Multi‐instance inflated 3D CNN for classifying urine red blood cells from multi‐focus videos

Classifying urine red blood cells (U-RBCs) is the core operation in diagnosing urinary system diseases (USDs). In this paper, based on a novel data type named multi-focus video, multi-instance inflated 3D convolutional neural network (MI3D) proposed. order to accurately classifying U-RBCs, MI3D integrates inception-V1 with learning models. Compared existent U-RBC classification methods relying single focus images, using videos effectively avoids misclassification caused by significant deformation of U-RBCs microscope changing. addition, can learn typical shapes and patterns from multi-focal simultaneously. Therefore, accuracy exceeds mainstream video There are totally 597 that include four types collected verify effectiveness MI3D. Experimental results show inspiring 94.4%, which obviously higher than existed method (85.6%). The also achieves comparable level junior microscopist (95.6%). Lastly, has powerful real-time performance, whose speed reaches 1.4 times microscopist.