Hardware Accelerators for Classification of Thoracic Disorders: A Survey

Deep learning, specifically convolutional neural networks (CNNs), has been successfully applied to the classification of chest X-ray images for diagnosing thoracic diseases. CNNs are effective in learning intricate patterns and features from images by training on large, labeled datasets. With this training, they can identify specific patterns associated with different thoracic diseases and classify new, unseen chest X-ray images into relevant disease categories. This application of Deep Neural Networks aids in the diagnosis and treatment of patients with thoracic diseases. The paper discusses the role of hardware accelerators, specifically Field-Programmable Gate Arrays (FPGAs), in accelerating deep learning model training processes. It emphasizes the importance of powerful hardware for deep learning advancements due to the need for vast amounts of data and high computing capabilities. The paper surveys techniques for implementing and optimizing convolutional neural networks (CNNs) on FPGA platforms, highlighting their efficiency and re-configurability. It also addresses the challenges in designing related embedded systems and proposes a complete acceleration design flow that explores different interfaces between the CPU and hardware accelerators.