Table 3. Current deep learning models in medical imaging.

Architecture	First applied in medical image processing	Features	Difficulties
CNN	1998	1. Suitable for image classification, segmentation, and detection	1. Extract local detailed information from images
		2. Extract image features through multiple layers of convolutional and pooling operations	2. Sensitive to changes in the input image size
		3. Pretrained models for transfer learning
RNN	2001	1. Handle text and speech	1. Long sequential data
		2. Model input data through the calculation of recurrent neurons	2. Tackle noise or outliers in the input data
		3. Able to process current information while retaining historical information	3. Training process is prone to the vanishing or exploding gradient problem
U-Net	2015	1. Specialized for medical image segmentation	Extract complex shapes and texture information from images
		2. Effectively compress image information while preserving high-resolution features
		3. Fewer parameters and trained faster
Transformer	2021	1. Handle text, audio, and images	1. A lot of computing resources and time
		2. Self-attention to learn global dependencies	2. Long-term dependency problems when processing long sequential data
		3. Parallelizes the processing of multiple sequence elements	3. Extract local information in input data
		4. Pretrained models for transfer learning

CNN, Convolutional Neural Network; RNN, Recurrent Neural Network.