Key points are not available for this paper at this time.
Abstract Background Pulmonary nodule detection in CXR is challenging. Recently, the use of artificial intelligence (AI) has been a major attraction. The current study aimed to evaluate the diagnostic performance of the AI in the detection of pulmonary nodules or masses on CXR compared to the radiologist’s interpretation and to assess its impact on the reporting process. The current study included 150 patients who had CXR interpreted by radiologists and by AI software. Results CT detected pulmonary nodules in 99 cases (66%) while the visual model of analysis, as well as AI, detected nodules among 92 cases (61.3%) compared to 93 (62%) cases detected by combined visual/AI model. A total of 216 nodules were detected by CT (64.4% solid and 31.5% GG). Only 188 nodules were detected by the AI while 170 nodules were detected by visual analysis. As per case classification or nodule analysis, the AI showed the highest area under curve (AUC) (0.890, 95% CI) and (0.875, 95% CI), respectively, followed by the combined visual/AI model. Regarding the nodules’ texture, the AI model’s sensitivity for solid nodules was 91.4% which was greater than the combined visual/AI and visual models alone, while in GG nodules, the combined visual/AI model’s sensitivity was higher than the AI and visual models. The probability score using the combined visual/AI model was significantly higher than using the visual model alone ( P value = 0.001). Conclusions The use of the AI model in CXR interpretation regarding nodules and masses detection helps in more accurate decision-making and increases the diagnostic performance affecting the patient’s morbidity and mortality.
Ibrahim et al. (Fri,) studied this question.