Assessing the Inter-Operator Variability in Dynamic Chest Radiography Image Interpretation

Abstract

RATIONALE Dynamic chest radiography (DCR) is a new imaging technique that involves taking sequential chest radiographs throughout the respiratory cycle. This allows the real time observation of the changes in lung and diaphragmatic movement occurring during respiration. Recent studies have correlated DCR acquired measures with those of pulmonary function testing, indicating that it can be an alternative measure of pulmonary physiology. Currently DCR values are manually recorded introducing an element of subjectivity. If this technique is to be widely adopted the nature of the inter-operator variability will need to be characterised. This study evaluates the inter-operator variability between two observers recording two key DCR obtained values. METHODS Two independent reviewers separately recorded measurements from the DCR image sequences of 50 patients. Their values for 'maximal distance from lung apex to diaphragm' and 'total lung area' at the point of maximal inspiration and expiration for both the right and left lungs were collated in a database. The inter-operator variability (IOV) for their inspiratory and expiratory values was assessed using Bland-Altman plots and Deming regression analysis was used to investigate statistical error in the observations. RESULTS The Bland-Altman pIot demonstrates the majority of the reviewer's observations for 'maximum distance' in the inspiratory images were within the 95% limits of agreement, with only a few outliers (right lung maximum distance; slope= 1.01 (SE 0.02, 95% CI 0.97-1.05) left lung max distance; slope= 0.98 (SE 0.04 95% CI 0.90-1.05). This was a similar picture for lung area (right lung area; slope= 1.0 (SE 0.01 95% CI 0.98-1.03), left lung area; slope= 1.02 (SE 0.01 95% CI 0.97-1.05)). Deming regression analysis showed good agreement. The two reviewers also had good agreement for expiratory images (right lung maximum distance; slope= 1.01 (SE 0.06, 95% CI 0.98-1.04) left lung max distance; slope= 0.99 (SE 0.05 95% CI 0.89-1.09). Again 'lung area' measurements were similar (right lung area; slope= 0.16 (SE 0.86 95% CI -1.58-1.88), left lung area; slope= 1.03 (SE 0.01 95% CI 1.0-1.06)). Deming regression showed poor fit however this was influenced by an outlying measurement. CONCLUSIONS In general there was good agreement between the two reviewers when interpreting the various images. However, as this imaging technique becomes more widespread the image analysis would benefit from being uniformly interpreted.