Methods and materials
Seven hundred twenty-three (n = 723, 47.7% female 52.3% male) unprocessed DICOM images of digital radiographic (DR) chest X-rays of PICU patients were acquired from the local PACS at the author’s sponsoring institution. An approval by the local ethics committee (29-558 ex 16/17) was acquired prior to the collection of the images from the archives.
Data Acquisition: Images were acquired based on the following age groups suggested by the National Institute of Child Health and Development (NICHD)[7]: newborn (0–29 days, n = 120), infant (1–12 months, n = 120), toddler (13–24 months, n = 108), early childhood (2–5 years n = 125), middle childhood (6–11 years, n = 121), and early adolescence (12-18 years, n = 129). All Images had been acquired using the division’s mobile unit. 99.3% of all images were shot in supine anterior-posterior (AP) chest view, with the remainder in semi erect position.
Qualitative and Quantitative parameters: A semi-automatic analysis was performed using a custom Fiji ImageJ Macro proposed by a preceding study. [5, 8] The qualitative parameters were based on the EC guidelines as depicted in Table 1.
Quantitatively, field sizes according the recommendations suggested by the EC guidelines were defined (Figure 1). There, a minimal field size (MinFS) extending from above the lung apices to T12/L1 is recommended. The outside of the thoracic cage was chosen as a lateral border, as the EC guidelines do not specify such. The maximal allowed field size (MaxFS) was computed by adding an age-dependent tolerance (as listed in Table 1). An actual exposed area (AEA) was defined in order to compute overexposure. Additionally, we analysed rotation, distances between the margins of the measured field sizes (MinFS, MaxFs, and AEA), the position, and possible truncation of the costophrenic angles (Figure 2).
Data analysis: The data was analysed with SPSS Statistics Version 24 software using descriptive statistics, t-tests, ANOVA, and regression analyses. Observer 1 performed all analyses, and p-values less than 0.05 were considered significant. Inter- and intra-observer agreement were examined using Cohen's Kappa and Inter Class Correlations (n = 37). The second observer had developed the macro, and thus didn't require any further training.