P78\u2005The role of computer-assisted radiographer reporting in lung cancer screening programmes

Abstract

Introduction and objectives The success of lung cancer screening (LCS) with low-dose CT (LDCT) depends critically on delivering timely, accurate radiology reports. Its anticipated widespread introduction will place a significant burden on current thoracic radiologist capacity, mandating innovative solutions. We explored the role that trained radiographers, using computer-assisted nodule detection (CADe) software, might have in LCS reporting pathways. Methods 770 LDCTs performed as part of the Lung Screen Uptake trial (LSUT) were retrospectively reported by two radiographers (R1 and R2) using the Veolity™ CADe software. Radiographers could request the opinion of the study radiologist about uncertain findings. The original radiologists’ reports (read without CADe) were considered the reference standard. Studies were categorised as ‘positive’ (nodule or mass requiring nodule surveillance or MDT referral), ‘negative’ (no intrapulmonary findings requiring further imaging) or ‘ill-defined’ (indistinct focal abnormality requiring surveillance, e.g. consolidation). Reported outcomes were compared to the reference standard, with any discrepant (i.e. radiographer-only reported) nodules re-reviewed by both study radiologist and a second independent radiologist, and verified as either ‘true’ or ‘false positive’ (figure 1). Secondary outcomes included scan-reading times and identification of incidental findings. Results The reference standard dataset included 163 ‘positive’, 35 ‘ill-defined’, and 572 ‘negative’ studies, and 34 confirmed lung cancers. R1 and R2 requested radiologist confirmation for 6.5% and 10.4% of studies respectively. Following verification of discrepant nodules, reporting sensitivity varied significantly between radiographers at 67.3% (R1) and 74.0% (R2) for all ‘positive’ studies (OR 2.27, p=0.03): 77.4% and 93.9% for confirmed cancers. The majority of ‘missed’ lesions arose from inappropriate rejection of CADe-detected findings rather than being missed altogether. The radiographer plus CADe reading combination highlighted ten nodules previously dismissed by the study radiologist, that were subsequently recalled for further surveillance. Allowing for these, the rates of false positive reporting were 7.9% (R1) and 6.2% (R2). Conclusions Individual performance varied significantly between the two radiographers, but the overall results suggest inadequate sensitivity to recommend this strategy. As per previous observations elsewhere, using CADe software in LCS reporting pathways is likely to reduce reporting times whilst increasing reader sensitivity.Abstract P78 Figure 1 LDCT reporting and management of discrepant findings