EDI will develop a computer program that will detect abnormalities in the pediatric bone
structure through a convolutional neural network and alert the physician whether or not there is a
fracture. Furthermore, in the event of a fracture, the program will identify a region to look out for, as well
as classify it by classification codes for most common diagnoses.
CNN will be trained with anonymised, labeled patient data, customizing the CNN architecture as
necessary to perform the task with high accuracy, which is crucial when working with medical data.
After the implementation of the program in the daily routine of physicians, the examination and diagnosis
process will become smoother – after questioning and sending the patient for an X-ray and receiving a
response, the physician will upload the received image to the program on the computer. Basing it’s
decision on the training done before, the program will determine whether or not there is a fracture in the
X-ray image considering a predefined confidence level. In cases where a fracture is detected, the program
will continue the analysis by identifying the fracture region and will classify the fracture by classification
codes for most common diagnoses. Next, the doctor will carry out further treatment, considering the
recommendations given by the program.
For fracture recognition convolutional neural networks (CNN), a type of AI, will be used. In cases when
sufficient amount of data is available, the accuracy achieved by CNNs is at least as good as the analysis
of radiologists. The input data will be X-ray images from which CNN will use filters to find sets of
visually similar properties – features. Filters can be handcrafted, horizontal or vertical line detectors for
example, but the real advantage of CNNs is the ability to learn the necessary filters during training.
Training data sets can be tailored to specific fracture problems, making CNN superior to other methods,
such as classic neural networks or recurrent neural networks. Adaptation to a specific problems is necessary
because, unlike adult bones, the bones of children are not completely ossified, the cartilage is
invisible radiographically, and the nuclei of ossification can simulate fracture.