|
Publication | Type(s) of cancer | Type of data | Methods | Performance |
|
[50] | Astrocytic tumor | Microarray gene dataset | ANN | 96.15% accuracy |
[51] | Breast cancer | Nuclear morphometric features | ANNs | Good (>5 years) and bad (<5 years) prognoses |
[52] | Breast invasive carcinoma | Gene expression data | Multiomics neural networks | Improved performance using more omics data |
[53] | Breast cancer | TCGA | Random forest, neural network | Log-rank |
[54] | Malignant melanoma | Custom dataset | Nonlinear ANN model | ANN model performs better than Cox model |
[55] | Multiple | WHAS, SUPPORT, METABRIC, Rotterdam tumor bank | Deep feedforward neural network | Better prognostic accuracy than the clinical experts for the prognosis of nasopharyngeal carcinoma |
[56] | Glioblastoma multiforme | TCGA | Pathway-associated sparse deep neural network | , |
[57] | Breast cancer | Gene expression profile+copy number alteration profile+clinical data | Multimodal deep neural network | The proposed method achieves better performance than the prediction methods with single-dimensional data and other existing approaches |
[58] | Hepatocellular carcinoma | TCGA | DL-based model | value = Concordance |
[59] | Colorectal cancer | Images of tumor tissue samples | Combined convolutional and recurrent architectures | Prediction with only small tissue areas (hazard ratio 2.3), tissue microarray spot (hazard ratio 1.67), and whole-slide level (hazard ratio 1.65) |
[60] | Ovarian cancer | CT images | Combined DL and Cox proportional hazards model | Concordance index was 0.713 and 0.694 |
[61] | Multiple | TCGA | ANN framework | Same or better predictive accuracy compared to other methods |
[62] | Multiple | WHAS, SUPPORT, & METABRIC | Cox proportional hazards deep neural network | Superior in predicting personalized treatment recommendations |
[63] | Lower-grade glioma and glioblastoma | TCGA | CNNs | Median concordance |
[64] | Mesothelioma | TCGA+French source | CNNs | Concordance index of 0.656 on TCGA cohort |
[65] | Multiple | TCGA+Gene Expression Omnibus dataset | DL-based model | For both marker types, the specificity of normal whole blood was 100% |
|