Fig. 4From: Multimodal deep learning-based diagnostic model for BPPVModel structure. In the short-term module, the convolutional kernel is a 4\(\times\)1\(\times\)1 3D convolutional module for downsampling in the time dimension. In the big kernel long-term module, the large convolution kernel expands the model perceptual field and extracts eye-movement featuresBack to article page