Body Part Classification from Medical Single Channel Images Using Deep Neural Networks

In general image recognition is aimed for classification of objects in RGB images and photos. However, many medical devices produce single channel images such as X-ray and MRI. There are several successful architectures with pre-trained input layers which are used for many classification tasks. In this paper ResNet with several configurations and one custom neural network architecture were used to classify human body parts in medical single channel images. From testing results it can be concluded that designed network provided better results than pre-trained ResNet. Medical MNIST dataset was used for the testing. Achieved results can be used in development of intelligent medical databases with AI filtering and processing of images.