The objective is to build a machine learning model that can detect an abnormality in the X-Ray radiographs. These models can help towards providing healthcare access to the parts of the world where access to skilled radiologists is limited. According to a study on the Global Burden of Disease and the worldwide impact of all diseases found that, “musculoskeletal conditions affect more than 1.7 billion people worldwide. They are the 2nd greatest cause of disabilities, and have the 4th greatest impact on the overall health of the world population when considering both death and disabilities”. (www.usbji.org, n.d.).
Stanford University Machine Learning Group has published a paper related to this problem and provided one of the world’s largest public radiographic images dataset called MURA. MURA is short for Musculoskeletal Radiographs. Stanford university ML group has used DenseNet169 algorithm to train a deep neural network which can detect abnormalities in the radiographs with the accuracy closer to the top radiologists. This project attempts to implement deep neural network using DenseNet169 inspired from the Stanford Paper Rajpurkar, et al., 2018.
I have written separate blog and a youtube presentation in detail about this project. Please check out https://ramsrajkumar.com/2019/02/10/densenet-project/
https://www.youtube.com/watch?v=Evmlhr9_XvM
http://cs231n.github.io/convolutional-networks/ https://arxiv.org/pdf/1608.06993.pdf https://arxiv.org/abs/1406.4729