|
| S. no. | Author name and year | Model | Recent application in healthcare | Accuracy | Limitation |
|
| 1 | Yamashita et al. (2018) | CNN | Radiology [15] | 99.3% confidence | Need lots of labeled data for classification |
| 2 | Humayun et al. (2018) | To detect the abnormal heart sound [16] | Cross-fold Macc of 87.10, an absolute improvement of 9.54% over the baseline CNN system |
| 3 | Ismail et al. (2020) | Health model for regular health factor analysis [25] | Accuracy reaches 95.60% | Only two layers are used to classify the positive and negative correlated factors |
| 4 | Choi et al. (2017) | RNN | To detect the onset of heart failure [17] | The AUC for the RNN model increased to 0.883 | Require a massive volume of datasets
Have various problems due to gradient vanishing |
| 5 | Khodabakhshi et al. (2018) | To classify the abnormalities in the lungs [19] | Classification accuracy of 91% |
| 6 | Maragatham et al. (2019) | Prediction of heart failure in big data [14] | 0.894 AUC | Delineates the time taken for the training of two diverse LSTM models |
| 7 | Gharehbaghi et al. (2018) | DNN | Phonocardiography [20] | Accuracy reaches 92.60% | The learning process is too slow |
| 8 | Chen et al. (2018) | DBN | To detect type 1 diabetes [26] | 71.5%, recall of 60.2%, and score of 65.4% | The training process is computationally expensive |
| 9 | Seeliger et al. (2018) | GAN | Reconstructing natural images from brain activity [22] | 72.2% correct identifications | Hard to learn to generate discrete data |
| 10 | Emami et al. (2018) | Generating synthetic brain CTs [23] | PSNR was and SSIM was | Very hard to train |
| 11 | San et al. (2016) | DBN | To detect the hypoglycemic episodes in children with type 1 diabetes [24] |
| The initialization process makes expensive computational overhead |
|
