Pulmonary nodule assessment by autonomous computer software and radiologists

Introduction/Aim: Pulmonary nodules are frequent incidental findings and follow‐up should facilitate early diagnosis while minimizing negative sequalae of investigation. Radiologist assessment of computed tomography (CT) images and use of clinical guidelines (e.g. Fleischner Society Guidelines; FSG) typically determine management. This study assesses the use of autonomous software aiding radiologist assessment of pulmonary nodules. Methods: CT chest scans performed for pulmonary nodule surveillance between September 2018 and March 2019 were retrospectively identified from the Gold Coast Hospital and Health Service imaging database. Data from three reports were collected: initial radiologist report (RAD), Philips Pulmonary Nodule Analysis® autonomous software report (CAD) and radiologists who reviewed CT images and CAD results (RAD+CAD). Lung cancer risk factors were collected from medical records. follow‐up based on the 2017 FSG were derived from each report. Results: CT scans of 100 patients were identified. Mean age was 68 years (SD 11 years) and 57% were female (n=54). The largest nodule's median maximal diameter as estimated by RAD and RAD+CAD were similar at 9.3 mm (n=74, IQR 6.0 mm) and 9.0 mm (n=87, IQR 5.6 mm), respectively (p=0.01). The same variable was reported by CAD as 13.9 mm (IQR 9.3 mm), different from the other modalities (p<0.001). There was agreement between RAD and RAD+CAD regarding nodule density in 45.5% (K=0.32, p=0.01), nodule spiculation in 78.6% (K=0.16, p=0.19) and lobar location in 62.1% (K=0.51, p<0.001). Guideline‐based follow‐up derived from RAD and RAD+CAD reports were different in 57% (n=57) with RAD+CAD suggesting follow‐up be earlier in 36% (n=36) and later in 21% (n=21). RAD did not report nodule volume while RAD+CAD reported it 87% (n=87). Conclusion: This pilot retrospective study provides evidence that access to autonomous software can alter the assessment of pulmonary nodules by radiologists resulting in differences in clinical follow‐up.