| dc.description.abstract | Fractures in children are among the medical cases that take a lot of doctors’ attention and time while analyzing the images to identify the presence or absence of fracture. More so a tired doctor may fail to identify the presence of fracture after looking at many images with health bones which may result in poor treatment. Our method of fracture can help doctors in faster detection, meet their deadline but also can help in obtaining accurate results hence administer proper treatment. In this research, we used X-ray images which we obtained from Mengo hospital. We obtained 60,140 Images and with the help of a radiologist we were able to find 211 images of children from the entire data set and also, we were able to separate images with fracture from those with fracture. we split our data into 3 different sets (train set, validation set, and test set). We used a convolution neural network (CNN) a method of automatic detection that is mostly preferred for cases that require deep analysis. We trained our model on 211 x-ray images and used a kernel to generate different features of interest. we obtained 4 convolution layers while developing our model and a max-pooling layer was placed in between each 2d convolutions layers to record the weights of the parameters but also to reduce over-fitting. We used 3 different matrices to evaluate our model which include accuracy, Precision, and recall. On training the model, we obtained Test accuracy levels of 47%, 59% for test precision, and Test recall of 76%. while using our data-set and we noticed an over-fit because of the small data set used, we then applied data augmentation and obtained an accuracy level of 56%, test precision 49%, and test recall of 76%. We also trained our model on VGG Net a model and we obtained an accuracy level of 63%, test precision of 66%, and test recall of 79%. | en_US |
