Lightweight Combinational Machine Learning Algorithm for Sorting Canine Torso Radiographs

The veterinary field lacks automation in contrast to the tremendous technological advances made in the human medical field. Implementation of machine learning technology can shorten any step of the automation process. This paper explores these core concepts and starts with automation in sorting radiographs for canines by view and anatomy. There is a wide range of deep learning approaches explored in the literature; while these models can obtain good accuracy, they require a considerable amount of memory, which is an undesirable trait. As a result, in this paper, a new lightweight algorithm is proposed, inspired by AlexNet, Inception, and SqueezeNet. The proposed model has only 0.1 million parameters which is about 100 times less than the number of parameters in the state-of-the-art models such as AlexNet and ResNet. Moreover, it achieves higher accuracy than AlexNet, ResNet, DenseNet, and SqueezeNet, while having much less running time, and requiring less memory.