Fast Semantic Segmentation of Medical Images

Semantic segmentation of medical images enables automatic identification and localization of anatomical structures and pathological regions. The technique classifies image regions at the pixel level, offering a powerful method for accurate diagnosis of diseases. Despite many advancements in deep learning models with high accuracy, the deployment of these systems in real-world clinical settings is slowed down by the computational requirements of these models, leading to inefficient systems with slow processing times. The work proposed in this paper is an efficient and fast semantic segmentation network tailored for deployment on an embedded computing system. By optimizing the model to reduce its size and computational complexity and applying hardware-oriented optimizations, this work aims to overcome current limitations in computational efficiency and latency. The results show a reduction of more than 10× in model size compared to state-of-the-art works with a similar accuracy.