objectively assess the performance of conventional CT (c-CT) and small-field-of-view reconstruction CT (sFOV-CT) by radiologists and a multi-instance learning (MIL) based lung nodule system. Methods A cohort of 112 patients were enrolled between July 2021 to March 2022 in this prospective study. After taking c-CT examinations, the sFOV-CT images with small-field-of-view were reconstructed. All the c-CT and sFOV-CT images were subjectively analyzed by two radiologists, including features of location, nodule type, size, CT values, and shape sign. Then a lung nodule system based on MIL objectively analyzed the c-CT (c-MIL) and sFOV-CT (sFOV-MIL) to explore their difference. Additionally, the signal-to-noise ratio of lung (SNR-lung) and contrast-to-noise ratio of nodule(CNR-nodule) were calculated to evaluate the quality of CT images. Results The subjective evaluation by two radiologists showed that only the features of minimal CT value ( p  = 0.019) had statistical significance between c-CT and sFOV-CT. While, to the objective analysis by MIL system, the most features (all the p  < 0.05) except for nodule type, location, volume, mean CT value, and vacuole sign ( p  = 0.056-1.000) had statistical difference between c-MIL and sFOV-MIL. The SNR-lung between c-CT and sFOV-CT had no statistical significance, while the CNR-nodule showed statistical difference ( p  = 0.007) and the CNR of sFOV-CT was higher than that of c-CT (median, inter-quartile: 2.203, 1.620–2.793 vs. 1.850, 1.027–2.558). Conclusions The MIL system were more sensitive than radiologists’ subjective evaluation in detecting the difference between c-CT and sFOV-CT. The image quality of those two CT images were different, and the CNR-nodule of sFOV-CT was higher than that of c-CT.