Close
  Indian J Med Microbiol
 

Figure 5: Diagram of EBUS multimodal image prediction model based on deep learning. Frames extracted from three modes of EBUS videos are input into the LSTM model, which can build the relationship between frames. With the relation between frames, we can obtain the importance of each frame and select the typical frame from the whole video. ROI is delineated to obtain the lymph node area from the whole image using FCN and CRF. Then, ROI images from three modes are input into CNN under a multi-modal manner which can facilitate the diagnosis of the CNN model. EBUS: Endobronchial ultrasound; LSTM: Long-short term memory; FCN: Fully connected network; CRF: Conditional random filed; ROI: Region of interest; CNN: Convolutional neural networks

Figure 5: Diagram of EBUS multimodal image prediction model based on deep learning. Frames extracted from three modes of EBUS videos are input into the LSTM model, which can build the relationship between frames. With the relation between frames, we can obtain the importance of each frame and select the typical frame from the whole video. ROI is delineated to obtain the lymph node area from the whole image using FCN and CRF. Then, ROI images from three modes are input into CNN under a multi-modal manner which can facilitate the diagnosis of the CNN model. EBUS: Endobronchial ultrasound; LSTM: Long-short term memory; FCN: Fully connected network; CRF: Conditional random filed; ROI: Region of interest; CNN: Convolutional neural networks