Evaluation of the post-processing algorithms SimGrid and S-Enhance for paediatric intensive care patients and neonates

Abstract

Background: Post-processing software can be used in digital radiography to achieve higher image quality, especially in cases of scattered radiation. SimGrid is a grid-like software based on a Convolutional Neuronal Network that estimates the distribution and degree of scattered radiation in radiographs and thus improves image quality by simulating an anti-scatter grid. S-Enhance is an algorithm programmed to improve contrast visibility of foreign material.

Objective: The objective of this study was to evaluate the SimGrid and S-Enhance digital radiography post-processing methods for neonatology and paediatric intensive care.

Materials and methods: Two hundred and ten radiographs from the neonatal (n = 101, 0 to 6 months of age) and paediatric (n = 109, 6 months to 18 years of age) intensive care units performed in daily clinical routine using a mobile digital radiography system were post-processed with one of the algorithms, anonymized and then evaluated comparatively by two experienced paediatric radiologists. For every radiograph, patient data and exposure data were collected and analysed.

Results: Analysis of different radiographs showed that SimGrid significantly improves image quality for patients with a weight above 10 kg (range: 10-30 kg: odds ratio [OR] = 6.683, P < 0.0001), especially regarding the tracheobronchial system, intestinal gas, and bones. Utilizing S-Enhance significantly advances the assessment of foreign material (OR = 136.111, P < 0.0001) and bones (OR = 34.917, P < 0.0001) for children of all ages and weight, whereas overall image quality decreases.

Conclusion: SimGrid offers a differentiated spectrum in image improvement for children beyond the neonatal period whereas S-Enhance especially improves visibility of foreign material and bones for all patients.

Keywords: Deep learning; Digital radiography; Image enhancement; Neonatal intensive care unit; Pediatric intensive care unit; Radiation dosage; Radiography.

Fig. 1

A block diagram illustrates the image processing with SimGrid [5] in an 8-year-old boy (weight: 25 kg) with pneumonia. The original image (near right) after post-processing with convolutional neural networks shows an estimated scatter distribution map resulting in the post-processed image (far right) with improved assessment of lung structures as well as improved contrast

Fig. 2

A block diagram illustrates the image processing with S-Enhance [23] in a 2-year-old girl (weight: 11 kg) with postinterventional foreign material in projection to the left part of the chest. The original image (near right) is composed in two different ways with detail and contrast enhancement as well as an algorithm for transparency amplification resulting in the post-processed image (far right)

Fig. 3

A bar chart shows the odds ratio (OR) for improved interpretability of the images through SimGrid influenced by patient weight, reference line for OR > 1 (improvement)

Fig. 4

An anteroposterior chest radiograph (69.6 kV, 4.0 mAs, 4.51 cGy·cm², exposure index: 345.50), unprocessed image (a) versus a SimGrid image (b) in an 8-year-old boy (weight: 25 kg) with pneumonia

Fig. 5

An anteroposterior abdominal radiograph (59.8 kV, 2.0 mAs, 0.57 cGy·cm²), unprocessed image (a) versus a SimGrid image (b) in a 1-year-old boy with central venous catheter from right femoral vein with the tip in the inferior vena cava (small black ring) and intestinal gas in the small bowel (large black ring)

Fig. 6

Results differentiating between different evaluation parameters for SimGrid with a red line marking the odds ratio (OR) = 1

Fig. 7

A bar chart shows the odds ratio (OR) for an improvement of interpretability of the images using S-Enhance influenced by patient weight, reference line for OR > 1 (improvement)

Fig. 8

An anteroposterior chest radiograph (69.6 kV, 4.0 mAs, 3.91 cGy·cm², exposure index: 376.78) unprocessed image (a) versus an S-Enhance image (b) in a 2-year-old girl (weight: 11 kg) with postinterventional foreign material (x-ray marker on compress) in projection to the left part of the chest (black circle)

Fig. 9

Results differentiate between different evaluation parameters for S-Enhance with a red line marking odds ratio (OR) = 1