Skip to main content
. 2023 Jun 21;12(7):893. doi: 10.3390/biology12070893

Table 1.

Summary of deep learning methods in cancer prognosis studies.

Deep Learning Category Brief Description
Autoencoders [31,32,33,34,35,36,37,38,39,40,41] A specific type of artificial neural network architecture designed for unsupervised learning of efficient codings. Autoencoders learn to compress input data into a coded representation and then uncompress it back to the original input. This capability makes them especially useful for dimensionality reduction and feature extraction tasks, where they learn to preserve as much information as possible while representing data in a reduced-dimensional space.
Ensemble of deep learning methods/models [42,43,44,45,46,47,48,49] An ensemble method combines predictions from several different deep learning models to improve the overall predictive accuracy. Each model in the ensemble contributes a portion of the overall prediction. Through the combination of diverse models, the ensemble method can exploit the strengths of each individual model while mitigating their weaknesses, leading to a more robust and accurate overall prediction.
Multi-modal/multi-omics deep learning models [50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76] These are sophisticated techniques designed to integrate and analyze data from different sources or modalities (such as imaging, genomics, proteomics, etc.) to enhance learning performance. By integrating data from diverse sources, these models can capture complex and hidden patterns that may not be evident when each data type is analyzed separately.
Deep learning with transfer learning [77,78,79,80] In the context of deep learning, transfer learning involves leveraging a pre-trained model (usually trained on a large-scale dataset) or transferring learned features from one task to another. The key advantage of transfer learning is that it can significantly improve learning efficiency and performance, especially when the target task has limited training data.
Deep learning for manifold representation [81,82,83,84,85,86,87,88,89,90,91] These methods employ deep learning architectures to generate numerical vector representations or embeddings of features in a lower-dimensional space (a manifold). Such representations can be utilized to reveal and preserve the intrinsic structure and relationships within the data, which are instrumental for downstream tasks.
Deep learning (unspecified/generic) [92,93,94,95,96,97,98,99,100,101,102,103,104,105,106,107,108,109] This category encompasses general deep learning approaches, which may include a variety of architectures and techniques. These approaches often involve several minor modifications or adaptations to cater to the specificities of the task at hand, without specializing in a particular method or model like the other categories.