Developing an artificial intelligence diagnostic tool for paediatric distal radius fractures, a proof of concept study

Abstract

Introduction: In the UK, 1 in 50 children sustain a fractured bone yearly, yet studies have shown that 34% of children sustaining an injury do not have a visible fracture on initial radiographs. Wrist fractures are particularly difficult to identify because the growth plate poses diagnostic challenges when interpreting radiographs.

Methods: We developed Convolutional Neural Network (CNN) image recognition software to detect fractures in radiographs from children. A consecutive data set of 5,000 radiographs of the distal radius in children aged less than 19 years from 2014 to 2019 was used to train the CNN. In addition, transfer learning from a VGG16 CNN pretrained on non-radiological images was applied to improve generalisation of the network and the classification of radiographs. Hyperparameter tuning techniques were used to compare the model with the radiology reports that accompanied the original images to determine diagnostic test accuracy.

Results: The training set consisted of 2,881 radiographs with a fracture and 1,571 without; 548 radiographs were outliers. With additional augmentation, the final data set consisted of 15,498 images. The data set was randomly split into three subsets: training (70%), validation (10%) and test (20%). After training for 20 epochs, the diagnostic test accuracy was 85%.

Discussion: A CNN model is feasible in diagnosing paediatric wrist fractures. We demonstrated that this application could be utilised as a tool for improving diagnostic accuracy. Future work would involve developing automated treatment pathways for diagnosis, reducing unnecessary hospital visits and allowing staff redeployment to other areas.

Keywords: Artificial intelligence; Convolutional Neural Network; Image classification; Radiographs; Wrist fracture; X-rays.

Figure 1

Schematic of image labelling, training and validation prior to testing

Figure 2

Graph of accuracy with respect to epochs showing that training accuracy reached its peak by the end of training, whereas validation accuracy peaked at 86% for 20 epochs

Figure 3

Graph illustrating that training loss decreased with each epoch, whereas the validation loss flattened after ten epochs

Figure 4

Decision-making process of the Convolutional Neural Network (CNN) classifier layer illustrating ‘flatten’, ‘activation’ and the final dense layer with two neurons. The output of the SoftMax function is the probability of the class of the image being ‘fracture’ or ‘no fracture’

Figure 5

Heat map of the features learnt by the model at ‘block 2’ convolution on lateral view (red greater significance, blue lower significance)

Figure 6

Heat map of the features learnt by model at block2_convolution on antero-posterior view (red greater significance, blue lower significance)

Figure 7

Heat map of the features learnt by model at block5_pooling on antero-posterior view (red greater significance, blue lower significance)