Copyright ©The Author(s) 2021.
World J Gastroenterol. Jun 7, 2021; 27(21): 2818-2833
Published online Jun 7, 2021. doi: 10.3748/wjg.v27.i21.2818
Table 3 Advantages and disadvantages of representative machine-learning methods in the development of artificial intelligence-models for gastrointestinal pathology
AI model
Conventional ML (supervised)User can reflect domain knowledge to featuresRequires hand-crafted features; Accuracy depends heavily on the quality of feature extraction
Conventional ML (unsupervised)Executable without labelsResults are often unstable; Interpretability of the results
Deep neural networks (CNN)Automatic feature extraction; High accuracyRequires a large dataset; Low explainability (Black box)
Multi-instance learningExecutable without detailed labelsRequires a large dataset; High computational cost
Semantic segmentation (FCN, U-Net)Pixel-level detection gives the position, size, and shape of the targetHigh labeling cost
Recurrent neural networksLearn sequential dataHigh computational cost
Generative adversarial networksLearn to synthesize new realistic dataComplexity and instability in training