BlazeNeo: Blazing Fast Polyp Segmentation and Neoplasm Detection

In recent years, computer-aided automatic polyp segmentation and neoplasm detection have been an emerging topic in medical image analysis, providing valuable support to colonoscopy procedures. Attentions paid improving the accuracy of segmentation. However, not much focus has given latency throughput for performing these tasks on dedicated devices, which can be crucial practical applications. This paper introduces a novel deep neural network architecture called BlazeNeo, task with emphasis compactness speed while maintaining high accuracy. The model leverages highly efficient HarDNet backbone alongside lightweight Receptive Field Blocks feature aggregation mechanism computational efficiency. An auxiliary training strategy is proposed take full advantage data quality. Our experiments challenging dataset show that BlazeNeo achieves improvements size comparable against state-of-the-art methods. We obtain over 155 fps outperforming all compared models terms INT8 precision when deploying edge device conventional configuration.