Interobserver agreement in interpretation of radiographic pulmonary changes in dogs in relation to radiology training

Abstract

Interpretation of pulmonary radiographs is one of the most difficult aspects of radiology and interobserver variability is high. The aim of this study was to assess variations in interpretation of pulmonary pathology amongst Brazilian veterinarians with different levels of training and experience, using the interpretation by American board-certified radiologists as a reference. We identified areas where interpretation is particularly challenging. Sixty digital canine thoracic radiographic examinations were interpreted by four groups of three Brazilian observers, each group being defined by different levels of training and experience. The radiographic findings of the 4 groups of observers in the study were compared to a reference interpretation established from the findings of three ACVR board-certified radiologists. The degree of discrepancy for each list between each group and the reference interpretation was assessed according to a three-level scoring system: no discrepancy, minor discrepancy, or major discrepancy. Data was analyzed using a Kappa and Cochran-Mantel-Haenszel tests. Brazilian veterinarians with the most training and experience showed the least interobserver variation and best performance when compared to the reference interpretation, followed by those with practical training, but with little work experience in professional practice. The radiographic patterns that were associated with the highest interobserver variability were the vascular, unstructured interstitial and bronchial patterns. Interobserver major discrepancies occurred in all groups, but is more evident in groups with the least training (44.4%) and the general practitioners (26.7%) group. It can be concluded that training positively influences the accuracy of radiographic interpretation and is recommended to reduce erroneous diagnoses. Key words: Training, interobserver agreement, thoracic radiographs, pulmonary patterns ResumoUma das maiores dificuldades na interpretação radiográfica em cães está nas alterações pulmonares sendo a variabilidade interobservador alta. O objetivo desse estudo é detectar as variações de interpretação entre radiologistas brasileiros em diferentes graus de treinamento e experiência, utilizando a interpretação de consenso feito por radiologistas americanos certificados pelo Colégio Americano de Radiologia Veterinária. Na tentativa de identificar os desafios e as particularidades dessa interpretação. Sessenta exames radiográficos digitais do tórax de cães foram interpretados por quatro grupos de observadores com diferentes graus de treinamento em leitura de exames radiográficos. O grau de discrepância entre as observadores foram comparados seguindo um escore com três subclassificações: sem discrepância, discrepância leve e maiores discrepâncias. Para análise dos dados os métodos estatísticos utilizados foram o Kappa e Cochran-Mantel-Haenszel. Os veterinários brasileiros com maior grau de treinamento e experiência foram o que apresentaram menores variações de interpretação quando comparado aos dados d...

Automatic classification of canine thoracic radiographs using deep learning

“…In human medicine, despite the efforts to improve radiology residents’ training programmes, the prevalence of interpretation errors has not significantly improved in recent decades 1 , 2 . The prevalence and the impact of interpretation errors on thoracic radiographs have only seldom been investigated in veterinary medicine 4 . Conversely, this topic has been widely studied in human medicine and the most common causes of interpretation errors have been identified 5 – 7 .…”